Split NodePerformanceProfile state storage into separate mappings

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

.github/workflows/build-app.yml

@@ -1,5 +1,16 @@
 name: Build EXO macOS DMG
 
+# Release workflow:
+# 1. Create a draft GitHub Release with the tag name (e.g. v1.0.0) and write release notes in markdown
+# 2. Push the tag: git tag v1.0.0 && git push origin v1.0.0
+# 3. This workflow builds, signs, and notarizes the DMG
+# 4. Release notes are embedded in appcast.xml for Sparkle (rendered as markdown)
+# 5. DMG and appcast.xml are uploaded to S3
+# 6. The draft GitHub Release is published with the DMG attached
+#
+# For alpha releases (e.g. v1.0.0-alpha.1): draft release and notes are optional.
+# If no draft exists, a release is auto-created with generated notes.
+
 on:
   workflow_dispatch:
   push:
@@ -11,8 +22,10 @@ on:
 jobs:
   build-macos-app:
     runs-on: "macos-26"
+    permissions:
+      contents: write
     env:
-      SPARKLE_VERSION: 2.8.1
+      SPARKLE_VERSION: 2.9.0-beta.1
       SPARKLE_DOWNLOAD_PREFIX: ${{ secrets.SPARKLE_DOWNLOAD_PREFIX }}
       SPARKLE_FEED_URL: ${{ secrets.SPARKLE_FEED_URL }}
       SPARKLE_ED25519_PUBLIC: ${{ secrets.SPARKLE_ED25519_PUBLIC }}
@@ -87,6 +100,52 @@ jobs:
             exit 1
           fi
 
+      - name: Fetch and validate release notes
+        if: github.ref_type == 'tag'
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          # Find draft release by name using gh release list (more reliable with default token)
+          echo "Looking for draft release named '$GITHUB_REF_NAME'..."
+          DRAFT_EXISTS=$(gh release list --json name,isDraft --jq ".[] | select(.isDraft == true) | select(.name == \"$GITHUB_REF_NAME\") | .name" 2>/dev/null || echo "")
+
+          if [[ -z "$DRAFT_EXISTS" ]]; then
+            if [[ "$IS_ALPHA" == "true" ]]; then
+              echo "No draft release found for alpha tag $GITHUB_REF_NAME (optional for alphas)"
+              echo "HAS_RELEASE_NOTES=false" >> $GITHUB_ENV
+              exit 0
+            fi
+            echo "ERROR: No draft release found for tag $GITHUB_REF_NAME"
+            echo "Please create a draft release with release notes before pushing the tag."
+            exit 1
+          fi
+
+          # Fetch full release details via API to get body and ID
+          echo "Found draft release, fetching details..."
+          RELEASE_JSON=$(gh api repos/${{ github.repository }}/releases --jq ".[] | select(.draft == true) | select(.name == \"$GITHUB_REF_NAME\")" 2>/dev/null || echo "")
+
+          # Extract release notes
+          NOTES=$(echo "$RELEASE_JSON" | jq -r '.body // ""')
+          if [[ -z "$NOTES" || "$NOTES" == "null" ]]; then
+            if [[ "$IS_ALPHA" == "true" ]]; then
+              echo "Draft release has no notes (optional for alphas)"
+              echo "HAS_RELEASE_NOTES=false" >> $GITHUB_ENV
+              exit 0
+            fi
+            echo "ERROR: Draft release exists but has no release notes"
+            echo "Please add release notes to the draft release before pushing the tag."
+            exit 1
+          fi
+
+          # Save release ID for later publishing
+          RELEASE_ID=$(echo "$RELEASE_JSON" | jq -r '.id')
+          echo "DRAFT_RELEASE_ID=$RELEASE_ID" >> $GITHUB_ENV
+          echo "HAS_RELEASE_NOTES=true" >> $GITHUB_ENV
+
+          echo "Found draft release (ID: $RELEASE_ID), saving release notes..."
+          echo "$NOTES" > /tmp/release_notes.md
+          echo "RELEASE_NOTES_FILE=/tmp/release_notes.md" >> $GITHUB_ENV
+
       # ============================================================
       # Install dependencies
       # ============================================================
@@ -304,6 +363,28 @@ jobs:
             $CHANNEL_FLAG \
             .
 
+      - name: Inject release notes into appcast
+        if: github.ref_type == 'tag' && env.HAS_RELEASE_NOTES == 'true'
+        env:
+          RELEASE_VERSION: ${{ env.RELEASE_VERSION }}
+        run: |
+          # Inject markdown release notes with sparkle:format="markdown" (Sparkle 2.9+)
+          export NOTES=$(cat "$RELEASE_NOTES_FILE")
+
+          # Insert description after the enclosure tag for this version
+          awk '
+            /<enclosure[^>]*>/ && index($0, ENVIRON["RELEASE_VERSION"]) {
+              print
+              print "            <description sparkle:format=\"markdown\"><![CDATA["
+              print ENVIRON["NOTES"]
+              print "            ]]></description>"
+              next
+            }
+            { print }
+          ' output/appcast.xml > output/appcast.xml.tmp && mv output/appcast.xml.tmp output/appcast.xml
+
+          echo "Injected markdown release notes for version $RELEASE_VERSION"
+
       # ============================================================
       # Upload artifacts
       # ============================================================
@@ -336,3 +417,26 @@ jobs:
             aws s3 cp "$DMG_NAME" "s3://${SPARKLE_S3_BUCKET}/${PREFIX}EXO-latest.dmg"
             aws s3 cp appcast.xml "s3://${SPARKLE_S3_BUCKET}/${PREFIX}appcast.xml" --content-type application/xml --cache-control no-cache
           fi
+
+      - name: Publish GitHub Release
+        if: github.ref_type == 'tag'
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          DMG_PATH="output/EXO-${RELEASE_VERSION}.dmg"
+
+          if [[ "$HAS_RELEASE_NOTES" == "true" ]]; then
+            # Update the draft release with the tag and upload DMG
+            gh api --method PATCH "repos/${{ github.repository }}/releases/$DRAFT_RELEASE_ID" \
+              -f tag_name="$GITHUB_REF_NAME" \
+              -F draft=false
+            gh release upload "$GITHUB_REF_NAME" "$DMG_PATH" --clobber
+            echo "Published release $GITHUB_REF_NAME with DMG attached"
+          else
+            # Alpha without draft release - create one with auto-generated notes
+            gh release create "$GITHUB_REF_NAME" "$DMG_PATH" \
+              --title "$GITHUB_REF_NAME" \
+              --generate-notes \
+              --prerelease
+            echo "Created alpha release $GITHUB_REF_NAME with auto-generated notes"
+          fi

TODO.md

@@ -19,7 +19,6 @@
 25. Rethink retry logic
 26. Task cancellation. When API http request gets cancelled, it should cancel corresponding task.
 27. Log cleanup - per-module log filters and default to DEBUG log levels
-28. Validate RDMA connections with ibv_devinfo in the info gatherer
 
 Potential refactors:
 

app/EXO/EXO.xcodeproj/project.pbxproj

@@ -585,7 +585,7 @@
 			repositoryURL = "https://github.com/sparkle-project/Sparkle.git";
 			requirement = {
 				kind = upToNextMajorVersion;
-				minimumVersion = 2.8.1;
+				minimumVersion = 2.9.0-beta.1;
 			};
 		};
 /* End XCRemoteSwiftPackageReference section */

app/EXO/EXO.xcodeproj/project.xcworkspace/xcshareddata/swiftpm/Package.resolved

@@ -6,8 +6,8 @@
       "kind" : "remoteSourceControl",
       "location" : "https://github.com/sparkle-project/Sparkle.git",
       "state" : {
-        "revision" : "5581748cef2bae787496fe6d61139aebe0a451f6",
-        "version" : "2.8.1"
+        "revision" : "e641adb41915a8409895e2e30666aa64e487b637",
+        "version" : "2.9.0-beta.1"
       }
     }
   ],

app/EXO/EXO/Services/NetworkSetupHelper.swift

@@ -6,7 +6,7 @@ enum NetworkSetupHelper {
     private static let logger = Logger(subsystem: "io.exo.EXO", category: "NetworkSetup")
     private static let daemonLabel = "io.exo.networksetup"
     private static let scriptDestination =
-        "/Library/Application Support/EXO/disable_bridge.sh"
+        "/Library/Application Support/EXO/disable_bridge_enable_dhcp.sh"
     private static let plistDestination = "/Library/LaunchDaemons/io.exo.networksetup.plist"
     private static let requiredStartInterval: Int = 1791
 
@@ -28,6 +28,35 @@ enum NetworkSetupHelper {
         # Remove Thunderbolt Bridge from VirtualNetworkInterfaces in preferences.plist
         /usr/libexec/PlistBuddy -c "Delete :VirtualNetworkInterfaces:Bridge:bridge0" "$PREFS" 2>/dev/null || true
 
+        networksetup -listlocations | grep -q exo || {
+          networksetup -createlocation exo
+        }
+
+        networksetup -switchtolocation exo
+        networksetup -listallhardwareports \\
+          | awk -F': ' '/Hardware Port: / {print $2}' \\
+          | while IFS=":" read -r name; do
+              case "$name" in
+                "Ethernet Adapter"*)
+                        ;;
+                "Thunderbolt Bridge")
+                        ;;
+                "Thunderbolt "*)
+                  networksetup -listallnetworkservices \\
+                    | grep -q "EXO $name" \\
+                      || networksetup -createnetworkservice "EXO $name" "$name" 2>/dev/null \\
+                      || continue
+                  networksetup -setdhcp "EXO $name"
+                        ;;
+                *)
+                  networksetup -listallnetworkservices \\
+                    | grep -q "$name" \\
+                      || networksetup -createnetworkservice "$name" "$name" 2>/dev/null \\
+                      || continue
+                        ;;
+              esac
+            done
+
         networksetup -listnetworkservices | grep -q "Thunderbolt Bridge" && {
           networksetup -setnetworkserviceenabled "Thunderbolt Bridge" off
         } || true

bench/exo_bench.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import argparse
+import contextlib
 import http.client
 import json
 import os
@@ -26,7 +27,7 @@ class ExoHttpError(RuntimeError):
 
 
 class ExoClient:
-    def __init__(self, host: str, port: int, timeout_s: float = 2400.0):
+    def __init__(self, host: str, port: int, timeout_s: float = 600.0):
         self.host = host
         self.port = port
         self.timeout_s = timeout_s
@@ -104,22 +105,46 @@ def runner_ready(runner: dict[str, Any]) -> bool:
     return "RunnerReady" in runner
 
 
+def runner_failed(runner: dict[str, Any]) -> bool:
+    return "RunnerFailed" in runner
+
+
+def get_runner_failed_message(runner: dict[str, Any]) -> str | None:
+    if "RunnerFailed" in runner:
+        return runner["RunnerFailed"].get("errorMessage")
+    return None
+
+
 def wait_for_instance_ready(
     client: ExoClient, instance_id: str, timeout: float = 24000.0
 ) -> None:
     start_time = time.time()
+    instance_existed = False
     while time.time() - start_time < timeout:
         state = client.request_json("GET", "/state")
         instances = state.get("instances", {})
 
         if instance_id not in instances:
+            if instance_existed:
+                # Instance was deleted after being created - likely due to runner failure
+                raise RuntimeError(
+                    f"Instance {instance_id} was deleted (runner may have failed)"
+                )
             time.sleep(0.1)
             continue
 
+        instance_existed = True
         instance = instances[instance_id]
         runner_ids = runner_ids_from_instance(instance)
         runners = state.get("runners", {})
 
+        # Check for failed runners first
+        for rid in runner_ids:
+            runner = runners.get(rid, {})
+            if runner_failed(runner):
+                error_msg = get_runner_failed_message(runner) or "Unknown error"
+                raise RuntimeError(f"Runner {rid} failed: {error_msg}")
+
         if all(runner_ready(runners.get(rid, {})) for rid in runner_ids):
             return
 
@@ -299,6 +324,12 @@ def main() -> int:
         default=4,
         help="Only consider placements using <= this many nodes.",
     )
+    ap.add_argument(
+        "--min-nodes",
+        type=int,
+        default=1,
+        help="Only consider placements using >= this many nodes.",
+    )
     ap.add_argument(
         "--instance-meta", choices=["ring", "jaccl", "both"], default="both"
     )
@@ -320,7 +351,7 @@ def main() -> int:
         help="Warmup runs per placement (uses first pp/tg).",
     )
     ap.add_argument(
-        "--timeout", type=float, default=2400.0, help="HTTP timeout (seconds)."
+        "--timeout", type=float, default=600.0, help="HTTP timeout (seconds)."
     )
     ap.add_argument(
         "--json-out",
@@ -399,7 +430,7 @@ def main() -> int:
         ):
             continue
 
-        if 0 < n <= args.max_nodes:
+        if args.min_nodes <= n <= args.max_nodes:
             selected.append(p)
 
     if not selected:
@@ -441,7 +472,13 @@ def main() -> int:
         )
 
         client.request_json("POST", "/instance", body={"instance": instance})
-        wait_for_instance_ready(client, instance_id)
+        try:
+            wait_for_instance_ready(client, instance_id)
+        except (RuntimeError, TimeoutError) as e:
+            logger.error(f"Failed to initialize placement: {e}")
+            with contextlib.suppress(ExoHttpError):
+                client.request_json("DELETE", f"/instance/{instance_id}")
+            continue
 
         time.sleep(1)
 

dashboard/src/lib/components/ModelCard.svelte

@@ -197,7 +197,7 @@ function toggleNodeDetails(nodeId: string): void {
 	// Uses API preview data when available, falls back to local estimation
 	const placementPreview = $derived(() => {
 		const nodeArray = nodeList();
-		if (nodeArray.length === 0) return { nodes: [], canFit: false, totalAvailable: 0, topoWidth: 260, topoHeight: 90, error: null };
+		if (nodeArray.length === 0) return { nodes: [], canFit: false, totalAvailable: 0, error: null };
 		
 		const numNodes = nodeArray.length;
 		const iconSize = numNodes === 1 ? 50 : 36;

dashboard/src/lib/components/TopologyGraph.svelte

@@ -1,7 +1,7 @@
 <script lang="ts">
 	import { onMount, onDestroy } from 'svelte';
 	import * as d3 from 'd3';
-import { topologyData, isTopologyMinimized, debugMode, type NodeInfo } from '$lib/stores/app.svelte';
+import { topologyData, isTopologyMinimized, debugMode } from '$lib/stores/app.svelte';
 
 	interface Props {
 		class?: string;
@@ -24,14 +24,14 @@ function getNodeLabel(nodeId: string): string {
 
 function getInterfaceLabel(nodeId: string, ip?: string): { label: string; missing: boolean } {
 	if (!ip) return { label: '?', missing: true };
-
+	
 	// Strip port if present (e.g., "192.168.1.1:8080" -> "192.168.1.1")
 	const cleanIp = ip.includes(':') && !ip.includes('[') ? ip.split(':')[0] : ip;
-
+	
 	// Helper to check a node's interfaces
-	function checkNode(node: NodeInfo | undefined): string | null {
+	function checkNode(node: typeof data.nodes[string]): string | null {
 		if (!node) return null;
-
+		
 		const matchFromInterfaces = node.network_interfaces?.find((iface) =>
 			(iface.addresses || []).some((addr) => addr === cleanIp || addr === ip)
 		);
@@ -39,19 +39,17 @@ function getInterfaceLabel(nodeId: string, ip?: string): { label: string; missin
 			return matchFromInterfaces.name;
 		}
 
-		if (node.ip_to_interface) {
-			const mapped = node.ip_to_interface[cleanIp] || (ip ? node.ip_to_interface[ip] : undefined);
-			if (mapped && mapped.trim().length > 0) {
-				return mapped;
-			}
+		const mapped = node.ip_to_interface?.[cleanIp] || node.ip_to_interface?.[ip];
+		if (mapped && mapped.trim().length > 0) {
+			return mapped;
 		}
 		return null;
 	}
-
+	
 	// Try specified node first
 	const result = checkNode(data?.nodes?.[nodeId]);
 	if (result) return { label: result, missing: false };
-
+	
 	// Fallback: search all nodes for this IP
 	for (const [, otherNode] of Object.entries(data?.nodes || {})) {
 		const otherResult = checkNode(otherNode);
@@ -257,24 +255,21 @@ function wrapLine(text: string, maxLen: number): string[] {
 		const arrowsGroup = svg.append('g').attr('class', 'arrows-group');
 		const debugLabelsGroup = svg.append('g').attr('class', 'debug-edge-labels');
 
-		type ConnectionInfo = { from: string; to: string; ip: string; ifaceLabel: string; missingIface: boolean };
-		type PairEntry = { a: string; b: string; aToB: boolean; bToA: boolean; connections: ConnectionInfo[] };
-		type DebugEdgeLabelEntry = { connections: ConnectionInfo[]; isLeft: boolean; isTop: boolean; mx: number; my: number };
-		const pairMap = new Map<string, PairEntry>();
-		const debugEdgeLabels: DebugEdgeLabelEntry[] = [];
+		const pairMap = new Map<string, { a: string; b: string; aToB: boolean; bToA: boolean; connections: Array<{ from: string; to: string; ip: string; ifaceLabel: string; missingIface: boolean }> }>();
+		let debugEdgeLabels: Array<{ connections: typeof pairMap extends Map<string, infer V> ? V['connections'] : never; isLeft: boolean; isTop: boolean; mx: number; my: number }> | null = null;
 		edges.forEach(edge => {
 			if (!edge.source || !edge.target || edge.source === edge.target) return;
 			if (!positionById[edge.source] || !positionById[edge.target]) return;
-
+			
 			const a = edge.source < edge.target ? edge.source : edge.target;
 			const b = edge.source < edge.target ? edge.target : edge.source;
 			const key = `${a}|${b}`;
 			const entry = pairMap.get(key) || { a, b, aToB: false, bToA: false, connections: [] };
-
+			
 			if (edge.source === a) entry.aToB = true;
 			else entry.bToA = true;
 
-			const ip = edge.sendBackIp || '?';
+			const ip = edge.sendBackIp || edge.sendBackMultiaddr?.ip_address || '?';
 			const ifaceInfo = getInterfaceLabel(edge.source, ip);
 			entry.connections.push({
 				from: edge.source,
@@ -343,8 +338,9 @@ function wrapLine(text: string, maxLen: number): string[] {
 				// Determine which side of viewport based on edge midpoint
 				const isLeft = mx < centerX;
 				const isTop = my < safeCenterY;
-
+				
 				// Store for batch rendering after all edges processed
+				if (!debugEdgeLabels) debugEdgeLabels = [];
 				debugEdgeLabels.push({
 					connections: entry.connections,
 					isLeft,
@@ -385,32 +381,32 @@ function wrapLine(text: string, maxLen: number): string[] {
 			}
 			
 			// Group by quadrant: topLeft, topRight, bottomLeft, bottomRight
-			const quadrants: Record<string, DebugEdgeLabelEntry[]> = {
+			const quadrants: Record<string, typeof debugEdgeLabels> = {
 				topLeft: [],
 				topRight: [],
 				bottomLeft: [],
 				bottomRight: []
 			};
-
+			
 			debugEdgeLabels.forEach(edge => {
 				const key = (edge.isTop ? 'top' : 'bottom') + (edge.isLeft ? 'Left' : 'Right');
 				quadrants[key].push(edge);
 			});
-
+			
 			// Render each quadrant
-			Object.entries(quadrants).forEach(([quadrant, quadrantEdges]) => {
-				if (quadrantEdges.length === 0) return;
-
+			Object.entries(quadrants).forEach(([quadrant, edges]) => {
+				if (edges.length === 0) return;
+				
 				const isLeft = quadrant.includes('Left');
 				const isTop = quadrant.includes('top');
-
+				
 				let baseX = isLeft ? padding : width - padding;
 				let baseY = isTop ? padding : height - padding;
 				const textAnchor = isLeft ? 'start' : 'end';
-
+				
 				let currentY = baseY;
-
-				quadrantEdges.forEach(edge => {
+				
+				edges.forEach(edge => {
 					edge.connections.forEach(conn => {
 						const arrow = getArrow(conn.from, conn.to);
 						const label = `${arrow} ${conn.ip} ${conn.ifaceLabel}`;

dashboard/src/lib/stores/app.svelte.ts

@@ -71,68 +71,51 @@ export interface Instance {
 	};
 }
 
-interface RawNodeProfile {
-	modelId?: string;
-	chipId?: string;
-	friendlyName?: string;
-	networkInterfaces?: Array<{
-		name?: string;
-		ipAddress?: string;
-		addresses?: Array<{ address?: string } | string>;
-		ipv4?: string;
-		ipv6?: string;
-		ipAddresses?: string[];
-		ips?: string[];
-	}>;
-	memory?: {
-		ramTotal?: { inBytes: number };
-		ramAvailable?: { inBytes: number };
-		swapTotal?: { inBytes: number };
-		swapAvailable?: { inBytes: number };
-	};
-	system?: {
-		gpuUsage?: number;
-		temp?: number;
-		sysPower?: number;
-	};
+// Split state interfaces
+interface RawNodeIdentity {
+	modelId: string;
+	chipId: string;
+	friendlyName: string;
+}
+
+interface RawNodeMemory {
+	ramTotal: { inBytes: number };
+	ramAvailable: { inBytes: number };
+	swapTotal: { inBytes: number };
+	swapAvailable: { inBytes: number };
+}
+
+interface RawNodeSystem {
+	gpuUsage?: number;
+	temp?: number;
+	sysPower?: number;
+	pcpuUsage?: number;
+	ecpuUsage?: number;
+	anePower?: number;
+}
+
+interface RawNetworkInterface {
+	name: string;
+	ipAddress: string;
 }
 
 interface RawTopologyNode {
 	nodeId: string;
-	nodeProfile?: RawNodeProfile;
 }
 
-// New connection edge types from Python SocketConnection/RDMAConnection
-interface RawSocketConnection {
-	sinkMultiaddr?: {
-		address?: string;
-		// Multiaddr uses snake_case (no camelCase alias)
-		ip_address?: string;
-		ipAddress?: string; // fallback in case it changes
-		address_type?: string;
-		port?: number;
-	};
+interface RawTopologyConnection {
+	localNodeId: string;
+	sendBackNodeId: string;
+	sendBackMultiaddr?:
+		| { multiaddr?: string; address?: string; ip_address?: string }
+		| string;
 }
 
-interface RawRDMAConnection {
-	sourceRdmaIface?: string;
-	sinkRdmaIface?: string;
-}
-
-type RawConnectionEdge = RawSocketConnection | RawRDMAConnection;
-
-// New nested mapping format: { source: { sink: [edge1, edge2, ...] } }
-type RawConnectionsMap = Record<string, Record<string, RawConnectionEdge[]>>;
-
 interface RawTopology {
-	// nodes can be array of strings (node IDs) or array of objects with nodeId/nodeProfile
-	nodes: (string | RawTopologyNode)[];
-	// New nested mapping format
-	connections?: RawConnectionsMap;
+	nodes: RawTopologyNode[];
+	connections?: RawTopologyConnection[];
 }
 
-type RawNodeProfiles = Record<string, RawNodeProfile>;
-
 export interface DownloadProgress {
 	totalBytes: number;
 	downloadedBytes: number;
@@ -187,7 +170,11 @@ interface RawStateResponse {
 	>;
 	runners?: Record<string, unknown>;
 	downloads?: Record<string, unknown[]>;
-	nodeProfiles?: RawNodeProfiles;
+	// Split state fields
+	nodeIdentities?: Record<string, RawNodeIdentity>;
+	nodeMemories?: Record<string, RawNodeMemory>;
+	nodeSystems?: Record<string, RawNodeSystem>;
+	nodeNetworks?: Record<string, RawNetworkInterface[]>;
 }
 
 export interface MessageAttachment {
@@ -224,75 +211,41 @@ const STORAGE_KEY = "exo-conversations";
 
 function transformTopology(
 	raw: RawTopology,
-	profiles?: RawNodeProfiles,
+	identities?: Record<string, RawNodeIdentity>,
+	memories?: Record<string, RawNodeMemory>,
+	systems?: Record<string, RawNodeSystem>,
+	networks?: Record<string, RawNetworkInterface[]>,
 ): TopologyData {
 	const nodes: Record<string, NodeInfo> = {};
 	const edges: TopologyEdge[] = [];
 
-	// Handle nodes - can be array of strings (node IDs) or array of objects with nodeId/nodeProfile
 	for (const node of raw.nodes || []) {
-		// Determine the node ID - could be a string or an object with nodeId property
-		const nodeId = typeof node === "string" ? node : node.nodeId;
-		if (!nodeId) continue;
+		// Get split state fields (may be undefined if events haven't arrived yet)
+		const identity = identities?.[node.nodeId];
+		const memory = memories?.[node.nodeId];
+		const system = systems?.[node.nodeId];
+		const network = networks?.[node.nodeId];
 
-		// Get the profile - from the separate profiles map or from the node object itself
-		const profileFromMap = profiles?.[nodeId];
-		const profileFromNode =
-			typeof node === "object" ? node.nodeProfile : undefined;
-		const profile = { ...(profileFromNode ?? {}), ...(profileFromMap ?? {}) };
-
-		const ramTotal = profile?.memory?.ramTotal?.inBytes ?? 0;
-		const ramAvailable = profile?.memory?.ramAvailable?.inBytes ?? 0;
+		const ramTotal = memory?.ramTotal?.inBytes ?? 0;
+		const ramAvailable = memory?.ramAvailable?.inBytes ?? 0;
 		const ramUsage = Math.max(ramTotal - ramAvailable, 0);
 
-		const networkInterfaces = (profile?.networkInterfaces || []).map(
-			(iface) => {
-				const addresses: string[] = [];
-				if (iface.ipAddress && typeof iface.ipAddress === "string") {
-					addresses.push(iface.ipAddress);
-				}
-				if (Array.isArray(iface.addresses)) {
-					for (const addr of iface.addresses) {
-						if (typeof addr === "string") addresses.push(addr);
-						else if (addr && typeof addr === "object" && addr.address)
-							addresses.push(addr.address);
-					}
-				}
-				if (Array.isArray(iface.ipAddresses)) {
-					addresses.push(
-						...iface.ipAddresses.filter(
-							(a): a is string => typeof a === "string",
-						),
-					);
-				}
-				if (Array.isArray(iface.ips)) {
-					addresses.push(
-						...iface.ips.filter((a): a is string => typeof a === "string"),
-					);
-				}
-				if (iface.ipv4 && typeof iface.ipv4 === "string")
-					addresses.push(iface.ipv4);
-				if (iface.ipv6 && typeof iface.ipv6 === "string")
-					addresses.push(iface.ipv6);
-
-				return {
-					name: iface.name,
-					addresses: Array.from(new Set(addresses)),
-				};
-			},
-		);
+		const networkInterfaces = (network ?? []).map((iface) => ({
+			name: iface.name,
+			addresses: [iface.ipAddress],
+		}));
 
 		const ipToInterface: Record<string, string> = {};
 		for (const iface of networkInterfaces) {
-			for (const addr of iface.addresses || []) {
-				ipToInterface[addr] = iface.name ?? "";
+			for (const addr of iface.addresses) {
+				ipToInterface[addr] = iface.name;
 			}
 		}
 
-		nodes[nodeId] = {
+		nodes[node.nodeId] = {
 			system_info: {
-				model_id: profile?.modelId ?? "Unknown",
-				chip: profile?.chipId,
+				model_id: identity?.modelId ?? "Unknown",
+				chip: identity?.chipId,
 				memory: ramTotal,
 			},
 			network_interfaces: networkInterfaces,
@@ -303,48 +256,41 @@ function transformTopology(
 					ram_total: ramTotal,
 				},
 				temp:
-					profile?.system?.temp !== undefined
-						? { gpu_temp_avg: profile.system.temp }
+					system?.temp !== undefined
+						? { gpu_temp_avg: system.temp }
 						: undefined,
 				gpu_usage:
-					profile?.system?.gpuUsage !== undefined
-						? [0, profile.system.gpuUsage]
-						: undefined,
-				sys_power: profile?.system?.sysPower,
+					system?.gpuUsage !== undefined ? [0, system.gpuUsage] : undefined,
+				sys_power: system?.sysPower,
 			},
 			last_macmon_update: Date.now() / 1000,
-			friendly_name: profile?.friendlyName,
+			friendly_name: identity?.friendlyName,
 		};
 	}
 
-	// Handle connections - nested mapping format { source: { sink: [edges] } }
-	const connections = raw.connections;
-	if (connections && typeof connections === "object") {
-		for (const [source, sinks] of Object.entries(connections)) {
-			if (!sinks || typeof sinks !== "object") continue;
-			for (const [sink, edgeList] of Object.entries(sinks)) {
-				if (!Array.isArray(edgeList)) continue;
-				for (const edge of edgeList) {
-					// Extract IP from SocketConnection (uses snake_case: ip_address)
-					let sendBackIp: string | undefined;
-					if (edge && typeof edge === "object" && "sinkMultiaddr" in edge) {
-						const multiaddr = edge.sinkMultiaddr;
-						if (multiaddr) {
-							// Try both snake_case (actual) and camelCase (in case it changes)
-							sendBackIp =
-								multiaddr.ip_address ||
-								multiaddr.ipAddress ||
-								extractIpFromMultiaddr(multiaddr.address);
-						}
-					}
-					// RDMAConnection (sourceRdmaIface/sinkRdmaIface) has no IP - edge just shows connection exists
+	for (const conn of raw.connections || []) {
+		if (!conn.localNodeId || !conn.sendBackNodeId) continue;
+		if (conn.localNodeId === conn.sendBackNodeId) continue;
+		if (!nodes[conn.localNodeId] || !nodes[conn.sendBackNodeId]) continue;
 
-					if (nodes[source] && nodes[sink] && source !== sink) {
-						edges.push({ source, target: sink, sendBackIp });
-					}
-				}
+		let sendBackIp: string | undefined;
+		if (conn.sendBackMultiaddr) {
+			const multi = conn.sendBackMultiaddr;
+			if (typeof multi === "string") {
+				sendBackIp = extractIpFromMultiaddr(multi);
+			} else {
+				sendBackIp =
+					multi.ip_address ||
+					extractIpFromMultiaddr(multi.multiaddr) ||
+					extractIpFromMultiaddr(multi.address);
 			}
 		}
+
+		edges.push({
+			source: conn.localNodeId,
+			target: conn.sendBackNodeId,
+			sendBackIp,
+		});
 	}
 
 	return { nodes, edges };
@@ -898,7 +844,13 @@ class AppStore {
 			const data: RawStateResponse = await response.json();
 
 			if (data.topology) {
-				this.topologyData = transformTopology(data.topology, data.nodeProfiles);
+				this.topologyData = transformTopology(
+					data.topology,
+					data.nodeIdentities,
+					data.nodeMemories,
+					data.nodeSystems,
+					data.nodeNetworks,
+				);
 			}
 			if (data.instances) {
 				this.instances = data.instances;

dashboard/src/routes/+page.svelte

@@ -915,7 +915,7 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 		const runnerEntries = Object.entries(runnerToShard).map(([runnerId, shardWrapped]) => {
 			const [tag, shard] = getTagged(shardWrapped);
 			const meta = (shard as { modelMeta?: { worldSize?: number; nLayers?: number; deviceRank?: number } } | undefined);
-			const deviceRank = meta?.modelMeta?.deviceRank ?? 0;
+			const deviceRank = (meta?.deviceRank as number | undefined) ?? 0;
 			return { runnerId, tag, deviceRank };
 		});
 

src/exo/main.py

@@ -205,6 +205,14 @@ def main():
     logger.info("Starting EXO")
     logger.info(f"EXO_LIBP2P_NAMESPACE: {os.getenv('EXO_LIBP2P_NAMESPACE')}")
 
+    # Set FAST_SYNCH override env var for runner subprocesses
+    if args.fast_synch is True:
+        os.environ["EXO_FAST_SYNCH"] = "on"
+        logger.info("FAST_SYNCH forced ON")
+    elif args.fast_synch is False:
+        os.environ["EXO_FAST_SYNCH"] = "off"
+        logger.info("FAST_SYNCH forced OFF")
+
     node = anyio.run(Node.create, args)
     anyio.run(node.run)
     logger.info("EXO Shutdown complete")
@@ -218,6 +226,7 @@ class Args(CamelCaseModel):
     api_port: PositiveInt = 52415
     tb_only: bool = False
     no_worker: bool = False
+    fast_synch: bool | None = None  # None = auto, True = force on, False = force off
 
     @classmethod
     def parse(cls) -> Self:
@@ -259,6 +268,20 @@ class Args(CamelCaseModel):
             "--no-worker",
             action="store_true",
         )
+        fast_synch_group = parser.add_mutually_exclusive_group()
+        fast_synch_group.add_argument(
+            "--fast-synch",
+            action="store_true",
+            dest="fast_synch",
+            default=None,
+            help="Force MLX FAST_SYNCH on (for JACCL backend)",
+        )
+        fast_synch_group.add_argument(
+            "--no-fast-synch",
+            action="store_false",
+            dest="fast_synch",
+            help="Force MLX FAST_SYNCH off",
+        )
 
         args = parser.parse_args()
         return cls(**vars(args))  # pyright: ignore[reportAny] - We are intentionally validating here, we can't do it statically

src/exo/master/api.py

@@ -1,13 +1,14 @@
 import time
 from collections.abc import AsyncGenerator
+from http import HTTPStatus
 from typing import cast
 
 import anyio
-from anyio import create_task_group
+from anyio import BrokenResourceError, create_task_group
 from anyio.abc import TaskGroup
-from fastapi import FastAPI, HTTPException
+from fastapi import FastAPI, HTTPException, Request
 from fastapi.middleware.cors import CORSMiddleware
-from fastapi.responses import StreamingResponse
+from fastapi.responses import JSONResponse, StreamingResponse
 from fastapi.staticfiles import StaticFiles
 from hypercorn.asyncio import serve  # pyright: ignore[reportUnknownVariableType]
 from hypercorn.config import Config
@@ -29,6 +30,8 @@ from exo.shared.types.api import (
     CreateInstanceParams,
     CreateInstanceResponse,
     DeleteInstanceResponse,
+    ErrorInfo,
+    ErrorResponse,
     FinishReason,
     GenerationStats,
     ModelList,
@@ -49,7 +52,12 @@ from exo.shared.types.commands import (
     TaskFinished,
 )
 from exo.shared.types.common import CommandId, NodeId, SessionId
-from exo.shared.types.events import ChunkGenerated, Event, ForwarderEvent, IndexedEvent
+from exo.shared.types.events import (
+    ChunkGenerated,
+    Event,
+    ForwarderEvent,
+    IndexedEvent,
+)
 from exo.shared.types.memory import Memory
 from exo.shared.types.models import ModelId, ModelMetadata
 from exo.shared.types.state import State
@@ -115,6 +123,7 @@ class API:
         self.paused_ev: anyio.Event = anyio.Event()
 
         self.app = FastAPI()
+        self._setup_exception_handlers()
         self._setup_cors()
         self._setup_routes()
 
@@ -145,6 +154,20 @@ class API:
         self.paused_ev.set()
         self.paused_ev = anyio.Event()
 
+    def _setup_exception_handlers(self) -> None:
+        @self.app.exception_handler(HTTPException)
+        async def http_exception_handler(  # pyright: ignore[reportUnusedFunction]
+            _: Request, exc: HTTPException
+        ) -> JSONResponse:
+            err = ErrorResponse(
+                error=ErrorInfo(
+                    message=exc.detail,
+                    type=HTTPStatus(exc.status_code).phrase,
+                    code=exc.status_code,
+                )
+            )
+            return JSONResponse(err.model_dump(), status_code=exc.status_code)
+
     def _setup_cors(self) -> None:
         self.app.add_middleware(
             CORSMiddleware,
@@ -228,7 +251,6 @@ class API:
                     instance_meta=instance_meta,
                     min_nodes=min_nodes,
                 ),
-                node_profiles=self.state.node_profiles,
                 topology=self.state.topology,
                 current_instances=self.state.instances,
             )
@@ -284,7 +306,6 @@ class API:
                             instance_meta=instance_meta,
                             min_nodes=min_nodes,
                         ),
-                        node_profiles=self.state.node_profiles,
                         topology=self.state.topology,
                         current_instances=self.state.instances,
                     )
@@ -408,6 +429,18 @@ class API:
         """Generate chat completion stream as JSON strings."""
 
         async for chunk in self._chat_chunk_stream(command_id):
+            if chunk.finish_reason == "error":
+                error_response = ErrorResponse(
+                    error=ErrorInfo(
+                        message=chunk.error_message or "Internal server error",
+                        type="InternalServerError",
+                        code=500,
+                    )
+                )
+                yield f"data: {error_response.model_dump_json()}\n\n"
+                yield "data: [DONE]\n\n"
+                return
+
             chunk_response: ChatCompletionResponse = chunk_to_response(
                 chunk, command_id
             )
@@ -428,6 +461,12 @@ class API:
         finish_reason: FinishReason | None = None
 
         async for chunk in self._chat_chunk_stream(command_id):
+            if chunk.finish_reason == "error":
+                raise HTTPException(
+                    status_code=500,
+                    detail=chunk.error_message or "Internal server error",
+                )
+
             if model is None:
                 model = chunk.model
 
@@ -465,6 +504,12 @@ class API:
         stats: GenerationStats | None = None
 
         async for chunk in self._chat_chunk_stream(command_id):
+            if chunk.finish_reason == "error":
+                raise HTTPException(
+                    status_code=500,
+                    detail=chunk.error_message or "Internal server error",
+                )
+
             if model is None:
                 model = chunk.model
 
@@ -554,8 +599,8 @@ class API:
         """Calculate total available memory across all nodes in bytes."""
         total_available = Memory()
 
-        for profile in self.state.node_profiles.values():
-            total_available += profile.memory.ram_available
+        for memory in self.state.node_memories.values():
+            total_available += memory.ram_available
 
         return total_available
 
@@ -608,14 +653,14 @@ class API:
                 for idx, event in self.event_buffer.drain_indexed():
                     self._event_log.append(event)
                     self.state = apply(self.state, IndexedEvent(event=event, idx=idx))
-                    if (
-                        isinstance(event, ChunkGenerated)
-                        and event.command_id in self._chat_completion_queues
-                    ):
+                    if isinstance(event, ChunkGenerated):
                         assert isinstance(event.chunk, TokenChunk)
-                        await self._chat_completion_queues[event.command_id].send(
-                            event.chunk
-                        )
+                        queue = self._chat_completion_queues.get(event.command_id)
+                        if queue is not None:
+                            try:
+                                await queue.send(event.chunk)
+                            except BrokenResourceError:
+                                self._chat_completion_queues.pop(event.command_id, None)
 
     async def _pause_on_new_election(self):
         with self.election_receiver as ems:

src/exo/master/main.py

@@ -27,7 +27,6 @@ from exo.shared.types.events import (
     ForwarderEvent,
     IndexedEvent,
     InstanceDeleted,
-    NodeGatheredInfo,
     NodeTimedOut,
     TaskCreated,
     TaskDeleted,
@@ -159,7 +158,6 @@ class Master:
                                 command,
                                 self.state.topology,
                                 self.state.instances,
-                                self.state.node_profiles,
                             )
                             transition_events = get_transition_events(
                                 self.state.instances, placement
@@ -202,7 +200,9 @@ class Master:
     async def _plan(self) -> None:
         while True:
             # kill broken instances
-            connected_node_ids = set(self.state.topology.list_nodes())
+            connected_node_ids = set(
+                [x.node_id for x in self.state.topology.list_nodes()]
+            )
             for instance_id, instance in self.state.instances.items():
                 for node_id in instance.shard_assignments.node_to_runner:
                     if node_id not in connected_node_ids:
@@ -237,8 +237,6 @@ class Master:
                     self.state = apply(self.state, indexed)
 
                     event._master_time_stamp = datetime.now(tz=timezone.utc)  # pyright: ignore[reportPrivateUsage]
-                    if isinstance(event, NodeGatheredInfo):
-                        event.when = str(datetime.now(tz=timezone.utc))
 
                     self._event_log.append(event)
                     await self._send_event(indexed)

src/exo/master/placement.py

@@ -6,10 +6,9 @@ from typing import Sequence
 from loguru import logger
 
 from exo.master.placement_utils import (
-    Cycle,
     filter_cycles_by_memory,
+    get_mlx_ibv_devices_matrix,
     get_mlx_jaccl_coordinators,
-    get_mlx_jaccl_devices_matrix,
     get_mlx_ring_hosts_by_node,
     get_shard_assignments,
     get_smallest_cycles,
@@ -20,11 +19,10 @@ from exo.shared.types.commands import (
     DeleteInstance,
     PlaceInstance,
 )
-from exo.shared.types.common import NodeId
 from exo.shared.types.events import Event, InstanceCreated, InstanceDeleted
 from exo.shared.types.memory import Memory
 from exo.shared.types.models import ModelId
-from exo.shared.types.profiling import NodePerformanceProfile
+from exo.shared.types.topology import NodeInfo
 from exo.shared.types.worker.instances import (
     Instance,
     InstanceId,
@@ -54,14 +52,19 @@ def place_instance(
     command: PlaceInstance,
     topology: Topology,
     current_instances: Mapping[InstanceId, Instance],
-    node_profiles: Mapping[NodeId, NodePerformanceProfile],
 ) -> dict[InstanceId, Instance]:
+    all_nodes = list(topology.list_nodes())
+
+    logger.info("finding cycles:")
     cycles = topology.get_cycles()
-    candidate_cycles = list(filter(lambda it: len(it) >= command.min_nodes, cycles))
-    cycles_with_sufficient_memory = filter_cycles_by_memory(
-        candidate_cycles, node_profiles, command.model_meta.storage_size
+    singleton_cycles = [[node] for node in all_nodes]
+    candidate_cycles = list(
+        filter(lambda it: len(it) >= command.min_nodes, cycles + singleton_cycles)
     )
-    if len(cycles_with_sufficient_memory) == 0:
+    cycles_with_sufficient_memory = filter_cycles_by_memory(
+        candidate_cycles, command.model_meta.storage_size
+    )
+    if not cycles_with_sufficient_memory:
         raise ValueError("No cycles found with sufficient memory")
 
     if command.sharding == Sharding.Tensor:
@@ -89,38 +92,45 @@ def place_instance(
     smallest_cycles = get_smallest_cycles(cycles_with_sufficient_memory)
 
     smallest_tb_cycles = [
-        cycle for cycle in smallest_cycles if topology.is_thunderbolt_cycle(cycle)
+        cycle
+        for cycle in smallest_cycles
+        if topology.get_subgraph_from_nodes(cycle).is_thunderbolt_cycle(cycle)
     ]
 
     if smallest_tb_cycles != []:
         smallest_cycles = smallest_tb_cycles
 
-    cycles_with_leaf_nodes: list[Cycle] = [
+    cycles_with_leaf_nodes: list[list[NodeInfo]] = [
         cycle
         for cycle in smallest_cycles
-        if any(topology.node_is_leaf(node_id) for node_id in cycle)
+        if any(topology.node_is_leaf(node.node_id) for node in cycle)
     ]
 
     selected_cycle = max(
         cycles_with_leaf_nodes if cycles_with_leaf_nodes != [] else smallest_cycles,
         key=lambda cycle: sum(
-            (node_profiles[node_id].memory.ram_available for node_id in cycle),
+            (
+                node.node_profile.memory.ram_available
+                for node in cycle
+                if node.node_profile is not None
+                and node.node_profile.memory is not None
+            ),
             start=Memory(),
         ),
     )
 
     shard_assignments = get_shard_assignments(
-        command.model_meta, selected_cycle, command.sharding, node_profiles
+        command.model_meta, selected_cycle, command.sharding
     )
 
-    cycle_digraph: Topology = topology.get_subgraph_from_nodes(selected_cycle.node_ids)
+    cycle_digraph: Topology = topology.get_subgraph_from_nodes(selected_cycle)
 
     instance_id = InstanceId()
     target_instances = dict(deepcopy(current_instances))
 
     if len(selected_cycle) == 1:
         logger.warning(
-            "You have likely selected jaccl for a single node instance; falling back to MlxRing"
+            "You have likely selected ibv for a single node instance; falling back to MlxRing"
         )
 
         command.instance_meta = InstanceMeta.MlxRing
@@ -128,20 +138,19 @@ def place_instance(
     # TODO: Single node instances
     match command.instance_meta:
         case InstanceMeta.MlxJaccl:
-            mlx_jaccl_devices = get_mlx_jaccl_devices_matrix(
-                [node_id for node_id in selected_cycle],
+            mlx_ibv_devices = get_mlx_ibv_devices_matrix(
+                selected_cycle,
                 cycle_digraph,
             )
             mlx_jaccl_coordinators = get_mlx_jaccl_coordinators(
-                coordinator=selected_cycle.node_ids[0],
+                selected_cycle,
                 coordinator_port=random_ephemeral_port(),
                 cycle_digraph=cycle_digraph,
-                node_profiles=node_profiles,
             )
             target_instances[instance_id] = MlxJacclInstance(
                 instance_id=instance_id,
                 shard_assignments=shard_assignments,
-                jaccl_devices=mlx_jaccl_devices,
+                ibv_devices=mlx_ibv_devices,
                 jaccl_coordinators=mlx_jaccl_coordinators,
             )
         case InstanceMeta.MlxRing:
@@ -150,7 +159,6 @@ def place_instance(
                 selected_cycle=selected_cycle,
                 cycle_digraph=cycle_digraph,
                 ephemeral_port=ephemeral_port,
-                node_profiles=node_profiles,
             )
             target_instances[instance_id] = MlxRingInstance(
                 instance_id=instance_id,

src/exo/master/placement_utils.py

@@ -1,13 +1,15 @@
-from collections.abc import Generator, Mapping
+from collections.abc import Generator
+from typing import TypeGuard, cast
 
 from loguru import logger
+from pydantic import BaseModel
 
 from exo.shared.topology import Topology
 from exo.shared.types.common import Host, NodeId
 from exo.shared.types.memory import Memory
 from exo.shared.types.models import ModelMetadata
 from exo.shared.types.profiling import NodePerformanceProfile
-from exo.shared.types.topology import Cycle, RDMAConnection, SocketConnection
+from exo.shared.types.topology import NodeInfo
 from exo.shared.types.worker.runners import RunnerId, ShardAssignments
 from exo.shared.types.worker.shards import (
     PipelineShardMetadata,
@@ -17,55 +19,74 @@ from exo.shared.types.worker.shards import (
 )
 
 
+class NodeWithProfile(BaseModel):
+    node_id: NodeId
+    node_profile: NodePerformanceProfile
+
+
+def narrow_all_nodes(nodes: list[NodeInfo]) -> TypeGuard[list[NodeWithProfile]]:
+    return all(
+        node.node_profile is not None and node.node_profile.memory is not None
+        for node in nodes
+    )
+
+
 def filter_cycles_by_memory(
-    cycles: list[Cycle],
-    node_profiles: Mapping[NodeId, NodePerformanceProfile],
-    required_memory: Memory,
-) -> list[Cycle]:
-    filtered_cycles: list[Cycle] = []
+    cycles: list[list[NodeInfo]], required_memory: Memory
+) -> list[list[NodeInfo]]:
+    filtered_cycles: list[list[NodeInfo]] = []
     for cycle in cycles:
-        if not all(node in node_profiles for node in cycle):
+        if not narrow_all_nodes(cycle):
             continue
 
+        # narrow_all_nodes guarantees memory is not None
         total_mem = sum(
-            (node_profiles[node_id].memory.ram_available for node_id in cycle.node_ids),
+            (
+                node.node_profile.memory.ram_available
+                for node in cycle
+                if node.node_profile.memory is not None
+            ),
             start=Memory(),
         )
         if total_mem >= required_memory:
-            filtered_cycles.append(cycle)
+            filtered_cycles.append(cast(list[NodeInfo], cycle))
     return filtered_cycles
 
 
-def get_smallest_cycles(
-    cycles: list[Cycle],
-) -> list[Cycle]:
+def get_smallest_cycles(cycles: list[list[NodeInfo]]) -> list[list[NodeInfo]]:
     min_nodes = min(len(cycle) for cycle in cycles)
     return [cycle for cycle in cycles if len(cycle) == min_nodes]
 
 
 def get_shard_assignments_for_pipeline_parallel(
     model_meta: ModelMetadata,
-    cycle: Cycle,
-    node_profiles: Mapping[NodeId, NodePerformanceProfile],
+    selected_cycle: list[NodeWithProfile],
 ):
+    # NodeWithProfile guarantees memory is not None
     cycle_memory = sum(
-        (node_profiles[node_id].memory.ram_available for node_id in cycle.node_ids),
+        (
+            node.node_profile.memory.ram_available
+            for node in selected_cycle
+            if node.node_profile.memory is not None
+        ),
         start=Memory(),
     )
     total_layers = model_meta.n_layers
-    world_size = len(cycle)
+    world_size = len(selected_cycle)
     runner_to_shard: dict[RunnerId, ShardMetadata] = {}
     node_to_runner: dict[NodeId, RunnerId] = {}
 
     layers_assigned = 0
-    for i, node_id in enumerate(cycle):
-        if i == len(cycle) - 1:
+    for i, node in enumerate(selected_cycle):
+        if i == len(selected_cycle) - 1:
             node_layers = total_layers - layers_assigned
         else:
+            # NodeWithProfile guarantees memory is not None
+            assert node.node_profile.memory is not None
             node_layers = round(
                 total_layers
                 * (
-                    node_profiles[node_id].memory.ram_available.in_bytes
+                    node.node_profile.memory.ram_available.in_bytes
                     / cycle_memory.in_bytes
                 )
             )
@@ -83,7 +104,7 @@ def get_shard_assignments_for_pipeline_parallel(
         )
 
         runner_to_shard[runner_id] = shard
-        node_to_runner[node_id] = runner_id
+        node_to_runner[node.node_id] = runner_id
         layers_assigned += node_layers
 
     shard_assignments = ShardAssignments(
@@ -97,14 +118,14 @@ def get_shard_assignments_for_pipeline_parallel(
 
 def get_shard_assignments_for_tensor_parallel(
     model_meta: ModelMetadata,
-    cycle: Cycle,
+    selected_cycle: list[NodeWithProfile],
 ):
     total_layers = model_meta.n_layers
-    world_size = len(cycle)
+    world_size = len(selected_cycle)
     runner_to_shard: dict[RunnerId, ShardMetadata] = {}
     node_to_runner: dict[NodeId, RunnerId] = {}
 
-    for i, node_id in enumerate(cycle):
+    for i, node in enumerate(selected_cycle):
         shard = TensorShardMetadata(
             model_meta=model_meta,
             device_rank=i,
@@ -117,7 +138,7 @@ def get_shard_assignments_for_tensor_parallel(
         runner_id = RunnerId()
 
         runner_to_shard[runner_id] = shard
-        node_to_runner[node_id] = runner_id
+        node_to_runner[node.node_id] = runner_id
 
     shard_assignments = ShardAssignments(
         model_id=model_meta.model_id,
@@ -130,21 +151,21 @@ def get_shard_assignments_for_tensor_parallel(
 
 def get_shard_assignments(
     model_meta: ModelMetadata,
-    cycle: Cycle,
+    selected_cycle: list[NodeInfo],
     sharding: Sharding,
-    node_profiles: Mapping[NodeId, NodePerformanceProfile],
 ) -> ShardAssignments:
+    if not narrow_all_nodes(selected_cycle):
+        raise ValueError("All nodes must have profiles to create shard assignments")
     match sharding:
         case Sharding.Pipeline:
             return get_shard_assignments_for_pipeline_parallel(
                 model_meta=model_meta,
-                cycle=cycle,
-                node_profiles=node_profiles,
+                selected_cycle=selected_cycle,
             )
         case Sharding.Tensor:
             return get_shard_assignments_for_tensor_parallel(
                 model_meta=model_meta,
-                cycle=cycle,
+                selected_cycle=selected_cycle,
             )
 
 
@@ -159,40 +180,38 @@ def get_hosts_from_subgraph(cycle_digraph: Topology) -> list[Host]:
             )
         return []
 
-    cycle = cycles[0]
-
     get_thunderbolt = False
-    if cycle_digraph.is_thunderbolt_cycle(cycle):
+    if cycle_digraph.is_thunderbolt_cycle(cycles[0]):
         get_thunderbolt = True
 
     logger.info(f"Using thunderbolt cycle: {get_thunderbolt}")
 
+    cycle = cycles[0]
     hosts: list[Host] = []
     for i in range(len(cycle)):
-        current_node = cycle.node_ids[i]
-        next_node = cycle.node_ids[(i + 1) % len(cycle)]
+        current_node = cycle[i]
+        next_node = cycle[(i + 1) % len(cycle)]
 
-        for connection in cycle_digraph.get_all_connections_between(
-            source=current_node, sink=next_node
-        ):
-            if not isinstance(connection, SocketConnection):
-                continue
-
-            if get_thunderbolt and not connection.is_thunderbolt():
-                continue
-
-            host = Host(
-                ip=connection.sink_multiaddr.ip_address,
-                port=connection.sink_multiaddr.port,
-            )
-            hosts.append(host)
-            break
+        for connection in cycle_digraph.list_connections():
+            if (
+                connection.local_node_id == current_node.node_id
+                and connection.send_back_node_id == next_node.node_id
+            ):
+                if get_thunderbolt and not connection.is_thunderbolt():
+                    continue
+                assert connection.send_back_multiaddr is not None
+                host = Host(
+                    ip=connection.send_back_multiaddr.ip_address,
+                    port=connection.send_back_multiaddr.port,
+                )
+                hosts.append(host)
+                break
 
     return hosts
 
 
-def get_mlx_jaccl_devices_matrix(
-    selected_cycle: list[NodeId],
+def get_mlx_ibv_devices_matrix(
+    selected_cycle: list[NodeInfo],
     cycle_digraph: Topology,
 ) -> list[list[str | None]]:
     """Build connectivity matrix mapping device i to device j via RDMA interface names.
@@ -211,37 +230,72 @@ def get_mlx_jaccl_devices_matrix(
             if i == j:
                 continue
 
-            for conn in cycle_digraph.get_all_connections_between(node_i, node_j):
-                if isinstance(conn, RDMAConnection):
-                    matrix[i][j] = conn.source_rdma_iface
+            # Find the IP J uses to talk to I
+            for connection_ip, _ in _find_connection_ip(node_j, node_i, cycle_digraph):
+                # This is a local IP on I, which is attached to an interface: find that interface
+                if interface_name := _find_rdma_interface_name_for_ip(
+                    connection_ip, node_i
+                ):
+                    matrix[i][j] = interface_name
+                    logger.info(
+                        f"Interface name for {connection_ip} on {node_i.node_id}: {interface_name}"
+                    )
                     break
             else:
                 logger.warning(
-                    f"Failed to find interface name between {node_i} and {node_j}"
+                    f"Failed to find interface name between {node_i.node_id} and {node_j.node_id}"
                 )
                 raise ValueError(
-                    "Current jaccl backend requires all-to-all RDMA connections"
+                    "Current ibv backend requires all-to-all rdma connections"
                 )
 
     return matrix
 
 
 def _find_connection_ip(
-    node_i: NodeId,
-    node_j: NodeId,
+    node_i: NodeInfo,
+    node_j: NodeInfo,
     cycle_digraph: Topology,
 ) -> Generator[tuple[str, bool]]:
-    """Find all IP addresses that connect node i to node j."""
-    for connection in cycle_digraph.get_all_connections_between(node_i, node_j):
-        if isinstance(connection, SocketConnection):
-            yield connection.sink_multiaddr.ip_address, connection.is_thunderbolt()
+    """Find all IP addresses that connect node i to node j, with thunderbolt flag."""
+    for connection in cycle_digraph.list_connections():
+        if (
+            connection.local_node_id == node_i.node_id
+            and connection.send_back_node_id == node_j.node_id
+        ):
+            yield connection.send_back_multiaddr.ip_address, connection.is_thunderbolt()
+
+
+def _find_rdma_interface_name_for_ip(
+    ip_address: str,
+    node_info: NodeInfo,
+) -> str | None:
+    if node_info.node_profile is None:
+        return None
+
+    logger.info(f"Searching {node_info.node_id} for ip {ip_address}:")
+    for interface in node_info.node_profile.network_interfaces:
+        if interface.name not in ["en2", "en3", "en4", "en5", "en6", "en7"]:
+            continue
+        logger.info(f" | {interface.name}: {interface.ip_address}")
+        if interface.ip_address != ip_address:
+            continue
+
+        logger.info("Found")
+        return f"rdma_{interface.name}"
+
+    return None
 
 
 def _find_interface_name_for_ip(
-    ip_address: str, node_profile: NodePerformanceProfile
+    ip_address: str,
+    node_info: NodeInfo,
 ) -> str | None:
     """Find the interface name for an IP address on a node (any interface)."""
-    for interface in node_profile.network_interfaces:
+    if node_info.node_profile is None:
+        return None
+
+    for interface in node_info.node_profile.network_interfaces:
         if interface.ip_address == ip_address:
             return interface.name
 
@@ -249,10 +303,7 @@ def _find_interface_name_for_ip(
 
 
 def _find_ip_prioritised(
-    node_id: NodeId,
-    other_node_id: NodeId,
-    cycle_digraph: Topology,
-    node_profiles: Mapping[NodeId, NodePerformanceProfile],
+    node: NodeInfo, other_node: NodeInfo, cycle_digraph: Topology
 ) -> str | None:
     # TODO: Actually prioritize in the correct Ethernet > Wifi > Non-TB > TB order.
     """Find an IP address between nodes with prioritization.
@@ -263,12 +314,9 @@ def _find_ip_prioritised(
     3. Non-Thunderbolt connections
     4. Any other IP address
     """
-    ips = list(_find_connection_ip(node_id, other_node_id, cycle_digraph))
+    ips = list(_find_connection_ip(node, other_node, cycle_digraph))
     # We expect a unique iface -> ip mapping
-    iface_map = {
-        _find_interface_name_for_ip(ip, node_profiles[other_node_id]): ip
-        for ip, _ in ips
-    }
+    iface_map = {_find_interface_name_for_ip(ip, other_node): ip for ip, _ in ips}
 
     en0_ip = iface_map.get("en0")
     if en0_ip:
@@ -292,10 +340,9 @@ def _find_ip_prioritised(
 
 
 def get_mlx_ring_hosts_by_node(
-    selected_cycle: Cycle,
+    selected_cycle: list[NodeInfo],
     cycle_digraph: Topology,
     ephemeral_port: int,
-    node_profiles: Mapping[NodeId, NodePerformanceProfile],
 ) -> dict[NodeId, list[Host]]:
     """Generate per-node host lists for MLX ring backend.
 
@@ -310,13 +357,14 @@ def get_mlx_ring_hosts_by_node(
 
     hosts_by_node: dict[NodeId, list[Host]] = {}
 
-    for rank, node_id in enumerate(selected_cycle):
+    for rank, node in enumerate(selected_cycle):
+        node_id = node.node_id
         left_rank = (rank - 1) % world_size
         right_rank = (rank + 1) % world_size
 
         hosts_for_node: list[Host] = []
 
-        for idx, other_node_id in enumerate(selected_cycle):
+        for idx, other_node in enumerate(selected_cycle):
             if idx == rank:
                 hosts_for_node.append(Host(ip="0.0.0.0", port=ephemeral_port))
                 continue
@@ -326,12 +374,10 @@ def get_mlx_ring_hosts_by_node(
                 hosts_for_node.append(Host(ip="198.51.100.1", port=0))
                 continue
 
-            connection_ip = _find_ip_prioritised(
-                node_id, other_node_id, cycle_digraph, node_profiles
-            )
+            connection_ip = _find_ip_prioritised(node, other_node, cycle_digraph)
             if connection_ip is None:
                 logger.warning(
-                    f"Failed to find prioritised connection IP between {node_id} and {other_node_id}"
+                    f"Failed to find prioritised connection IP between {node_id} and {other_node.node_id}"
                 )
                 raise ValueError(
                     "MLX ring backend requires connectivity between neighbouring nodes"
@@ -345,34 +391,31 @@ def get_mlx_ring_hosts_by_node(
 
 
 def get_mlx_jaccl_coordinators(
-    coordinator: NodeId,
+    selected_cycle: list[NodeInfo],
     coordinator_port: int,
     cycle_digraph: Topology,
-    node_profiles: Mapping[NodeId, NodePerformanceProfile],
 ) -> dict[NodeId, str]:
-    """Get the coordinator addresses for MLX JACCL (rank 0 device).
+    """Get the coordinator addresses for MLX Jaccl (rank 0 device).
 
     Select an IP address that each node can reach for the rank 0 node. Returns
     address in format "X.X.X.X:PORT" per node.
     """
-    logger.info(f"Selecting coordinator: {coordinator}")
+    rank_0_node = selected_cycle[0]
+    logger.debug(f"Selecting coordinator from rank 0 node: {rank_0_node.node_id}")
 
-    def get_ip_for_node(n: NodeId) -> str:
-        if n == coordinator:
+    def get_ip_for_node(n: NodeInfo) -> str:
+        if n.node_id == rank_0_node.node_id:
             return "0.0.0.0"
 
-        ip = _find_ip_prioritised(n, coordinator, cycle_digraph, node_profiles)
-        if ip is not None:
+        ip = _find_ip_prioritised(n, rank_0_node, cycle_digraph)
+        if ip:
             return ip
 
         logger.warning(
-            f"Failed to find directly connected ip between {n} and {coordinator}"
-        )
-        raise ValueError(
-            "Current jaccl backend requires all participating devices to be able to communicate"
+            f"Failed to find directly connected ip between {n.node_id} and {rank_0_node.node_id}"
         )
+        raise ValueError("Current ibv backend requires all-to-all rdma connections")
 
     return {
-        n: f"{get_ip_for_node(n)}:{coordinator_port}"
-        for n in cycle_digraph.list_nodes()
+        n.node_id: f"{get_ip_for_node(n)}:{coordinator_port}" for n in selected_cycle
     }

src/exo/master/tests/conftest.py

@@ -1,39 +1,67 @@
+from typing import Callable
+
+import pytest
+
+from exo.shared.types.common import NodeId
 from exo.shared.types.multiaddr import Multiaddr
 from exo.shared.types.profiling import (
-    MemoryUsage,
-    NetworkInterfaceInfo,
+    MemoryPerformanceProfile,
     NodePerformanceProfile,
     SystemPerformanceProfile,
 )
-from exo.shared.types.topology import RDMAConnection, SocketConnection
+from exo.shared.types.topology import Connection, ConnectionProfile, NodeInfo
 
 
-def create_node_profile(memory: int) -> NodePerformanceProfile:
-    return NodePerformanceProfile(
-        model_id="test",
-        chip_id="test",
-        friendly_name="test",
-        memory=MemoryUsage.from_bytes(
-            ram_total=1000,
-            ram_available=memory,
-            swap_total=1000,
-            swap_available=1000,
-        ),
-        network_interfaces=[
-            NetworkInterfaceInfo(name="en0", ip_address=f"169.254.0.{i}")
-            for i in range(10)
-        ],
-        system=SystemPerformanceProfile(),
-    )
+@pytest.fixture
+def create_node():
+    def _create_node(memory: int, node_id: NodeId | None = None) -> NodeInfo:
+        if node_id is None:
+            node_id = NodeId()
+        return NodeInfo(
+            node_id=node_id,
+            node_profile=NodePerformanceProfile(
+                model_id="test",
+                chip_id="test",
+                friendly_name="test",
+                memory=MemoryPerformanceProfile.from_bytes(
+                    ram_total=1000,
+                    ram_available=memory,
+                    swap_total=1000,
+                    swap_available=1000,
+                ),
+                network_interfaces=[],
+                system=SystemPerformanceProfile(),
+            ),
+        )
+
+    return _create_node
 
 
-def create_socket_connection(ip: int, sink_port: int = 1234) -> SocketConnection:
-    return SocketConnection(
-        sink_multiaddr=Multiaddr(address=f"/ip4/169.254.0.{ip}/tcp/{sink_port}"),
-    )
+# TODO: this is a hack to get the port for the send_back_multiaddr
+@pytest.fixture
+def create_connection() -> Callable[[NodeId, NodeId, int | None], Connection]:
+    port_counter = 1235
+    ip_counter = 1
 
+    def _create_connection(
+        source_node_id: NodeId, sink_node_id: NodeId, send_back_port: int | None = None
+    ) -> Connection:
+        nonlocal port_counter
+        nonlocal ip_counter
+        # assign unique ips
+        ip_counter += 1
+        if send_back_port is None:
+            send_back_port = port_counter
+            port_counter += 1
+        return Connection(
+            local_node_id=source_node_id,
+            send_back_node_id=sink_node_id,
+            send_back_multiaddr=Multiaddr(
+                address=f"/ip4/169.254.0.{ip_counter}/tcp/{send_back_port}"
+            ),
+            connection_profile=ConnectionProfile(
+                throughput=1000, latency=1000, jitter=1000
+            ),
+        )
 
-def create_rdma_connection(iface: int) -> RDMAConnection:
-    return RDMAConnection(
-        source_rdma_iface=f"rdma_en{iface}", sink_rdma_iface=f"rdma_en{iface}"
-    )
+    return _create_connection

src/exo/master/tests/test_api_error_handling.py

@@ -0,0 +1,107 @@
+# pyright: reportUnusedFunction=false, reportAny=false
+from typing import Any, get_args
+
+from fastapi import FastAPI, HTTPException
+from fastapi.testclient import TestClient
+
+from exo.shared.types.api import ErrorInfo, ErrorResponse, FinishReason
+from exo.shared.types.chunks import TokenChunk
+from exo.worker.tests.constants import MODEL_A_ID
+
+
+def test_http_exception_handler_formats_openai_style() -> None:
+    """Test that HTTPException is converted to OpenAI-style error format."""
+    from exo.master.api import API
+
+    app = FastAPI()
+
+    # Setup exception handler
+    api = object.__new__(API)
+    api.app = app
+    api._setup_exception_handlers()  # pyright: ignore[reportPrivateUsage]
+
+    # Add test routes that raise HTTPException
+    @app.get("/test-error")
+    async def _test_error() -> None:
+        raise HTTPException(status_code=500, detail="Test error message")
+
+    @app.get("/test-not-found")
+    async def _test_not_found() -> None:
+        raise HTTPException(status_code=404, detail="Resource not found")
+
+    client = TestClient(app)
+
+    # Test 500 error
+    response = client.get("/test-error")
+    assert response.status_code == 500
+    data: dict[str, Any] = response.json()
+    assert "error" in data
+    assert data["error"]["message"] == "Test error message"
+    assert data["error"]["type"] == "Internal Server Error"
+    assert data["error"]["code"] == 500
+
+    # Test 404 error
+    response = client.get("/test-not-found")
+    assert response.status_code == 404
+    data = response.json()
+    assert "error" in data
+    assert data["error"]["message"] == "Resource not found"
+    assert data["error"]["type"] == "Not Found"
+    assert data["error"]["code"] == 404
+
+
+def test_finish_reason_includes_error() -> None:
+    valid_reasons = get_args(FinishReason)
+    assert "error" in valid_reasons
+
+
+def test_token_chunk_with_error_fields() -> None:
+    chunk = TokenChunk(
+        idx=0,
+        model=MODEL_A_ID,
+        text="",
+        token_id=0,
+        finish_reason="error",
+        error_message="Something went wrong",
+    )
+
+    assert chunk.finish_reason == "error"
+    assert chunk.error_message == "Something went wrong"
+
+
+def test_token_chunk_without_error() -> None:
+    chunk = TokenChunk(
+        idx=1,
+        model=MODEL_A_ID,
+        text="Hello",
+        token_id=42,
+        finish_reason=None,
+    )
+
+    assert chunk.finish_reason is None
+    assert chunk.error_message is None
+
+
+def test_error_response_construction() -> None:
+    error_response = ErrorResponse(
+        error=ErrorInfo(
+            message="Generation failed",
+            type="InternalServerError",
+            code=500,
+        )
+    )
+
+    assert error_response.error.message == "Generation failed"
+    assert error_response.error.code == 500
+
+
+def test_normal_finish_reasons_still_work() -> None:
+    for reason in ["stop", "length", "tool_calls", "content_filter", "function_call"]:
+        chunk = TokenChunk(
+            idx=0,
+            model=MODEL_A_ID,
+            text="done",
+            token_id=100,
+            finish_reason=reason,  # type: ignore[arg-type]
+        )
+        assert chunk.finish_reason == reason

src/exo/master/tests/test_master.py

@@ -19,14 +19,13 @@ from exo.shared.types.events import (
     ForwarderEvent,
     IndexedEvent,
     InstanceCreated,
-    NodeGatheredInfo,
+    NodeIdentityMeasured,
+    NodeMemoryMeasured,
     TaskCreated,
 )
 from exo.shared.types.memory import Memory
 from exo.shared.types.models import ModelId, ModelMetadata
-from exo.shared.types.profiling import (
-    MemoryUsage,
-)
+from exo.shared.types.profiling import MemoryPerformanceProfile
 from exo.shared.types.tasks import ChatCompletion as ChatCompletionTask
 from exo.shared.types.tasks import TaskStatus
 from exo.shared.types.worker.instances import (
@@ -73,18 +72,35 @@ async def test_master():
         tg.start_soon(master.run)
 
         sender_node_id = NodeId(f"{keypair.to_peer_id().to_base58()}_sender")
-        # inject a NodePerformanceProfile event
-        logger.info("inject a NodePerformanceProfile event")
+        # inject NodeIdentityMeasured and NodeMemoryMeasured events
+        logger.info("inject NodeIdentityMeasured event")
         await local_event_sender.send(
             ForwarderEvent(
                 origin_idx=0,
                 origin=sender_node_id,
                 session=session_id,
                 event=(
-                    NodeGatheredInfo(
+                    NodeIdentityMeasured(
                         when=str(datetime.now(tz=timezone.utc)),
                         node_id=node_id,
-                        info=MemoryUsage(
+                        model_id="maccy",
+                        chip_id="arm",
+                        friendly_name="test",
+                    )
+                ),
+            )
+        )
+        logger.info("inject NodeMemoryMeasured event")
+        await local_event_sender.send(
+            ForwarderEvent(
+                origin_idx=1,
+                origin=sender_node_id,
+                session=session_id,
+                event=(
+                    NodeMemoryMeasured(
+                        when=str(datetime.now(tz=timezone.utc)),
+                        node_id=node_id,
+                        memory=MemoryPerformanceProfile(
                             ram_total=Memory.from_bytes(678948 * 1024),
                             ram_available=Memory.from_bytes(678948 * 1024),
                             swap_total=Memory.from_bytes(0),
@@ -99,7 +115,7 @@ async def test_master():
         logger.info("wait for initial topology event")
         while len(list(master.state.topology.list_nodes())) == 0:
             await anyio.sleep(0.001)
-        while len(master.state.node_profiles) == 0:
+        while len(master.state.node_identities) == 0:
             await anyio.sleep(0.001)
 
         logger.info("inject a CreateInstance Command")
@@ -146,17 +162,19 @@ async def test_master():
                 ),
             )
         )
-        while len(_get_events()) < 3:
+        while len(_get_events()) < 4:
             await anyio.sleep(0.01)
 
         events = _get_events()
-        assert len(events) == 3
+        assert len(events) == 4
         assert events[0].idx == 0
         assert events[1].idx == 1
         assert events[2].idx == 2
-        assert isinstance(events[0].event, NodeGatheredInfo)
-        assert isinstance(events[1].event, InstanceCreated)
-        created_instance = events[1].event.instance
+        assert events[3].idx == 3
+        assert isinstance(events[0].event, NodeIdentityMeasured)
+        assert isinstance(events[1].event, NodeMemoryMeasured)
+        assert isinstance(events[2].event, InstanceCreated)
+        created_instance = events[2].event.instance
         assert isinstance(created_instance, MlxRingInstance)
         runner_id = list(created_instance.shard_assignments.runner_to_shard.keys())[0]
         # Validate the shard assignments
@@ -188,10 +206,10 @@ async def test_master():
         assert len(created_instance.hosts_by_node[node_id]) == 1
         assert created_instance.hosts_by_node[node_id][0].ip == "0.0.0.0"
         assert created_instance.ephemeral_port > 0
-        assert isinstance(events[2].event, TaskCreated)
-        assert events[2].event.task.task_status == TaskStatus.Pending
-        assert isinstance(events[2].event.task, ChatCompletionTask)
-        assert events[2].event.task.task_params == ChatCompletionTaskParams(
+        assert isinstance(events[3].event, TaskCreated)
+        assert events[3].event.task.task_status == TaskStatus.Pending
+        assert isinstance(events[3].event.task, ChatCompletionTask)
+        assert events[3].event.task.task_params == ChatCompletionTaskParams(
             model="llama-3.2-1b",
             messages=[
                 ChatCompletionMessage(role="user", content="Hello, how are you?")

src/exo/master/tests/test_placement.py

@@ -1,23 +1,20 @@
+from typing import Callable
+
 import pytest
+from loguru import logger
 
 from exo.master.placement import (
     get_transition_events,
     place_instance,
 )
-from exo.master.tests.conftest import (
-    create_node_profile,
-    create_rdma_connection,
-    create_socket_connection,
-)
 from exo.shared.topology import Topology
 from exo.shared.types.commands import PlaceInstance
 from exo.shared.types.common import CommandId, NodeId
 from exo.shared.types.events import InstanceCreated, InstanceDeleted
 from exo.shared.types.memory import Memory
 from exo.shared.types.models import ModelId, ModelMetadata
-from exo.shared.types.multiaddr import Multiaddr
-from exo.shared.types.profiling import NetworkInterfaceInfo
-from exo.shared.types.topology import Connection, SocketConnection
+from exo.shared.types.profiling import NetworkInterfaceInfo, NodePerformanceProfile
+from exo.shared.types.topology import Connection, NodeInfo
 from exo.shared.types.worker.instances import (
     Instance,
     InstanceId,
@@ -29,6 +26,11 @@ from exo.shared.types.worker.runners import ShardAssignments
 from exo.shared.types.worker.shards import Sharding
 
 
+@pytest.fixture
+def topology() -> Topology:
+    return Topology()
+
+
 @pytest.fixture
 def instance() -> Instance:
     return MlxRingInstance(
@@ -75,57 +77,34 @@ def test_get_instance_placements_create_instance(
     available_memory: tuple[int, int, int],
     total_layers: int,
     expected_layers: tuple[int, int, int],
+    topology: Topology,
     model_meta: ModelMetadata,
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+    create_connection: Callable[[NodeId, NodeId], Connection],
 ):
     # arrange
     model_meta.n_layers = total_layers
     model_meta.storage_size.in_bytes = sum(
         available_memory
     )  # make it exactly fit across all nodes
-    topology = Topology()
 
     cic = place_instance_command(model_meta)
     node_id_a = NodeId()
     node_id_b = NodeId()
     node_id_c = NodeId()
-
-    # fully connected (directed) between the 3 nodes
-    conn_a_b = Connection(
-        source=node_id_a, sink=node_id_b, edge=create_socket_connection(1)
-    )
-    conn_b_c = Connection(
-        source=node_id_b, sink=node_id_c, edge=create_socket_connection(2)
-    )
-    conn_c_a = Connection(
-        source=node_id_c, sink=node_id_a, edge=create_socket_connection(3)
-    )
-    conn_c_b = Connection(
-        source=node_id_c, sink=node_id_b, edge=create_socket_connection(4)
-    )
-    conn_a_c = Connection(
-        source=node_id_a, sink=node_id_c, edge=create_socket_connection(5)
-    )
-    conn_b_a = Connection(
-        source=node_id_b, sink=node_id_a, edge=create_socket_connection(6)
-    )
-
-    profiles = {
-        node_id_a: create_node_profile(available_memory[0]),
-        node_id_b: create_node_profile(available_memory[1]),
-        node_id_c: create_node_profile(available_memory[2]),
-    }
-    topology.add_node(node_id_a)
-    topology.add_node(node_id_b)
-    topology.add_node(node_id_c)
-    topology.add_connection(conn_a_b)
-    topology.add_connection(conn_b_c)
-    topology.add_connection(conn_c_a)
-    topology.add_connection(conn_c_b)
-    topology.add_connection(conn_a_c)
-    topology.add_connection(conn_b_a)
+    topology.add_node(create_node(available_memory[0], node_id_a))
+    topology.add_node(create_node(available_memory[1], node_id_b))
+    topology.add_node(create_node(available_memory[2], node_id_c))
+    # Add bidirectional connections for ring topology
+    topology.add_connection(create_connection(node_id_a, node_id_b))
+    topology.add_connection(create_connection(node_id_b, node_id_a))
+    topology.add_connection(create_connection(node_id_b, node_id_c))
+    topology.add_connection(create_connection(node_id_c, node_id_b))
+    topology.add_connection(create_connection(node_id_c, node_id_a))
+    topology.add_connection(create_connection(node_id_a, node_id_c))
 
     # act
-    placements = place_instance(cic, topology, {}, profiles)
+    placements = place_instance(cic, topology, {})
 
     # assert
     assert len(placements) == 1
@@ -151,11 +130,12 @@ def test_get_instance_placements_create_instance(
     assert shards_sorted[-1].end_layer == total_layers
 
 
-def test_get_instance_placements_one_node_exact_fit() -> None:
+def test_get_instance_placements_one_node_exact_fit(
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+) -> None:
     topology = Topology()
     node_id = NodeId()
-    topology.add_node(node_id)
-    profiles = {node_id: create_node_profile(1000 * 1024)}
+    topology.add_node(create_node(1000 * 1024, node_id))
     cic = place_instance_command(
         ModelMetadata(
             model_id=ModelId("test-model"),
@@ -166,7 +146,7 @@ def test_get_instance_placements_one_node_exact_fit() -> None:
             supports_tensor=True,
         ),
     )
-    placements = place_instance(cic, topology, {}, profiles)
+    placements = place_instance(cic, topology, {})
 
     assert len(placements) == 1
     instance_id = list(placements.keys())[0]
@@ -177,11 +157,12 @@ def test_get_instance_placements_one_node_exact_fit() -> None:
     assert len(instance.shard_assignments.runner_to_shard) == 1
 
 
-def test_get_instance_placements_one_node_fits_with_extra_memory() -> None:
+def test_get_instance_placements_one_node_fits_with_extra_memory(
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+) -> None:
     topology = Topology()
     node_id = NodeId()
-    topology.add_node(node_id)
-    profiles = {node_id: create_node_profile(1001 * 1024)}
+    topology.add_node(create_node(1001 * 1024, node_id))
     cic = place_instance_command(
         ModelMetadata(
             model_id=ModelId("test-model"),
@@ -192,7 +173,7 @@ def test_get_instance_placements_one_node_fits_with_extra_memory() -> None:
             supports_tensor=True,
         ),
     )
-    placements = place_instance(cic, topology, {}, profiles)
+    placements = place_instance(cic, topology, {})
 
     assert len(placements) == 1
     instance_id = list(placements.keys())[0]
@@ -203,11 +184,12 @@ def test_get_instance_placements_one_node_fits_with_extra_memory() -> None:
     assert len(instance.shard_assignments.runner_to_shard) == 1
 
 
-def test_get_instance_placements_one_node_not_fit() -> None:
+def test_get_instance_placements_one_node_not_fit(
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+) -> None:
     topology = Topology()
     node_id = NodeId()
-    topology.add_node(node_id)
-    profiles = {node_id: create_node_profile(1000 * 1024)}
+    topology.add_node(create_node(1000 * 1024, node_id))
     cic = place_instance_command(
         model_meta=ModelMetadata(
             model_id=ModelId("test-model"),
@@ -220,7 +202,7 @@ def test_get_instance_placements_one_node_not_fit() -> None:
     )
 
     with pytest.raises(ValueError, match="No cycles found with sufficient memory"):
-        place_instance(cic, topology, {}, profiles)
+        place_instance(cic, topology, {})
 
 
 def test_get_transition_events_no_change(instance: Instance):
@@ -265,130 +247,179 @@ def test_get_transition_events_delete_instance(instance: Instance):
     assert events[0].instance_id == instance_id
 
 
-def test_placement_selects_leaf_nodes(
+def test_placement_selects_cycle_with_most_memory(
+    topology: Topology,
     model_meta: ModelMetadata,
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+    create_connection: Callable[[NodeId, NodeId], Connection],
 ):
-    # arrange
-    topology = Topology()
+    # Arrange two 3-node cycles with different total memory.
+    # With bidirectional connections for ring topology, both cycles have non-leaf nodes.
+    # The algorithm should select the cycle with the most available memory.
 
-    model_meta.storage_size = Memory.from_bytes(1000)
+    # Model requires more than any single node but fits within a 3-node cycle
+    model_meta.storage_size.in_bytes = 1500
+    model_meta.n_layers = 12
 
+    # Create node ids
     node_id_a = NodeId()
     node_id_b = NodeId()
     node_id_c = NodeId()
     node_id_d = NodeId()
+    node_id_e = NodeId()
+    node_id_f = NodeId()
 
-    profiles = {
-        node_id_a: create_node_profile(500),
-        node_id_b: create_node_profile(600),
-        node_id_c: create_node_profile(600),
-        node_id_d: create_node_profile(500),
-    }
+    # A-B-C cycle total memory = 1600 (< D-E-F total)
+    topology.add_node(create_node(400, node_id_a))
+    topology.add_node(create_node(400, node_id_b))
+    topology.add_node(create_node(800, node_id_c))
 
-    topology.add_node(node_id_a)
-    topology.add_node(node_id_b)
-    topology.add_node(node_id_c)
-    topology.add_node(node_id_d)
+    # D-E-F cycle total memory = 1800 (> A-B-C total)
+    topology.add_node(create_node(600, node_id_d))
+    topology.add_node(create_node(600, node_id_e))
+    topology.add_node(create_node(600, node_id_f))
 
-    # Daisy chain topology (directed)
-    topology.add_connection(
-        Connection(source=node_id_a, sink=node_id_b, edge=create_socket_connection(1))
-    )
-    topology.add_connection(
-        Connection(source=node_id_b, sink=node_id_a, edge=create_socket_connection(1))
-    )
-    topology.add_connection(
-        Connection(source=node_id_b, sink=node_id_c, edge=create_socket_connection(1))
-    )
-    topology.add_connection(
-        Connection(source=node_id_c, sink=node_id_b, edge=create_socket_connection(1))
-    )
-    topology.add_connection(
-        Connection(source=node_id_c, sink=node_id_d, edge=create_socket_connection(1))
-    )
-    topology.add_connection(
-        Connection(source=node_id_d, sink=node_id_c, edge=create_socket_connection(1))
+    # Build bidirectional cycles for ring topology
+    topology.add_connection(create_connection(node_id_a, node_id_b))
+    topology.add_connection(create_connection(node_id_b, node_id_a))
+    topology.add_connection(create_connection(node_id_b, node_id_c))
+    topology.add_connection(create_connection(node_id_c, node_id_b))
+    topology.add_connection(create_connection(node_id_c, node_id_a))
+    topology.add_connection(create_connection(node_id_a, node_id_c))
+
+    topology.add_connection(create_connection(node_id_d, node_id_e))
+    topology.add_connection(create_connection(node_id_e, node_id_d))
+    topology.add_connection(create_connection(node_id_e, node_id_f))
+    topology.add_connection(create_connection(node_id_f, node_id_e))
+    topology.add_connection(create_connection(node_id_f, node_id_d))
+    topology.add_connection(create_connection(node_id_d, node_id_f))
+
+    cic = place_instance_command(
+        model_meta=model_meta,
     )
 
-    cic = place_instance_command(model_meta=model_meta)
+    # Act
+    placements = place_instance(cic, topology, {})
 
-    # act
-    placements = place_instance(cic, topology, {}, profiles)
-
-    # assert
+    # Assert: D-E-F cycle should be selected as it has more total memory
     assert len(placements) == 1
-    instance = list(placements.values())[0]
+    instance_id = list(placements.keys())[0]
+    instance = placements[instance_id]
 
     assigned_nodes = set(instance.shard_assignments.node_to_runner.keys())
-    assert assigned_nodes == set((node_id_a, node_id_b)) or assigned_nodes == set(
-        (
-            node_id_c,
-            node_id_d,
-        )
-    )
+    less_memory_cycle_nodes = {node_id_a, node_id_b, node_id_c}
+    more_memory_cycle_nodes = {node_id_d, node_id_e, node_id_f}
+
+    assert more_memory_cycle_nodes.issubset(assigned_nodes)
+    assert assigned_nodes.isdisjoint(less_memory_cycle_nodes)
 
 
 def test_tensor_rdma_backend_connectivity_matrix(
+    topology: Topology,
     model_meta: ModelMetadata,
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+    create_connection: Callable[[NodeId, NodeId], Connection],
 ):
-    # arrange
-    topology = Topology()
     model_meta.n_layers = 12
     model_meta.storage_size.in_bytes = 1500
 
-    node_a = NodeId()
-    node_b = NodeId()
-    node_c = NodeId()
+    node_id_a = NodeId()
+    node_id_b = NodeId()
+    node_id_c = NodeId()
 
-    profiles = {
-        node_a: create_node_profile(500),
-        node_b: create_node_profile(500),
-        node_c: create_node_profile(500),
-    }
+    node_a = create_node(500, node_id_a)
+    node_b = create_node(500, node_id_b)
+    node_c = create_node(500, node_id_c)
 
     ethernet_interface = NetworkInterfaceInfo(
         name="en0",
-        ip_address="10.0.0.1",
-    )
-    ethernet_conn = SocketConnection(
-        sink_multiaddr=Multiaddr(address="/ip4/10.0.0.1/tcp/8000")
+        ip_address="192.168.1.100",
     )
 
-    profiles[node_a].network_interfaces = [ethernet_interface]
-    profiles[node_b].network_interfaces = [ethernet_interface]
-    profiles[node_c].network_interfaces = [ethernet_interface]
+    assert node_a.node_profile is not None
+    assert node_b.node_profile is not None
+    assert node_c.node_profile is not None
+
+    conn_a_b = create_connection(node_id_a, node_id_b)
+    conn_b_c = create_connection(node_id_b, node_id_c)
+    conn_c_a = create_connection(node_id_c, node_id_a)
+
+    conn_b_a = create_connection(node_id_b, node_id_a)
+    conn_c_b = create_connection(node_id_c, node_id_b)
+    conn_a_c = create_connection(node_id_a, node_id_c)
+
+    assert conn_a_b.send_back_multiaddr is not None
+    assert conn_b_c.send_back_multiaddr is not None
+    assert conn_c_a.send_back_multiaddr is not None
+
+    assert conn_b_a.send_back_multiaddr is not None
+    assert conn_c_b.send_back_multiaddr is not None
+    assert conn_a_c.send_back_multiaddr is not None
+
+    node_a.node_profile = NodePerformanceProfile(
+        model_id="test",
+        chip_id="test",
+        friendly_name="test",
+        memory=node_a.node_profile.memory,
+        network_interfaces=[
+            NetworkInterfaceInfo(
+                name="en3",
+                ip_address=conn_c_a.send_back_multiaddr.ip_address,
+            ),
+            NetworkInterfaceInfo(
+                name="en4",
+                ip_address=conn_b_a.send_back_multiaddr.ip_address,
+            ),
+            ethernet_interface,
+        ],
+        system=node_a.node_profile.system,
+    )
+    node_b.node_profile = NodePerformanceProfile(
+        model_id="test",
+        chip_id="test",
+        friendly_name="test",
+        memory=node_b.node_profile.memory,
+        network_interfaces=[
+            NetworkInterfaceInfo(
+                name="en3",
+                ip_address=conn_c_b.send_back_multiaddr.ip_address,
+            ),
+            NetworkInterfaceInfo(
+                name="en4",
+                ip_address=conn_a_b.send_back_multiaddr.ip_address,
+            ),
+            ethernet_interface,
+        ],
+        system=node_b.node_profile.system,
+    )
+    node_c.node_profile = NodePerformanceProfile(
+        model_id="test",
+        chip_id="test",
+        friendly_name="test",
+        memory=node_c.node_profile.memory,
+        network_interfaces=[
+            NetworkInterfaceInfo(
+                name="en3",
+                ip_address=conn_a_c.send_back_multiaddr.ip_address,
+            ),
+            NetworkInterfaceInfo(
+                name="en4",
+                ip_address=conn_b_c.send_back_multiaddr.ip_address,
+            ),
+            ethernet_interface,
+        ],
+        system=node_c.node_profile.system,
+    )
 
     topology.add_node(node_a)
     topology.add_node(node_b)
     topology.add_node(node_c)
-
-    # RDMA connections (directed)
-    topology.add_connection(
-        Connection(source=node_a, sink=node_b, edge=create_rdma_connection(3))
-    )
-    topology.add_connection(
-        Connection(source=node_b, sink=node_a, edge=create_rdma_connection(3))
-    )
-    topology.add_connection(
-        Connection(source=node_b, sink=node_c, edge=create_rdma_connection(4))
-    )
-    topology.add_connection(
-        Connection(source=node_c, sink=node_b, edge=create_rdma_connection(4))
-    )
-    topology.add_connection(
-        Connection(source=node_a, sink=node_c, edge=create_rdma_connection(5))
-    )
-    topology.add_connection(
-        Connection(source=node_c, sink=node_a, edge=create_rdma_connection(5))
-    )
-
-    # Ethernet connections (directed)
-    topology.add_connection(Connection(source=node_a, sink=node_b, edge=ethernet_conn))
-    topology.add_connection(Connection(source=node_b, sink=node_c, edge=ethernet_conn))
-    topology.add_connection(Connection(source=node_c, sink=node_a, edge=ethernet_conn))
-    topology.add_connection(Connection(source=node_a, sink=node_c, edge=ethernet_conn))
-    topology.add_connection(Connection(source=node_b, sink=node_a, edge=ethernet_conn))
-    topology.add_connection(Connection(source=node_c, sink=node_b, edge=ethernet_conn))
+    topology.add_connection(conn_a_b)
+    topology.add_connection(conn_b_c)
+    topology.add_connection(conn_c_a)
+    topology.add_connection(conn_b_a)
+    topology.add_connection(conn_c_b)
+    topology.add_connection(conn_a_c)
 
     cic = PlaceInstance(
         sharding=Sharding.Tensor,
@@ -398,34 +429,35 @@ def test_tensor_rdma_backend_connectivity_matrix(
         min_nodes=1,
     )
 
-    # act
-    placements = place_instance(cic, topology, {}, profiles)
+    placements = place_instance(cic, topology, {})
 
-    # assert
     assert len(placements) == 1
     instance_id = list(placements.keys())[0]
     instance = placements[instance_id]
 
     assert isinstance(instance, MlxJacclInstance)
 
-    assert instance.jaccl_devices is not None
+    assert instance.ibv_devices is not None
     assert instance.jaccl_coordinators is not None
 
-    matrix = instance.jaccl_devices
+    matrix = instance.ibv_devices
     assert len(matrix) == 3
+
     for i in range(3):
         assert matrix[i][i] is None
 
     assigned_nodes = list(instance.shard_assignments.node_to_runner.keys())
     node_to_idx = {node_id: idx for idx, node_id in enumerate(assigned_nodes)}
 
-    idx_a = node_to_idx[node_a]
-    idx_b = node_to_idx[node_b]
-    idx_c = node_to_idx[node_c]
+    idx_a = node_to_idx[node_id_a]
+    idx_b = node_to_idx[node_id_b]
+    idx_c = node_to_idx[node_id_c]
 
-    assert matrix[idx_a][idx_b] == "rdma_en3"
-    assert matrix[idx_b][idx_c] == "rdma_en4"
-    assert matrix[idx_c][idx_a] == "rdma_en5"
+    logger.info(matrix)
+
+    assert matrix[idx_a][idx_b] == "rdma_en4"
+    assert matrix[idx_b][idx_c] == "rdma_en3"
+    assert matrix[idx_c][idx_a] == "rdma_en3"
 
     # Verify coordinators are set for all nodes
     assert len(instance.jaccl_coordinators) == 3
@@ -437,5 +469,7 @@ def test_tensor_rdma_backend_connectivity_matrix(
         if node_id == assigned_nodes[0]:
             assert coordinator.startswith("0.0.0.0:")
         else:
+            # Non-rank-0 nodes should have valid IP addresses (can be link-local)
             ip_part = coordinator.split(":")[0]
+            # Just verify it's a valid IP format
             assert len(ip_part.split(".")) == 4

src/exo/master/tests/test_placement_utils.py

@@ -1,4 +1,4 @@
-from copy import copy
+from typing import Callable
 
 import pytest
 
@@ -9,178 +9,154 @@ from exo.master.placement_utils import (
     get_shard_assignments,
     get_smallest_cycles,
 )
-from exo.master.tests.conftest import create_node_profile, create_socket_connection
 from exo.shared.topology import Topology
 from exo.shared.types.common import Host, NodeId
 from exo.shared.types.memory import Memory
 from exo.shared.types.models import ModelId, ModelMetadata
-from exo.shared.types.profiling import (
-    MemoryUsage,
-    NetworkInterfaceInfo,
-    NodePerformanceProfile,
-    SystemPerformanceProfile,
-)
-from exo.shared.types.topology import Connection, SocketConnection
+from exo.shared.types.profiling import NetworkInterfaceInfo, NodePerformanceProfile
+from exo.shared.types.topology import Connection, NodeInfo
 from exo.shared.types.worker.shards import Sharding
 
 
-def test_filter_cycles_by_memory():
+@pytest.fixture
+def topology() -> Topology:
+    topology = Topology()
+    return topology
+
+
+def test_filter_cycles_by_memory(
+    topology: Topology,
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+    create_connection: Callable[[NodeId, NodeId], Connection],
+):
     # arrange
     node1_id = NodeId()
     node2_id = NodeId()
-    connection1 = Connection(
-        source=node1_id, sink=node2_id, edge=create_socket_connection(1)
-    )
-    connection2 = Connection(
-        source=node2_id, sink=node1_id, edge=create_socket_connection(2)
-    )
 
-    node1 = create_node_profile(1000 * 1024)
-    node2 = create_node_profile(1000 * 1024)
-    node_profiles = {node1_id: node1, node2_id: node2}
+    node1 = create_node(1000 * 1024, node1_id)
+    node2 = create_node(1000 * 1024, node2_id)
+
+    topology.add_node(node1)
+    topology.add_node(node2)
+
+    connection1 = create_connection(node1_id, node2_id)
+    connection2 = create_connection(node2_id, node1_id)
 
-    topology = Topology()
-    topology.add_node(node1_id)
-    topology.add_node(node2_id)
     topology.add_connection(connection1)
     topology.add_connection(connection2)
 
-    cycles = [c for c in topology.get_cycles() if len(c) != 1]
+    cycles = topology.get_cycles()
     assert len(cycles) == 1
     assert len(cycles[0]) == 2
 
     # act
-    filtered_cycles = filter_cycles_by_memory(
-        cycles, node_profiles, Memory.from_bytes(1)
-    )
+    filtered_cycles = filter_cycles_by_memory(cycles, Memory.from_bytes(1))
 
     # assert
     assert len(filtered_cycles) == 1
     assert len(filtered_cycles[0]) == 2
-    assert set(n for n in filtered_cycles[0]) == {node1_id, node2_id}
+    assert set(n.node_id for n in filtered_cycles[0]) == {node1_id, node2_id}
 
 
-def test_filter_cycles_by_insufficient_memory():
+def test_filter_cycles_by_insufficient_memory(
+    topology: Topology,
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+    create_connection: Callable[[NodeId, NodeId], Connection],
+):
     # arrange
     node1_id = NodeId()
     node2_id = NodeId()
-    connection1 = Connection(
-        source=node1_id, sink=node2_id, edge=create_socket_connection(1)
-    )
-    connection2 = Connection(
-        source=node2_id, sink=node1_id, edge=create_socket_connection(2)
-    )
 
-    node1 = create_node_profile(1000 * 1024)
-    node2 = create_node_profile(1000 * 1024)
-    node_profiles = {node1_id: node1, node2_id: node2}
+    node1 = create_node(1000 * 1024, node1_id)
+    node2 = create_node(1000 * 1024, node2_id)
+
+    topology.add_node(node1)
+    topology.add_node(node2)
+
+    connection1 = create_connection(node1_id, node2_id)
+    connection2 = create_connection(node2_id, node1_id)
 
-    topology = Topology()
-    topology.add_node(node1_id)
-    topology.add_node(node2_id)
     topology.add_connection(connection1)
     topology.add_connection(connection2)
 
     # act
     filtered_cycles = filter_cycles_by_memory(
-        topology.get_cycles(), node_profiles, Memory.from_kb(2001)
+        topology.get_cycles(), Memory.from_kb(2001)
     )
 
     # assert
     assert len(filtered_cycles) == 0
 
 
-def test_filter_multiple_cycles_by_memory():
+def test_filter_multiple_cycles_by_memory(
+    topology: Topology,
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+    create_connection: Callable[[NodeId, NodeId], Connection],
+):
     # arrange
     node_a_id = NodeId()
     node_b_id = NodeId()
     node_c_id = NodeId()
-    connection1 = Connection(
-        source=node_a_id, sink=node_b_id, edge=create_socket_connection(1)
-    )
-    connection2 = Connection(
-        source=node_b_id, sink=node_a_id, edge=create_socket_connection(2)
-    )
-    connection3 = Connection(
-        source=node_a_id, sink=node_c_id, edge=create_socket_connection(3)
-    )
-    connection4 = Connection(
-        source=node_c_id, sink=node_b_id, edge=create_socket_connection(4)
-    )
 
-    node_a = create_node_profile(500 * 1024)
-    node_b = create_node_profile(500 * 1024)
-    node_c = create_node_profile(1000 * 1024)
-    node_profiles = {
-        node_a_id: node_a,
-        node_b_id: node_b,
-        node_c_id: node_c,
-    }
+    node_a = create_node(500 * 1024, node_a_id)
+    node_b = create_node(500 * 1024, node_b_id)
+    node_c = create_node(1000 * 1024, node_c_id)
 
-    topology = Topology()
-    topology.add_node(node_a_id)
-    topology.add_node(node_b_id)
-    topology.add_node(node_c_id)
-    topology.add_connection(connection1)
-    topology.add_connection(connection2)
-    topology.add_connection(connection3)
-    topology.add_connection(connection4)
+    topology.add_node(node_a)
+    topology.add_node(node_b)
+    topology.add_node(node_c)
+
+    topology.add_connection(create_connection(node_a_id, node_b_id))
+    topology.add_connection(create_connection(node_b_id, node_a_id))
+
+    topology.add_connection(create_connection(node_a_id, node_c_id))
+    topology.add_connection(create_connection(node_c_id, node_b_id))
 
     cycles = topology.get_cycles()
 
     # act
-    filtered_cycles = filter_cycles_by_memory(
-        cycles, node_profiles, Memory.from_kb(1500)
-    )
+    filtered_cycles = filter_cycles_by_memory(cycles, Memory.from_kb(1500))
 
     # assert
     assert len(filtered_cycles) == 1
     assert len(filtered_cycles[0]) == 3
-    assert set(n for n in filtered_cycles[0]) == {
+    assert set(n.node_id for n in filtered_cycles[0]) == {
         node_a_id,
         node_b_id,
         node_c_id,
     }
 
 
-def test_get_smallest_cycles():
+def test_get_smallest_cycles(
+    topology: Topology,
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+    create_connection: Callable[[NodeId, NodeId], Connection],
+):
     # arrange
     node_a_id = NodeId()
     node_b_id = NodeId()
     node_c_id = NodeId()
 
-    topology = Topology()
-    topology.add_node(node_a_id)
-    topology.add_node(node_b_id)
-    topology.add_node(node_c_id)
+    node_a = create_node(500 * 1024, node_a_id)
+    node_b = create_node(500 * 1024, node_b_id)
+    node_c = create_node(1000 * 1024, node_c_id)
 
-    connection1 = Connection(
-        source=node_a_id, sink=node_b_id, edge=create_socket_connection(1)
-    )
-    connection2 = Connection(
-        source=node_b_id, sink=node_a_id, edge=create_socket_connection(2)
-    )
-    connection3 = Connection(
-        source=node_a_id, sink=node_c_id, edge=create_socket_connection(3)
-    )
-    connection4 = Connection(
-        source=node_c_id, sink=node_b_id, edge=create_socket_connection(4)
-    )
+    topology.add_node(node_a)
+    topology.add_node(node_b)
+    topology.add_node(node_c)
 
-    topology.add_connection(connection1)
-    topology.add_connection(connection2)
-    topology.add_connection(connection3)
-    topology.add_connection(connection4)
-
-    cycles = [c for c in topology.get_cycles() if len(c) != 1]  # ignore singletons
+    topology.add_connection(create_connection(node_a_id, node_b_id))
+    topology.add_connection(create_connection(node_b_id, node_c_id))
+    topology.add_connection(create_connection(node_c_id, node_a_id))
+    topology.add_connection(create_connection(node_b_id, node_a_id))
 
     # act
-    smallest_cycles = get_smallest_cycles(cycles)
+    smallest_cycles = get_smallest_cycles(topology.get_cycles())
 
     # assert
     assert len(smallest_cycles) == 1
     assert len(smallest_cycles[0]) == 2
-    assert set(n for n in smallest_cycles[0]) == {node_a_id, node_b_id}
+    assert set(n.node_id for n in smallest_cycles[0]) == {node_a_id, node_b_id}
 
 
 @pytest.mark.parametrize(
@@ -192,6 +168,9 @@ def test_get_smallest_cycles():
     ],
 )
 def test_get_shard_assignments(
+    topology: Topology,
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+    create_connection: Callable[[NodeId, NodeId], Connection],
     available_memory: tuple[int, int, int],
     total_layers: int,
     expected_layers: tuple[int, int, int],
@@ -201,37 +180,18 @@ def test_get_shard_assignments(
     node_b_id = NodeId()
     node_c_id = NodeId()
 
-    # create connections (A -> B -> C -> A forms a 3-cycle, plus B -> A also exists)
-    connection1 = Connection(
-        source=node_a_id, sink=node_b_id, edge=create_socket_connection(1)
-    )
-    connection2 = Connection(
-        source=node_b_id, sink=node_c_id, edge=create_socket_connection(2)
-    )
-    connection3 = Connection(
-        source=node_c_id, sink=node_a_id, edge=create_socket_connection(3)
-    )
-    connection4 = Connection(
-        source=node_b_id, sink=node_a_id, edge=create_socket_connection(4)
-    )
+    node_a = create_node(available_memory[0] * 1024, node_a_id)
+    node_b = create_node(available_memory[1] * 1024, node_b_id)
+    node_c = create_node(available_memory[2] * 1024, node_c_id)
 
-    topology = Topology()
-    topology.add_node(node_a_id)
-    topology.add_node(node_b_id)
-    topology.add_node(node_c_id)
-    topology.add_connection(connection1)
-    topology.add_connection(connection2)
-    topology.add_connection(connection3)
-    topology.add_connection(connection4)
+    topology.add_node(node_a)
+    topology.add_node(node_b)
+    topology.add_node(node_c)
 
-    node_a = create_node_profile(available_memory[0] * 1024)
-    node_b = create_node_profile(available_memory[1] * 1024)
-    node_c = create_node_profile(available_memory[2] * 1024)
-    node_profiles = {
-        node_a_id: node_a,
-        node_b_id: node_b,
-        node_c_id: node_c,
-    }
+    topology.add_connection(create_connection(node_a_id, node_b_id))
+    topology.add_connection(create_connection(node_b_id, node_c_id))
+    topology.add_connection(create_connection(node_c_id, node_a_id))
+    topology.add_connection(create_connection(node_b_id, node_a_id))
 
     model_meta = ModelMetadata(
         model_id=ModelId("test-model"),
@@ -241,22 +201,23 @@ def test_get_shard_assignments(
         hidden_size=1000,
         supports_tensor=True,
     )
-
     cycles = topology.get_cycles()
-
-    # pick the 3-node cycle deterministically (cycle ordering can vary)
-    selected_cycle = next(cycle for cycle in cycles if len(cycle) == 3)
+    selected_cycle = cycles[0]
 
     # act
     shard_assignments = get_shard_assignments(
-        model_meta, selected_cycle, Sharding.Pipeline, node_profiles=node_profiles
+        model_meta, selected_cycle, Sharding.Pipeline
     )
 
     # assert
     runner_id_a = shard_assignments.node_to_runner[node_a_id]
     runner_id_b = shard_assignments.node_to_runner[node_b_id]
     runner_id_c = shard_assignments.node_to_runner[node_c_id]
-
+    assert (
+        shard_assignments.runner_to_shard[runner_id_c].end_layer
+        - shard_assignments.runner_to_shard[runner_id_c].start_layer
+        == expected_layers[2]
+    )
     assert (
         shard_assignments.runner_to_shard[runner_id_a].end_layer
         - shard_assignments.runner_to_shard[runner_id_a].start_layer
@@ -267,37 +228,30 @@ def test_get_shard_assignments(
         - shard_assignments.runner_to_shard[runner_id_b].start_layer
         == expected_layers[1]
     )
-    assert (
-        shard_assignments.runner_to_shard[runner_id_c].end_layer
-        - shard_assignments.runner_to_shard[runner_id_c].start_layer
-        == expected_layers[2]
-    )
 
 
-def test_get_hosts_from_subgraph():
+def test_get_hosts_from_subgraph(
+    topology: Topology,
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+    create_connection: Callable[[NodeId, NodeId, int | None], Connection],
+):
     # arrange
     node_a_id = NodeId()
     node_b_id = NodeId()
     node_c_id = NodeId()
-    topology = Topology()
 
-    topology.add_node(node_a_id)
-    topology.add_node(node_b_id)
-    topology.add_node(node_c_id)
+    node_a = create_node(500, node_a_id)
+    node_b = create_node(500, node_b_id)
+    node_c = create_node(1000, node_c_id)
 
-    connection1 = Connection(
-        source=node_a_id, sink=node_b_id, edge=create_socket_connection(1)
-    )
-    connection2 = Connection(
-        source=node_b_id, sink=node_c_id, edge=create_socket_connection(2)
-    )
-    connection3 = Connection(
-        source=node_c_id, sink=node_a_id, edge=create_socket_connection(3)
-    )
+    topology.add_node(node_a)
+    topology.add_node(node_b)
+    topology.add_node(node_c)
 
-    topology.add_connection(connection1)
-    topology.add_connection(connection2)
-    topology.add_connection(connection3)
+    topology.add_connection(create_connection(node_a_id, node_b_id, 5001))
+    topology.add_connection(create_connection(node_b_id, node_c_id, 5002))
+    topology.add_connection(create_connection(node_c_id, node_a_id, 5003))
+    topology.add_connection(create_connection(node_b_id, node_a_id, 5004))
 
     # act
     hosts = get_hosts_from_subgraph(topology)
@@ -305,78 +259,95 @@ def test_get_hosts_from_subgraph():
     # assert
     assert len(hosts) == 3
     expected_hosts = [
-        Host(ip="169.254.0.1", port=1234),
-        Host(ip="169.254.0.2", port=1234),
-        Host(ip="169.254.0.3", port=1234),
+        Host(ip=("169.254.0.2"), port=5001),
+        Host(ip=("169.254.0.3"), port=5002),
+        Host(ip=("169.254.0.4"), port=5003),
     ]
     for expected_host in expected_hosts:
         assert expected_host in hosts
 
 
-def test_get_mlx_jaccl_coordinators():
+def test_get_mlx_jaccl_coordinators(
+    topology: Topology,
+    create_node: Callable[[int, NodeId | None], NodeInfo],
+    create_connection: Callable[[NodeId, NodeId, int | None], Connection],
+):
     # arrange
     node_a_id = NodeId()
     node_b_id = NodeId()
     node_c_id = NodeId()
 
-    # fully connected (directed) between the 3 nodes
-    conn_a_b = Connection(
-        source=node_a_id, sink=node_b_id, edge=create_socket_connection(1)
-    )
-    conn_b_a = Connection(
-        source=node_b_id, sink=node_a_id, edge=create_socket_connection(2)
-    )
-    conn_b_c = Connection(
-        source=node_b_id, sink=node_c_id, edge=create_socket_connection(3)
-    )
-    conn_c_b = Connection(
-        source=node_c_id, sink=node_b_id, edge=create_socket_connection(4)
-    )
-    conn_c_a = Connection(
-        source=node_c_id, sink=node_a_id, edge=create_socket_connection(5)
-    )
-    conn_a_c = Connection(
-        source=node_a_id, sink=node_c_id, edge=create_socket_connection(6)
-    )
+    node_a = create_node(500 * 1024, node_a_id)
+    node_b = create_node(500 * 1024, node_b_id)
+    node_c = create_node(1000 * 1024, node_c_id)
 
-    npp = NodePerformanceProfile(
+    conn_a_b = create_connection(node_a_id, node_b_id, 5001)
+    conn_b_a = create_connection(node_b_id, node_a_id, 5002)
+    conn_b_c = create_connection(node_b_id, node_c_id, 5003)
+    conn_c_b = create_connection(node_c_id, node_b_id, 5004)
+    conn_c_a = create_connection(node_c_id, node_a_id, 5005)
+    conn_a_c = create_connection(node_a_id, node_c_id, 5006)
+
+    # Update node profiles with network interfaces before adding to topology
+    assert node_a.node_profile is not None
+    assert node_b.node_profile is not None
+    assert node_c.node_profile is not None
+
+    node_a.node_profile = NodePerformanceProfile(
         model_id="test",
         chip_id="test",
         friendly_name="test",
-        memory=MemoryUsage.from_bytes(
-            ram_total=0,
-            ram_available=0,
-            swap_total=0,
-            swap_available=0,
-        ),
-        network_interfaces=[],
-        system=SystemPerformanceProfile(),
+        memory=node_a.node_profile.memory,
+        network_interfaces=[
+            NetworkInterfaceInfo(
+                name="en3",
+                ip_address=conn_a_b.send_back_multiaddr.ip_address,
+            ),
+            NetworkInterfaceInfo(
+                name="en4",
+                ip_address=conn_a_c.send_back_multiaddr.ip_address,
+            ),
+        ],
+        system=node_a.node_profile.system,
+    )
+    node_b.node_profile = NodePerformanceProfile(
+        model_id="test",
+        chip_id="test",
+        friendly_name="test",
+        memory=node_b.node_profile.memory,
+        network_interfaces=[
+            NetworkInterfaceInfo(
+                name="en3",
+                ip_address=conn_b_a.send_back_multiaddr.ip_address,
+            ),
+            NetworkInterfaceInfo(
+                name="en4",
+                ip_address=conn_b_c.send_back_multiaddr.ip_address,
+            ),
+        ],
+        system=node_b.node_profile.system,
+    )
+    node_c.node_profile = NodePerformanceProfile(
+        model_id="test",
+        chip_id="test",
+        friendly_name="test",
+        memory=node_c.node_profile.memory,
+        network_interfaces=[
+            NetworkInterfaceInfo(
+                name="en3",
+                ip_address=conn_c_b.send_back_multiaddr.ip_address,
+            ),
+            NetworkInterfaceInfo(
+                name="en4",
+                ip_address=conn_c_a.send_back_multiaddr.ip_address,
+            ),
+        ],
+        system=node_c.node_profile.system,
     )
-    npp_a = copy(npp)
-    npp_a.network_interfaces = [
-        NetworkInterfaceInfo(name="en0", ip_address="169.254.0.5"),
-        NetworkInterfaceInfo(name="en0", ip_address="169.254.0.2"),
-    ]
-    npp_b = copy(npp)
-    npp_b.network_interfaces = [
-        NetworkInterfaceInfo(name="en0", ip_address="169.254.0.1"),
-        NetworkInterfaceInfo(name="en0", ip_address="169.254.0.4"),
-    ]
-    npp_c = copy(npp)
-    npp_c.network_interfaces = [
-        NetworkInterfaceInfo(name="en0", ip_address="169.254.0.3"),
-        NetworkInterfaceInfo(name="en0", ip_address="169.254.0.6"),
-    ]
-    node_profiles = {
-        node_a_id: npp_a,
-        node_b_id: npp_b,
-        node_c_id: npp_c,
-    }
 
-    topology = Topology()
-    topology.add_node(node_a_id)
-    topology.add_node(node_b_id)
-    topology.add_node(node_c_id)
+    topology.add_node(node_a)
+    topology.add_node(node_b)
+    topology.add_node(node_c)
 
     topology.add_connection(conn_a_b)
     topology.add_connection(conn_b_a)
@@ -385,12 +356,11 @@ def test_get_mlx_jaccl_coordinators():
     topology.add_connection(conn_c_a)
     topology.add_connection(conn_a_c)
 
+    cycle = [node_a, node_b, node_c]
+
     # act
     coordinators = get_mlx_jaccl_coordinators(
-        node_a_id,
-        coordinator_port=5000,
-        cycle_digraph=topology,
-        node_profiles=node_profiles,
+        cycle, coordinator_port=5000, cycle_digraph=topology
     )
 
     # assert
@@ -411,20 +381,19 @@ def test_get_mlx_jaccl_coordinators():
             f"Coordinator for {node_id} should use port 5000"
         )
 
-    # Rank 0 (node_a) treats this as the listen socket so should listen on all IPs
+    # Rank 0 (node_a) treats this as the listen socket so should listen on all
+    # IPs
     assert coordinators[node_a_id].startswith("0.0.0.0:"), (
-        "Rank 0 node should use 0.0.0.0 as coordinator listen address"
+        "Rank 0 node should use localhost as coordinator"
     )
 
     # Non-rank-0 nodes should use the specific IP from their connection to rank 0
     # node_b uses the IP from conn_b_a (node_b -> node_a)
-    assert isinstance(conn_b_a.edge, SocketConnection)
-    assert (
-        coordinators[node_b_id] == f"{conn_b_a.edge.sink_multiaddr.ip_address}:5000"
+    assert coordinators[node_b_id] == (
+        f"{conn_b_a.send_back_multiaddr.ip_address}:5000"
     ), "node_b should use the IP from conn_b_a"
 
     # node_c uses the IP from conn_c_a (node_c -> node_a)
-    assert isinstance(conn_c_a.edge, SocketConnection)
-    assert (
-        coordinators[node_c_id] == f"{conn_c_a.edge.sink_multiaddr.ip_address}:5000"
+    assert coordinators[node_c_id] == (
+        f"{conn_c_a.send_back_multiaddr.ip_address}:5000"
     ), "node_c should use the IP from conn_c_a"

src/exo/master/tests/test_topology.py

@@ -1,14 +1,13 @@
 import pytest
 
 from exo.shared.topology import Topology
-from exo.shared.types.common import NodeId
 from exo.shared.types.multiaddr import Multiaddr
 from exo.shared.types.profiling import (
-    MemoryUsage,
+    MemoryPerformanceProfile,
     NodePerformanceProfile,
     SystemPerformanceProfile,
 )
-from exo.shared.types.topology import Connection, SocketConnection
+from exo.shared.types.topology import Connection, ConnectionProfile, NodeId, NodeInfo
 
 
 @pytest.fixture
@@ -17,15 +16,20 @@ def topology() -> Topology:
 
 
 @pytest.fixture
-def socket_connection() -> SocketConnection:
-    return SocketConnection(
-        sink_multiaddr=Multiaddr(address="/ip4/127.0.0.1/tcp/1235"),
+def connection() -> Connection:
+    return Connection(
+        local_node_id=NodeId(),
+        send_back_node_id=NodeId(),
+        send_back_multiaddr=Multiaddr(address="/ip4/127.0.0.1/tcp/1235"),
+        connection_profile=ConnectionProfile(
+            throughput=1000, latency=1000, jitter=1000
+        ),
     )
 
 
 @pytest.fixture
 def node_profile() -> NodePerformanceProfile:
-    memory_profile = MemoryUsage.from_bytes(
+    memory_profile = MemoryPerformanceProfile.from_bytes(
         ram_total=1000, ram_available=1000, swap_total=1000, swap_available=1000
     )
     system_profile = SystemPerformanceProfile()
@@ -39,91 +43,162 @@ def node_profile() -> NodePerformanceProfile:
     )
 
 
-def test_add_node(topology: Topology):
+@pytest.fixture
+def connection_profile() -> ConnectionProfile:
+    return ConnectionProfile(throughput=1000, latency=1000, jitter=1000)
+
+
+def test_add_node(topology: Topology, node_profile: NodePerformanceProfile):
     # arrange
     node_id = NodeId()
 
     # act
-    topology.add_node(node_id)
+    topology.add_node(NodeInfo(node_id=node_id, node_profile=node_profile))
 
     # assert
-    assert topology.node_is_leaf(node_id)
+    data = topology.get_node_profile(node_id)
+    assert data == node_profile
 
 
-def test_add_connection(topology: Topology, socket_connection: SocketConnection):
+def test_add_connection(
+    topology: Topology, node_profile: NodePerformanceProfile, connection: Connection
+):
     # arrange
-    node_a = NodeId()
-    node_b = NodeId()
-    connection = Connection(source=node_a, sink=node_b, edge=socket_connection)
-
-    topology.add_node(node_a)
-    topology.add_node(node_b)
+    topology.add_node(
+        NodeInfo(node_id=connection.local_node_id, node_profile=node_profile)
+    )
+    topology.add_node(
+        NodeInfo(node_id=connection.send_back_node_id, node_profile=node_profile)
+    )
     topology.add_connection(connection)
 
     # act
-    data = list(topology.list_connections())
+    data = topology.get_connection_profile(connection)
 
     # assert
-    assert data == [connection]
+    assert data == connection.connection_profile
 
-    assert topology.node_is_leaf(node_a)
-    assert topology.node_is_leaf(node_b)
+
+def test_update_node_profile(
+    topology: Topology, node_profile: NodePerformanceProfile, connection: Connection
+):
+    # arrange
+    topology.add_node(
+        NodeInfo(node_id=connection.local_node_id, node_profile=node_profile)
+    )
+    topology.add_node(
+        NodeInfo(node_id=connection.send_back_node_id, node_profile=node_profile)
+    )
+    topology.add_connection(connection)
+
+    new_node_profile = NodePerformanceProfile(
+        model_id="test",
+        chip_id="test",
+        friendly_name="test",
+        memory=MemoryPerformanceProfile.from_bytes(
+            ram_total=1000, ram_available=1000, swap_total=1000, swap_available=1000
+        ),
+        network_interfaces=[],
+        system=SystemPerformanceProfile(),
+    )
+
+    # act
+    topology.update_node_profile(
+        connection.local_node_id, node_profile=new_node_profile
+    )
+
+    # assert
+    data = topology.get_node_profile(connection.local_node_id)
+    assert data == new_node_profile
+
+
+def test_update_connection_profile(
+    topology: Topology, node_profile: NodePerformanceProfile, connection: Connection
+):
+    # arrange
+    topology.add_node(
+        NodeInfo(node_id=connection.local_node_id, node_profile=node_profile)
+    )
+    topology.add_node(
+        NodeInfo(node_id=connection.send_back_node_id, node_profile=node_profile)
+    )
+    topology.add_connection(connection)
+
+    new_connection_profile = ConnectionProfile(
+        throughput=2000, latency=2000, jitter=2000
+    )
+    connection = Connection(
+        local_node_id=connection.local_node_id,
+        send_back_node_id=connection.send_back_node_id,
+        send_back_multiaddr=connection.send_back_multiaddr,
+        connection_profile=new_connection_profile,
+    )
+
+    # act
+    topology.update_connection_profile(connection)
+
+    # assert
+    data = topology.get_connection_profile(connection)
+    assert data == new_connection_profile
 
 
 def test_remove_connection_still_connected(
-    topology: Topology, socket_connection: SocketConnection
+    topology: Topology, node_profile: NodePerformanceProfile, connection: Connection
 ):
     # arrange
-    node_a = NodeId()
-    node_b = NodeId()
-    conn = Connection(source=node_a, sink=node_b, edge=socket_connection)
-
-    topology.add_node(node_a)
-    topology.add_node(node_b)
-    topology.add_connection(conn)
+    topology.add_node(
+        NodeInfo(node_id=connection.local_node_id, node_profile=node_profile)
+    )
+    topology.add_node(
+        NodeInfo(node_id=connection.send_back_node_id, node_profile=node_profile)
+    )
+    topology.add_connection(connection)
 
     # act
-    topology.remove_connection(conn)
+    topology.remove_connection(connection)
 
     # assert
-    assert list(topology.get_all_connections_between(node_a, node_b)) == []
+    assert topology.get_connection_profile(connection) is None
 
 
 def test_remove_node_still_connected(
-    topology: Topology, socket_connection: SocketConnection
+    topology: Topology, node_profile: NodePerformanceProfile, connection: Connection
 ):
     # arrange
-    node_a = NodeId()
-    node_b = NodeId()
-    conn = Connection(source=node_a, sink=node_b, edge=socket_connection)
-
-    topology.add_node(node_a)
-    topology.add_node(node_b)
-    topology.add_connection(conn)
-    assert list(topology.out_edges(node_a)) == [conn]
+    topology.add_node(
+        NodeInfo(node_id=connection.local_node_id, node_profile=node_profile)
+    )
+    topology.add_node(
+        NodeInfo(node_id=connection.send_back_node_id, node_profile=node_profile)
+    )
+    topology.add_connection(connection)
 
     # act
-    topology.remove_node(node_b)
+    topology.remove_node(connection.local_node_id)
 
     # assert
-    assert list(topology.out_edges(node_a)) == []
+    assert topology.get_node_profile(connection.local_node_id) is None
 
 
-def test_list_nodes(topology: Topology, socket_connection: SocketConnection):
+def test_list_nodes(
+    topology: Topology, node_profile: NodePerformanceProfile, connection: Connection
+):
     # arrange
-    node_a = NodeId()
-    node_b = NodeId()
-    conn = Connection(source=node_a, sink=node_b, edge=socket_connection)
-
-    topology.add_node(node_a)
-    topology.add_node(node_b)
-    topology.add_connection(conn)
-    assert list(topology.out_edges(node_a)) == [conn]
+    topology.add_node(
+        NodeInfo(node_id=connection.local_node_id, node_profile=node_profile)
+    )
+    topology.add_node(
+        NodeInfo(node_id=connection.send_back_node_id, node_profile=node_profile)
+    )
+    topology.add_connection(connection)
 
     # act
     nodes = list(topology.list_nodes())
 
     # assert
     assert len(nodes) == 2
-    assert all(isinstance(node, NodeId) for node in nodes)
-    assert set(node for node in nodes) == set([node_a, node_b])
+    assert all(isinstance(node, NodeInfo) for node in nodes)
+    assert {node.node_id for node in nodes} == {
+        connection.local_node_id,
+        connection.send_back_node_id,
+    }

src/exo/shared/apply.py

@@ -11,8 +11,12 @@ from exo.shared.types.events import (
     IndexedEvent,
     InstanceCreated,
     InstanceDeleted,
+    NodeCreated,
     NodeDownloadProgress,
-    NodeGatheredInfo,
+    NodeIdentityMeasured,
+    NodeMemoryMeasured,
+    NodeNetworkMeasured,
+    NodeSystemMeasured,
     NodeTimedOut,
     RunnerDeleted,
     RunnerStatusUpdated,
@@ -25,23 +29,19 @@ from exo.shared.types.events import (
     TopologyEdgeCreated,
     TopologyEdgeDeleted,
 )
-from exo.shared.types.profiling import NodePerformanceProfile
+from exo.shared.types.profiling import (
+    MemoryPerformanceProfile,
+    NetworkInterfaceInfo,
+    NodeIdentity,
+    NodePerformanceProfile,
+    SystemPerformanceProfile,
+)
 from exo.shared.types.state import State
 from exo.shared.types.tasks import Task, TaskId, TaskStatus
-from exo.shared.types.topology import Connection, RDMAConnection
+from exo.shared.types.topology import NodeInfo
 from exo.shared.types.worker.downloads import DownloadProgress
 from exo.shared.types.worker.instances import Instance, InstanceId
 from exo.shared.types.worker.runners import RunnerId, RunnerStatus
-from exo.utils.info_gatherer.info_gatherer import (
-    MacmonMetrics,
-    MacThunderboltConnections,
-    MacThunderboltIdentifiers,
-    MemoryUsage,
-    MiscData,
-    NodeConfig,
-    NodeNetworkInterfaces,
-    StaticNodeInformation,
-)
 
 
 def event_apply(event: Event, state: State) -> State:
@@ -55,12 +55,20 @@ def event_apply(event: Event, state: State) -> State:
             return apply_instance_created(event, state)
         case InstanceDeleted():
             return apply_instance_deleted(event, state)
+        case NodeCreated():
+            return apply_topology_node_created(event, state)
         case NodeTimedOut():
             return apply_node_timed_out(event, state)
+        case NodeIdentityMeasured():
+            return apply_node_identity_measured(event, state)
+        case NodeSystemMeasured():
+            return apply_node_system_measured(event, state)
+        case NodeNetworkMeasured():
+            return apply_node_network_measured(event, state)
         case NodeDownloadProgress():
             return apply_node_download_progress(event, state)
-        case NodeGatheredInfo():
-            return apply_node_gathered_info(event, state)
+        case NodeMemoryMeasured():
+            return apply_node_memory_measured(event, state)
         case RunnerDeleted():
             return apply_runner_deleted(event, state)
         case RunnerStatusUpdated():
@@ -192,89 +200,188 @@ def apply_runner_deleted(event: RunnerDeleted, state: State) -> State:
 
 
 def apply_node_timed_out(event: NodeTimedOut, state: State) -> State:
-    topology = copy.deepcopy(state.topology)
+    topology = copy.copy(state.topology)
     state.topology.remove_node(event.node_id)
-    node_profiles = {
-        key: value for key, value in state.node_profiles.items() if key != event.node_id
+    node_identities = {
+        key: value
+        for key, value in state.node_identities.items()
+        if key != event.node_id
+    }
+    node_memories = {
+        key: value for key, value in state.node_memories.items() if key != event.node_id
+    }
+    node_systems = {
+        key: value for key, value in state.node_systems.items() if key != event.node_id
+    }
+    node_networks = {
+        key: value for key, value in state.node_networks.items() if key != event.node_id
     }
     last_seen = {
         key: value for key, value in state.last_seen.items() if key != event.node_id
     }
-    downloads = {
-        key: value for key, value in state.downloads.items() if key != event.node_id
+    return state.model_copy(
+        update={
+            "topology": topology,
+            "node_identities": node_identities,
+            "node_memories": node_memories,
+            "node_systems": node_systems,
+            "node_networks": node_networks,
+            "last_seen": last_seen,
+        }
+    )
+
+
+def _reconstruct_profile(
+    node_id: NodeId,
+    state: State,
+    *,
+    identity: NodeIdentity | None = None,
+    memory: MemoryPerformanceProfile | None = None,
+    system: SystemPerformanceProfile | None = None,
+    network_interfaces: list[NetworkInterfaceInfo] | None = None,
+) -> NodePerformanceProfile:
+    """Reconstruct a NodePerformanceProfile from split state storage.
+
+    Uses provided overrides, falling back to state values.
+    """
+    ident = identity or state.node_identities.get(node_id)
+    mem = memory or state.node_memories.get(node_id)
+    sys = system or state.node_systems.get(node_id)
+    nets = (
+        network_interfaces
+        if network_interfaces is not None
+        else state.node_networks.get(node_id, [])
+    )
+
+    return NodePerformanceProfile(
+        model_id=ident.model_id if ident else None,
+        chip_id=ident.chip_id if ident else None,
+        friendly_name=ident.friendly_name if ident else None,
+        memory=mem,
+        network_interfaces=nets,
+        system=sys,
+    )
+
+
+def apply_node_identity_measured(event: NodeIdentityMeasured, state: State) -> State:
+    topology = copy.copy(state.topology)
+
+    identity = NodeIdentity(
+        model_id=event.model_id,
+        chip_id=event.chip_id,
+        friendly_name=event.friendly_name,
+    )
+    new_identities: Mapping[NodeId, NodeIdentity] = {
+        **state.node_identities,
+        event.node_id: identity,
     }
+    last_seen: Mapping[NodeId, datetime] = {
+        **state.last_seen,
+        event.node_id: datetime.fromisoformat(event.when),
+    }
+    if not topology.contains_node(event.node_id):
+        topology.add_node(NodeInfo(node_id=event.node_id))
+    reconstructed = _reconstruct_profile(event.node_id, state, identity=identity)
+    topology.update_node_profile(event.node_id, reconstructed)
     return state.model_copy(
         update={
-            "downloads": downloads,
+            "node_identities": new_identities,
             "topology": topology,
-            "node_profiles": node_profiles,
             "last_seen": last_seen,
         }
     )
 
 
-def apply_node_gathered_info(event: NodeGatheredInfo, state: State) -> State:
-    topology = copy.deepcopy(state.topology)
-    topology.add_node(event.node_id)
-    info = event.info
-    profile = state.node_profiles.get(event.node_id, NodePerformanceProfile())
-    match info:
-        case MacmonMetrics():
-            profile.system = info.system_profile
-            profile.memory = info.memory
-        case MemoryUsage():
-            profile.memory = info
-        case NodeConfig():
-            pass
-        case MiscData():
-            profile.friendly_name = info.friendly_name
-        case StaticNodeInformation():
-            profile.model_id = info.model
-            profile.chip_id = info.chip
-        case NodeNetworkInterfaces():
-            profile.network_interfaces = info.ifaces
-        case MacThunderboltIdentifiers():
-            profile.tb_interfaces = info.idents
-        case MacThunderboltConnections():
-            conn_map = {
-                tb_ident.domain_uuid: (nid, tb_ident.rdma_interface)
-                for nid in state.node_profiles
-                for tb_ident in state.node_profiles[nid].tb_interfaces
-            }
-            as_rdma_conns = [
-                Connection(
-                    source=event.node_id,
-                    sink=conn_map[tb_conn.sink_uuid][0],
-                    edge=RDMAConnection(
-                        source_rdma_iface=conn_map[tb_conn.source_uuid][1],
-                        sink_rdma_iface=conn_map[tb_conn.sink_uuid][1],
-                    ),
-                )
-                for tb_conn in info.conns
-                if tb_conn.source_uuid in conn_map
-                if tb_conn.sink_uuid in conn_map
-            ]
-            topology.replace_all_out_rdma_connections(event.node_id, as_rdma_conns)
+def apply_node_system_measured(event: NodeSystemMeasured, state: State) -> State:
+    topology = copy.copy(state.topology)
 
-    last_seen = {**state.last_seen, event.node_id: datetime.fromisoformat(event.when)}
-    new_profiles = {**state.node_profiles, event.node_id: profile}
+    new_systems: Mapping[NodeId, SystemPerformanceProfile] = {
+        **state.node_systems,
+        event.node_id: event.system,
+    }
+    last_seen: Mapping[NodeId, datetime] = {
+        **state.last_seen,
+        event.node_id: datetime.fromisoformat(event.when),
+    }
+    if not topology.contains_node(event.node_id):
+        topology.add_node(NodeInfo(node_id=event.node_id))
+    reconstructed = _reconstruct_profile(event.node_id, state, system=event.system)
+    topology.update_node_profile(event.node_id, reconstructed)
     return state.model_copy(
         update={
-            "node_profiles": new_profiles,
-            "last_seen": last_seen,
+            "node_systems": new_systems,
             "topology": topology,
+            "last_seen": last_seen,
         }
     )
 
 
+def apply_node_network_measured(event: NodeNetworkMeasured, state: State) -> State:
+    topology = copy.copy(state.topology)
+
+    new_networks: Mapping[NodeId, list[NetworkInterfaceInfo]] = {
+        **state.node_networks,
+        event.node_id: event.network_interfaces,
+    }
+    last_seen: Mapping[NodeId, datetime] = {
+        **state.last_seen,
+        event.node_id: datetime.fromisoformat(event.when),
+    }
+    if not topology.contains_node(event.node_id):
+        topology.add_node(NodeInfo(node_id=event.node_id))
+    reconstructed = _reconstruct_profile(
+        event.node_id, state, network_interfaces=event.network_interfaces
+    )
+    topology.update_node_profile(event.node_id, reconstructed)
+    return state.model_copy(
+        update={
+            "node_networks": new_networks,
+            "topology": topology,
+            "last_seen": last_seen,
+        }
+    )
+
+
+def apply_node_memory_measured(event: NodeMemoryMeasured, state: State) -> State:
+    topology = copy.copy(state.topology)
+
+    new_memories: Mapping[NodeId, MemoryPerformanceProfile] = {
+        **state.node_memories,
+        event.node_id: event.memory,
+    }
+    last_seen: Mapping[NodeId, datetime] = {
+        **state.last_seen,
+        event.node_id: datetime.fromisoformat(event.when),
+    }
+    if not topology.contains_node(event.node_id):
+        topology.add_node(NodeInfo(node_id=event.node_id))
+    reconstructed = _reconstruct_profile(event.node_id, state, memory=event.memory)
+    topology.update_node_profile(event.node_id, reconstructed)
+    return state.model_copy(
+        update={
+            "node_memories": new_memories,
+            "topology": topology,
+            "last_seen": last_seen,
+        }
+    )
+
+
+def apply_topology_node_created(event: NodeCreated, state: State) -> State:
+    topology = copy.copy(state.topology)
+    topology.add_node(NodeInfo(node_id=event.node_id))
+    return state.model_copy(update={"topology": topology})
+
+
 def apply_topology_edge_created(event: TopologyEdgeCreated, state: State) -> State:
-    topology = copy.deepcopy(state.topology)
-    topology.add_connection(event.conn)
+    topology = copy.copy(state.topology)
+    topology.add_connection(event.edge)
     return state.model_copy(update={"topology": topology})
 
 
 def apply_topology_edge_deleted(event: TopologyEdgeDeleted, state: State) -> State:
-    topology = copy.deepcopy(state.topology)
-    topology.remove_connection(event.conn)
+    topology = copy.copy(state.topology)
+    if not topology.contains_connection(event.edge):
+        return state
+    topology.remove_connection(event.edge)
     # TODO: Clean up removing the reverse connection
     return state.model_copy(update={"topology": topology})

src/exo/shared/constants.py

@@ -38,7 +38,6 @@ EXO_TEST_LOG = EXO_CACHE_HOME / "exo_test.log"
 
 # Identity (config)
 EXO_NODE_ID_KEYPAIR = EXO_CONFIG_HOME / "node_id.keypair"
-EXO_CONFIG_FILE = EXO_CONFIG_HOME / "config.toml"
 
 # libp2p topics for event forwarding
 LIBP2P_LOCAL_EVENTS_TOPIC = "worker_events"

src/exo/shared/logging.py

@@ -11,6 +11,9 @@ class InterceptLogger(HypercornLogger):
     def __init__(self, config: Config):
         super().__init__(config)
         assert self.error_logger
+        # TODO: Decide if we want to provide access logs
+        # assert self.access_logger
+        # self.access_logger.handlers = [_InterceptHandler()]
         self.error_logger.handlers = [_InterceptHandler()]
 
 

src/exo/shared/tests/test_apply/test_apply_node_download.py

@@ -43,4 +43,7 @@ def test_apply_two_node_download_progress():
         NodeDownloadProgress(download_progress=event2), state
     )
 
+    # TODO: This test is failing. We should support the following:
+    # 1. Downloading multiple models concurrently on the same node (one per runner is fine).
+    # 2. Downloading a model, it completes, then downloading a different model on the same node.
     assert new_state.downloads == {NodeId("node-1"): [event1, event2]}

src/exo/shared/tests/test_state_serialization.py

@@ -1,7 +1,7 @@
 from exo.shared.types.common import NodeId
 from exo.shared.types.multiaddr import Multiaddr
 from exo.shared.types.state import State
-from exo.shared.types.topology import Connection, SocketConnection
+from exo.shared.types.topology import Connection
 
 
 def test_state_serialization_roundtrip() -> None:
@@ -12,11 +12,9 @@ def test_state_serialization_roundtrip() -> None:
     node_b = NodeId("node-b")
 
     connection = Connection(
-        source=node_a,
-        sink=node_b,
-        edge=SocketConnection(
-            sink_multiaddr=Multiaddr(address="/ip4/127.0.0.1/tcp/10001"),
-        ),
+        local_node_id=node_a,
+        send_back_node_id=node_b,
+        send_back_multiaddr=Multiaddr(address="/ip4/127.0.0.1/tcp/10001"),
     )
 
     state = State()
@@ -25,11 +23,5 @@ def test_state_serialization_roundtrip() -> None:
     json_repr = state.model_dump_json()
     restored_state = State.model_validate_json(json_repr)
 
-    assert (
-        state.topology.to_snapshot().nodes
-        == restored_state.topology.to_snapshot().nodes
-    )
-    assert set(state.topology.to_snapshot().connections) == set(
-        restored_state.topology.to_snapshot().connections
-    )
+    assert state.topology.to_snapshot() == restored_state.topology.to_snapshot()
     assert restored_state.model_dump_json() == json_repr

src/exo/shared/topology.py

@@ -1,227 +1,203 @@
 import contextlib
-from collections.abc import Mapping, Sequence
-from dataclasses import dataclass, field
 from typing import Iterable
 
 import rustworkx as rx
 from pydantic import BaseModel, ConfigDict
 
 from exo.shared.types.common import NodeId
-from exo.shared.types.topology import (
-    Connection,
-    Cycle,
-    RDMAConnection,
-    SocketConnection,
-)
+from exo.shared.types.profiling import ConnectionProfile, NodePerformanceProfile
+from exo.shared.types.topology import Connection, NodeInfo
 
 
 class TopologySnapshot(BaseModel):
-    nodes: Sequence[NodeId]
-    connections: Mapping[
-        NodeId, Mapping[NodeId, Sequence[SocketConnection | RDMAConnection]]
-    ]
+    nodes: list[NodeInfo]
+    connections: list[Connection]
 
-    model_config = ConfigDict(frozen=True, extra="forbid")
+    model_config = ConfigDict(frozen=True, extra="forbid", strict=True)
 
 
-@dataclass
 class Topology:
-    _graph: rx.PyDiGraph[NodeId, SocketConnection | RDMAConnection] = field(
-        init=False, default_factory=rx.PyDiGraph
-    )
-    _vertex_indices: dict[NodeId, int] = field(init=False, default_factory=dict)
+    def __init__(self) -> None:
+        self._graph: rx.PyDiGraph[NodeInfo, Connection] = rx.PyDiGraph()
+        self._node_id_to_rx_id_map: dict[NodeId, int] = dict()
+        self._rx_id_to_node_id_map: dict[int, NodeId] = dict()
+        self._edge_id_to_rx_id_map: dict[Connection, int] = dict()
 
     def to_snapshot(self) -> TopologySnapshot:
         return TopologySnapshot(
-            nodes=list(self.list_nodes()), connections=self.map_connections()
+            nodes=list(self.list_nodes()),
+            connections=list(self.list_connections()),
         )
 
     @classmethod
     def from_snapshot(cls, snapshot: TopologySnapshot) -> "Topology":
         topology = cls()
 
-        for node_id in snapshot.nodes:
+        for node in snapshot.nodes:
             with contextlib.suppress(ValueError):
-                topology.add_node(node_id)
+                topology.add_node(node)
 
-        for source in snapshot.connections:
-            for sink in snapshot.connections[source]:
-                for edge in snapshot.connections[source][sink]:
-                    topology.add_connection(
-                        Connection(source=source, sink=sink, edge=edge)
-                    )
+        for connection in snapshot.connections:
+            topology.add_connection(connection)
 
         return topology
 
-    def add_node(self, node_id: NodeId) -> None:
-        if node_id in self._vertex_indices:
+    def add_node(self, node: NodeInfo) -> None:
+        if node.node_id in self._node_id_to_rx_id_map:
             return
-        rx_id = self._graph.add_node(node_id)
-        self._vertex_indices[node_id] = rx_id
+        rx_id = self._graph.add_node(node)
+        self._node_id_to_rx_id_map[node.node_id] = rx_id
+        self._rx_id_to_node_id_map[rx_id] = node.node_id
 
     def node_is_leaf(self, node_id: NodeId) -> bool:
         return (
-            node_id in self._vertex_indices
-            and len(self._graph.neighbors(self._vertex_indices[node_id])) <= 1
+            node_id in self._node_id_to_rx_id_map
+            and len(self._graph.neighbors(self._node_id_to_rx_id_map[node_id])) == 1
         )
 
     def neighbours(self, node_id: NodeId) -> list[NodeId]:
         return [
-            self._graph[rx_id]
-            for rx_id in self._graph.neighbors(self._vertex_indices[node_id])
+            self._rx_id_to_node_id_map[rx_id]
+            for rx_id in self._graph.neighbors(self._node_id_to_rx_id_map[node_id])
         ]
 
-    def out_edges(self, node_id: NodeId) -> Iterable[Connection]:
-        if node_id not in self._vertex_indices:
+    def out_edges(self, node_id: NodeId) -> list[tuple[NodeId, Connection]]:
+        if node_id not in self._node_id_to_rx_id_map:
             return []
-        return (
-            Connection(source=self._graph[source], sink=self._graph[sink], edge=edge)
-            for source, sink, edge in self._graph.out_edges(
-                self._vertex_indices[node_id]
+        return [
+            (self._rx_id_to_node_id_map[nid], conn)
+            for _, nid, conn in self._graph.out_edges(
+                self._node_id_to_rx_id_map[node_id]
             )
-        )
+        ]
 
     def contains_node(self, node_id: NodeId) -> bool:
-        return node_id in self._vertex_indices
+        return node_id in self._node_id_to_rx_id_map
 
-    def add_connection(self, conn: Connection) -> None:
-        source, sink, edge = conn.source, conn.sink, conn.edge
-        del conn
-        if edge in self.get_all_connections_between(source, sink):
+    def contains_connection(self, connection: Connection) -> bool:
+        return connection in self._edge_id_to_rx_id_map
+
+    def add_connection(
+        self,
+        connection: Connection,
+    ) -> None:
+        if connection.local_node_id not in self._node_id_to_rx_id_map:
+            self.add_node(NodeInfo(node_id=connection.local_node_id))
+        if connection.send_back_node_id not in self._node_id_to_rx_id_map:
+            self.add_node(NodeInfo(node_id=connection.send_back_node_id))
+
+        if connection in self._edge_id_to_rx_id_map:
             return
 
-        if source not in self._vertex_indices:
-            self.add_node(source)
-        if sink not in self._vertex_indices:
-            self.add_node(sink)
+        src_id = self._node_id_to_rx_id_map[connection.local_node_id]
+        sink_id = self._node_id_to_rx_id_map[connection.send_back_node_id]
 
-        src_id = self._vertex_indices[source]
-        sink_id = self._vertex_indices[sink]
+        rx_id = self._graph.add_edge(src_id, sink_id, connection)
+        self._edge_id_to_rx_id_map[connection] = rx_id
 
-        _ = self._graph.add_edge(src_id, sink_id, edge)
+    def list_nodes(self) -> Iterable[NodeInfo]:
+        return (self._graph[i] for i in self._graph.node_indices())
 
-    def get_all_connections_between(
-        self, source: NodeId, sink: NodeId
-    ) -> Iterable[SocketConnection | RDMAConnection]:
-        if source not in self._vertex_indices:
-            return []
-        if sink not in self._vertex_indices:
-            return []
+    def list_connections(self) -> Iterable[Connection]:
+        return (connection for _, _, connection in self._graph.weighted_edge_list())
 
-        src_id = self._vertex_indices[source]
-        sink_id = self._vertex_indices[sink]
+    def get_node_profile(self, node_id: NodeId) -> NodePerformanceProfile | None:
         try:
-            return self._graph.get_all_edge_data(src_id, sink_id)
-        except rx.NoEdgeBetweenNodes:
-            return []
+            rx_idx = self._node_id_to_rx_id_map[node_id]
+            return self._graph.get_node_data(rx_idx).node_profile
+        except KeyError:
+            return None
 
-    def list_nodes(self) -> Iterable[NodeId]:
-        return self._graph.nodes()
+    def update_node_profile(
+        self, node_id: NodeId, node_profile: NodePerformanceProfile
+    ) -> None:
+        rx_idx = self._node_id_to_rx_id_map[node_id]
+        self._graph[rx_idx].node_profile = node_profile
 
-    def map_connections(
-        self,
-    ) -> Mapping[NodeId, Mapping[NodeId, Sequence[SocketConnection | RDMAConnection]]]:
-        base: dict[NodeId, dict[NodeId, list[SocketConnection | RDMAConnection]]] = {}
-        for src_id, sink_id, connection in self._graph.weighted_edge_list():
-            source = self._graph[src_id]
-            sink = self._graph[sink_id]
-            if source not in base:
-                base[source] = {}
-            if sink not in base[source]:
-                base[source][sink] = []
-            base[source][sink].append(connection)
-        return base
+    def update_connection_profile(self, connection: Connection) -> None:
+        rx_idx = self._edge_id_to_rx_id_map[connection]
+        self._graph.update_edge_by_index(rx_idx, connection)
 
-    def list_connections(
-        self,
-    ) -> Iterable[Connection]:
-        return (
-            (
-                Connection(
-                    source=self._graph[src_id],
-                    sink=self._graph[sink_id],
-                    edge=connection,
-                )
-            )
-            for src_id, sink_id, connection in self._graph.weighted_edge_list()
-        )
+    def get_connection_profile(
+        self, connection: Connection
+    ) -> ConnectionProfile | None:
+        try:
+            rx_idx = self._edge_id_to_rx_id_map[connection]
+            return self._graph.get_edge_data_by_index(rx_idx).connection_profile
+        except KeyError:
+            return None
 
     def remove_node(self, node_id: NodeId) -> None:
-        if node_id not in self._vertex_indices:
+        if node_id not in self._node_id_to_rx_id_map:
             return
 
-        rx_idx = self._vertex_indices[node_id]
+        for connection in self.list_connections():
+            if (
+                connection.local_node_id == node_id
+                or connection.send_back_node_id == node_id
+            ):
+                self.remove_connection(connection)
+
+        rx_idx = self._node_id_to_rx_id_map[node_id]
         self._graph.remove_node(rx_idx)
 
-        del self._vertex_indices[node_id]
+        del self._node_id_to_rx_id_map[node_id]
+        del self._rx_id_to_node_id_map[rx_idx]
 
-    def replace_all_out_rdma_connections(
-        self, source: NodeId, new_connections: Sequence[Connection]
-    ) -> None:
-        for conn_idx in self._graph.out_edge_indices(self._vertex_indices[source]):
-            if isinstance(self._graph.get_edge_data_by_index(conn_idx), RDMAConnection):
-                self._graph.remove_edge_from_index(conn_idx)
-        for conn in new_connections:
-            self.add_connection(conn)
-
-    def remove_connection(self, conn: Connection) -> None:
-        if (
-            conn.source not in self._vertex_indices
-            or conn.sink not in self._vertex_indices
-        ):
+    def remove_connection(self, connection: Connection) -> None:
+        if connection not in self._edge_id_to_rx_id_map:
             return
-        for conn_idx in self._graph.edge_indices_from_endpoints(
-            self._vertex_indices[conn.source], self._vertex_indices[conn.sink]
-        ):
-            if self._graph.get_edge_data_by_index(conn_idx) == conn.edge:
-                self._graph.remove_edge_from_index(conn_idx)
-
-    def get_cycles(self) -> list[Cycle]:
-        """Get simple cycles in the graph, including singleton cycles"""
+        rx_idx = self._edge_id_to_rx_id_map[connection]
+        self._graph.remove_edge_from_index(rx_idx)
+        del self._edge_id_to_rx_id_map[connection]
 
+    def get_cycles(self) -> list[list[NodeInfo]]:
         cycle_idxs = rx.simple_cycles(self._graph)
-        cycles: list[Cycle] = []
+        cycles: list[list[NodeInfo]] = []
         for cycle_idx in cycle_idxs:
-            cycle = Cycle(node_ids=[self._graph[idx] for idx in cycle_idx])
+            cycle = [self._graph[idx] for idx in cycle_idx]
             cycles.append(cycle)
-        for node_id in self.list_nodes():
-            cycles.append(Cycle(node_ids=[node_id]))
+
         return cycles
 
-    def get_cycles_tb(self) -> list[Cycle]:
+    def get_cycles_tb(self) -> list[list[NodeInfo]]:
         tb_edges = [
             (u, v, conn)
             for u, v, conn in self._graph.weighted_edge_list()
             if conn.is_thunderbolt()
         ]
 
-        tb_graph: rx.PyDiGraph[NodeId, SocketConnection] = rx.PyDiGraph()
+        tb_graph: rx.PyDiGraph[NodeInfo, Connection] = rx.PyDiGraph()
         tb_graph.add_nodes_from(self._graph.nodes())
 
         for u, v, conn in tb_edges:
-            if isinstance(conn, SocketConnection):
-                tb_graph.add_edge(u, v, conn)
+            tb_graph.add_edge(u, v, conn)
 
         cycle_idxs = rx.simple_cycles(tb_graph)
-        cycles: list[Cycle] = []
+        cycles: list[list[NodeInfo]] = []
         for cycle_idx in cycle_idxs:
-            cycle = Cycle(node_ids=[tb_graph[idx] for idx in cycle_idx])
+            cycle = [tb_graph[idx] for idx in cycle_idx]
             cycles.append(cycle)
 
         return cycles
 
-    def get_subgraph_from_nodes(self, node_ids: list[NodeId]) -> "Topology":
+    def get_subgraph_from_nodes(self, nodes: list[NodeInfo]) -> "Topology":
+        node_idxs = [node.node_id for node in nodes]
+        rx_idxs = [self._node_id_to_rx_id_map[idx] for idx in node_idxs]
         topology = Topology()
-        for node_id in node_ids:
-            topology.add_node(node_id)
+        for rx_idx in rx_idxs:
+            topology.add_node(self._graph[rx_idx])
         for connection in self.list_connections():
-            if connection.source in node_ids and connection.sink in node_ids:
+            if (
+                connection.local_node_id in node_idxs
+                and connection.send_back_node_id in node_idxs
+            ):
                 topology.add_connection(connection)
         return topology
 
-    def is_thunderbolt_cycle(self, cycle: Cycle) -> bool:
-        node_idxs = [node for node in cycle]
-        rx_idxs = [self._vertex_indices[idx] for idx in node_idxs]
+    def is_thunderbolt_cycle(self, cycle: list[NodeInfo]) -> bool:
+        node_idxs = [node.node_id for node in cycle]
+        rx_idxs = [self._node_id_to_rx_id_map[idx] for idx in node_idxs]
         for rid in rx_idxs:
             for neighbor_rid in self._graph.neighbors(rid):
                 if neighbor_rid not in rx_idxs:

src/exo/shared/types/api.py

@@ -11,10 +11,21 @@ from exo.shared.types.worker.instances import Instance, InstanceId, InstanceMeta
 from exo.shared.types.worker.shards import Sharding
 
 FinishReason = Literal[
-    "stop", "length", "tool_calls", "content_filter", "function_call"
+    "stop", "length", "tool_calls", "content_filter", "function_call", "error"
 ]
 
 
+class ErrorInfo(BaseModel):
+    message: str
+    type: str
+    param: str | None = None
+    code: int
+
+
+class ErrorResponse(BaseModel):
+    error: ErrorInfo
+
+
 class ModelListModel(BaseModel):
     id: str
     object: str = "model"

src/exo/shared/types/chunks.py

@@ -22,6 +22,7 @@ class TokenChunk(BaseChunk):
     token_id: int
     finish_reason: FinishReason | None = None
     stats: GenerationStats | None = None
+    error_message: str | None = None
 
 
 class ImageChunk(BaseChunk):

src/exo/shared/types/events.py

@@ -5,11 +5,15 @@ from pydantic import Field
 from exo.shared.topology import Connection
 from exo.shared.types.chunks import GenerationChunk
 from exo.shared.types.common import CommandId, Id, NodeId, SessionId
+from exo.shared.types.profiling import (
+    MemoryPerformanceProfile,
+    NetworkInterfaceInfo,
+    SystemPerformanceProfile,
+)
 from exo.shared.types.tasks import Task, TaskId, TaskStatus
 from exo.shared.types.worker.downloads import DownloadProgress
 from exo.shared.types.worker.instances import Instance, InstanceId
 from exo.shared.types.worker.runners import RunnerId, RunnerStatus
-from exo.utils.info_gatherer.info_gatherer import GatheredInfo
 from exo.utils.pydantic_ext import CamelCaseModel, TaggedModel
 
 
@@ -76,15 +80,47 @@ class RunnerDeleted(BaseEvent):
     runner_id: RunnerId
 
 
+# TODO
+class NodeCreated(BaseEvent):
+    node_id: NodeId
+
+
 class NodeTimedOut(BaseEvent):
     node_id: NodeId
 
 
-# TODO: bikeshed this name
-class NodeGatheredInfo(BaseEvent):
+class NodeIdentityMeasured(BaseEvent):
+    """Static identity info - emitted once at startup."""
+
     node_id: NodeId
     when: str  # this is a manually cast datetime overrode by the master when the event is indexed, rather than the local time on the device
-    info: GatheredInfo
+    model_id: str
+    chip_id: str
+    friendly_name: str
+
+
+class NodeSystemMeasured(BaseEvent):
+    """Dynamic system metrics (GPU, temp, power) - emitted at 1s intervals."""
+
+    node_id: NodeId
+    when: str  # this is a manually cast datetime overrode by the master when the event is indexed, rather than the local time on the device
+    system: SystemPerformanceProfile
+
+
+class NodeNetworkMeasured(BaseEvent):
+    """Semi-static network interface info - emitted at 30s intervals."""
+
+    node_id: NodeId
+    when: str  # this is a manually cast datetime overrode by the master when the event is indexed, rather than the local time on the device
+    network_interfaces: list[NetworkInterfaceInfo]
+
+
+class NodeMemoryMeasured(BaseEvent):
+    """Dynamic memory metrics - emitted at 0.5s intervals."""
+
+    node_id: NodeId
+    when: str  # this is a manually cast datetime overrode by the master when the event is indexed, rather than the local time on the device
+    memory: MemoryPerformanceProfile
 
 
 class NodeDownloadProgress(BaseEvent):
@@ -97,11 +133,11 @@ class ChunkGenerated(BaseEvent):
 
 
 class TopologyEdgeCreated(BaseEvent):
-    conn: Connection
+    edge: Connection
 
 
 class TopologyEdgeDeleted(BaseEvent):
-    conn: Connection
+    edge: Connection
 
 
 Event = (
@@ -115,8 +151,12 @@ Event = (
     | InstanceDeleted
     | RunnerStatusUpdated
     | RunnerDeleted
+    | NodeCreated
     | NodeTimedOut
-    | NodeGatheredInfo
+    | NodeIdentityMeasured
+    | NodeSystemMeasured
+    | NodeNetworkMeasured
+    | NodeMemoryMeasured
     | NodeDownloadProgress
     | ChunkGenerated
     | TopologyEdgeCreated

src/exo/shared/types/multiaddr.py

@@ -1,11 +1,10 @@
 import re
 from typing import ClassVar
 
-from pydantic import BaseModel, ConfigDict, computed_field, field_validator
+from pydantic import BaseModel, computed_field, field_validator
 
 
 class Multiaddr(BaseModel):
-    model_config = ConfigDict(frozen=True)
     address: str
 
     PATTERNS: ClassVar[list[str]] = [

src/exo/shared/types/profiling.py

@@ -1,14 +1,12 @@
-from collections.abc import Sequence
 from typing import Self
 
 import psutil
 
 from exo.shared.types.memory import Memory
-from exo.shared.types.thunderbolt import ThunderboltIdentifier
 from exo.utils.pydantic_ext import CamelCaseModel
 
 
-class MemoryUsage(CamelCaseModel):
+class MemoryPerformanceProfile(CamelCaseModel):
     ram_total: Memory
     ram_available: Memory
     swap_total: Memory
@@ -46,6 +44,7 @@ class SystemPerformanceProfile(CamelCaseModel):
     sys_power: float = 0.0
     pcpu_usage: float = 0.0
     ecpu_usage: float = 0.0
+    ane_power: float = 0.0
 
 
 class NetworkInterfaceInfo(CamelCaseModel):
@@ -53,13 +52,24 @@ class NetworkInterfaceInfo(CamelCaseModel):
     ip_address: str
 
 
+class NodeIdentity(CamelCaseModel):
+    """Static identity info for a node."""
+
+    model_id: str
+    chip_id: str
+    friendly_name: str
+
+
 class NodePerformanceProfile(CamelCaseModel):
-    model_id: str = "Unknown"
-    chip_id: str = "Unknown"
-    friendly_name: str = "Unknown"
-    memory: MemoryUsage = MemoryUsage.from_bytes(
-        ram_total=0, ram_available=0, swap_total=0, swap_available=0
-    )
-    network_interfaces: Sequence[NetworkInterfaceInfo] = []
-    tb_interfaces: Sequence[ThunderboltIdentifier] = []
-    system: SystemPerformanceProfile = SystemPerformanceProfile()
+    model_id: str | None = None
+    chip_id: str | None = None
+    friendly_name: str | None = None
+    memory: MemoryPerformanceProfile | None = None
+    network_interfaces: list[NetworkInterfaceInfo] = []
+    system: SystemPerformanceProfile | None = None
+
+
+class ConnectionProfile(CamelCaseModel):
+    throughput: float
+    latency: float
+    jitter: float

src/exo/shared/types/state.py

@@ -7,7 +7,12 @@ from pydantic.alias_generators import to_camel
 
 from exo.shared.topology import Topology, TopologySnapshot
 from exo.shared.types.common import NodeId
-from exo.shared.types.profiling import NodePerformanceProfile
+from exo.shared.types.profiling import (
+    MemoryPerformanceProfile,
+    NetworkInterfaceInfo,
+    NodeIdentity,
+    SystemPerformanceProfile,
+)
 from exo.shared.types.tasks import Task, TaskId
 from exo.shared.types.worker.downloads import DownloadProgress
 from exo.shared.types.worker.instances import Instance, InstanceId
@@ -35,7 +40,10 @@ class State(CamelCaseModel):
     runners: Mapping[RunnerId, RunnerStatus] = {}
     downloads: Mapping[NodeId, Sequence[DownloadProgress]] = {}
     tasks: Mapping[TaskId, Task] = {}
-    node_profiles: Mapping[NodeId, NodePerformanceProfile] = {}
+    node_identities: Mapping[NodeId, NodeIdentity] = {}
+    node_memories: Mapping[NodeId, MemoryPerformanceProfile] = {}
+    node_systems: Mapping[NodeId, SystemPerformanceProfile] = {}
+    node_networks: Mapping[NodeId, list[NetworkInterfaceInfo]] = {}
     last_seen: Mapping[NodeId, datetime] = {}
     topology: Topology = Field(default_factory=Topology)
     last_event_applied_idx: int = Field(default=-1, ge=-1)

src/exo/shared/types/thunderbolt.py

@@ -1,81 +0,0 @@
-import anyio
-from pydantic import BaseModel, Field
-
-from exo.utils.pydantic_ext import CamelCaseModel
-
-
-class ThunderboltConnection(CamelCaseModel):
-    source_uuid: str
-    sink_uuid: str
-
-
-class ThunderboltIdentifier(CamelCaseModel):
-    rdma_interface: str
-    domain_uuid: str
-
-
-## Intentionally minimal, only collecting data we care about - there's a lot more
-
-
-class _ReceptacleTag(BaseModel, extra="ignore"):
-    receptacle_id_key: str | None = None
-
-
-class _ConnectivityItem(BaseModel, extra="ignore"):
-    domain_uuid_key: str | None = None
-
-
-class ThunderboltConnectivityData(BaseModel, extra="ignore"):
-    domain_uuid_key: str | None = None
-    items: list[_ConnectivityItem] | None = Field(None, alias="_items")
-    receptacle_1_tag: _ReceptacleTag | None = None
-
-    def ident(self, ifaces: dict[str, str]) -> ThunderboltIdentifier | None:
-        if (
-            self.domain_uuid_key is None
-            or self.receptacle_1_tag is None
-            or self.receptacle_1_tag.receptacle_id_key is None
-        ):
-            return
-        tag = f"Thunderbolt {self.receptacle_1_tag.receptacle_id_key}"
-        assert tag in ifaces  # doesn't need to be an assertion but im confident
-        # if tag not in ifaces: return None
-        iface = f"rdma_{ifaces[tag]}"
-        return ThunderboltIdentifier(
-            rdma_interface=iface, domain_uuid=self.domain_uuid_key
-        )
-
-    def conn(self) -> ThunderboltConnection | None:
-        if self.domain_uuid_key is None or self.items is None:
-            return
-
-        sink_key = next(
-            (
-                item.domain_uuid_key
-                for item in self.items
-                if item.domain_uuid_key is not None
-            ),
-            None,
-        )
-        if sink_key is None:
-            return None
-
-        return ThunderboltConnection(
-            source_uuid=self.domain_uuid_key, sink_uuid=sink_key
-        )
-
-
-class ThunderboltConnectivity(BaseModel, extra="ignore"):
-    SPThunderboltDataType: list[ThunderboltConnectivityData] = []
-
-    @classmethod
-    async def gather(cls) -> list[ThunderboltConnectivityData] | None:
-        proc = await anyio.run_process(
-            ["system_profiler", "SPThunderboltDataType", "-json"], check=False
-        )
-        if proc.returncode != 0:
-            return None
-        # Saving you from PascalCase while avoiding too much pydantic
-        return ThunderboltConnectivity.model_validate_json(
-            proc.stdout
-        ).SPThunderboltDataType

src/exo/shared/types/topology.py

@@ -1,41 +1,37 @@
-from collections.abc import Iterator
-from dataclasses import dataclass
-
 from exo.shared.types.common import NodeId
 from exo.shared.types.multiaddr import Multiaddr
-from exo.utils.pydantic_ext import FrozenModel
+from exo.shared.types.profiling import ConnectionProfile, NodePerformanceProfile
+from exo.utils.pydantic_ext import CamelCaseModel
 
 
-@dataclass(frozen=True)
-class Cycle:
-    node_ids: list[NodeId]
-
-    def __len__(self) -> int:
-        return self.node_ids.__len__()
-
-    def __iter__(self) -> Iterator[NodeId]:
-        return self.node_ids.__iter__()
+class NodeInfo(CamelCaseModel):
+    node_id: NodeId
+    node_profile: NodePerformanceProfile | None = None
 
 
-class RDMAConnection(FrozenModel):
-    source_rdma_iface: str
-    sink_rdma_iface: str
+class Connection(CamelCaseModel):
+    local_node_id: NodeId
+    send_back_node_id: NodeId
+    send_back_multiaddr: Multiaddr
+    connection_profile: ConnectionProfile | None = None
+
+    def __hash__(self) -> int:
+        return hash(
+            (
+                self.local_node_id,
+                self.send_back_node_id,
+                self.send_back_multiaddr.address,
+            )
+        )
+
+    def __eq__(self, other: object) -> bool:
+        if not isinstance(other, Connection):
+            raise ValueError("Cannot compare Connection with non-Connection")
+        return (
+            self.local_node_id == other.local_node_id
+            and self.send_back_node_id == other.send_back_node_id
+            and self.send_back_multiaddr == other.send_back_multiaddr
+        )
 
     def is_thunderbolt(self) -> bool:
-        return True
-
-
-class SocketConnection(FrozenModel):
-    sink_multiaddr: Multiaddr
-
-    def __hash__(self):
-        return hash(self.sink_multiaddr.ip_address)
-
-    def is_thunderbolt(self) -> bool:
-        return str(self.sink_multiaddr.ipv4_address).startswith("169.254")
-
-
-class Connection(FrozenModel):
-    source: NodeId
-    sink: NodeId
-    edge: RDMAConnection | SocketConnection
+        return str(self.send_back_multiaddr.ipv4_address).startswith("169.254")

src/exo/shared/types/worker/instances.py

@@ -30,7 +30,7 @@ class MlxRingInstance(BaseInstance):
 
 
 class MlxJacclInstance(BaseInstance):
-    jaccl_devices: list[list[str | None]]
+    ibv_devices: list[list[str | None]]
     jaccl_coordinators: dict[NodeId, str]
 
 

src/exo/shared/types/worker/resource_monitor.py

@@ -0,0 +1,43 @@
+import asyncio
+from abc import ABC, abstractmethod
+from collections.abc import Coroutine
+from typing import Callable
+
+from exo.shared.types.profiling import (
+    MemoryPerformanceProfile,
+    SystemPerformanceProfile,
+)
+
+
+class ResourceCollector(ABC):
+    @abstractmethod
+    async def collect(self) -> SystemPerformanceProfile | MemoryPerformanceProfile: ...
+
+
+class SystemResourceCollector(ResourceCollector):
+    async def collect(self) -> SystemPerformanceProfile: ...
+
+
+class MemoryResourceCollector(ResourceCollector):
+    async def collect(self) -> MemoryPerformanceProfile: ...
+
+
+class ResourceMonitor:
+    data_collectors: list[ResourceCollector]
+    effect_handlers: set[
+        Callable[[SystemPerformanceProfile | MemoryPerformanceProfile], None]
+    ]
+
+    async def _collect(
+        self,
+    ) -> list[SystemPerformanceProfile | MemoryPerformanceProfile]:
+        tasks: list[
+            Coroutine[None, None, SystemPerformanceProfile | MemoryPerformanceProfile]
+        ] = [collector.collect() for collector in self.data_collectors]
+        return await asyncio.gather(*tasks)
+
+    async def collect(self) -> None:
+        profiles = await self._collect()
+        for profile in profiles:
+            for effect_handler in self.effect_handlers:
+                effect_handler(profile)

src/exo/utils/info_gatherer/info_gatherer.py

@@ -1,235 +0,0 @@
-import os
-import shutil
-import sys
-import tomllib
-from collections.abc import Sequence
-from dataclasses import dataclass, field
-from subprocess import CalledProcessError
-from typing import Self, cast
-
-import anyio
-from anyio import create_task_group, open_process
-from anyio.abc import TaskGroup
-from anyio.streams.buffered import BufferedByteReceiveStream
-from anyio.streams.text import TextReceiveStream
-from loguru import logger
-
-from exo.shared.constants import EXO_CONFIG_FILE
-from exo.shared.types.memory import Memory
-from exo.shared.types.profiling import (
-    MemoryUsage,
-    NetworkInterfaceInfo,
-)
-from exo.shared.types.thunderbolt import (
-    ThunderboltConnection,
-    ThunderboltConnectivity,
-    ThunderboltIdentifier,
-)
-from exo.utils.channels import Sender
-from exo.utils.pydantic_ext import TaggedModel
-
-from .macmon import MacmonMetrics
-from .system_info import get_friendly_name, get_model_and_chip, get_network_interfaces
-
-IS_DARWIN = sys.platform == "darwin"
-
-
-class StaticNodeInformation(TaggedModel):
-    """Node information that should NEVER change, to be gathered once at startup"""
-
-    model: str
-    chip: str
-
-    @classmethod
-    async def gather(cls) -> Self:
-        model, chip = await get_model_and_chip()
-        return cls(model=model, chip=chip)
-
-
-class NodeNetworkInterfaces(TaggedModel):
-    ifaces: Sequence[NetworkInterfaceInfo]
-
-
-class MacThunderboltIdentifiers(TaggedModel):
-    idents: Sequence[ThunderboltIdentifier]
-
-
-class MacThunderboltConnections(TaggedModel):
-    conns: Sequence[ThunderboltConnection]
-
-
-class NodeConfig(TaggedModel):
-    """Node configuration from EXO_CONFIG_FILE, reloaded from the file only at startup. Other changes should come in through the API and propagate from there"""
-
-    @classmethod
-    async def gather(cls) -> Self | None:
-        cfg_file = anyio.Path(EXO_CONFIG_FILE)
-        await cfg_file.touch(exist_ok=True)
-        async with await cfg_file.open("rb") as f:
-            try:
-                contents = (await f.read()).decode("utf-8")
-                data = tomllib.loads(contents)
-                return cls.model_validate(data)
-            except (tomllib.TOMLDecodeError, UnicodeDecodeError):
-                logger.warning("Invalid config file, skipping...")
-                return None
-
-
-class MiscData(TaggedModel):
-    """Node information that may slowly change that doesn't fall into the other categories"""
-
-    friendly_name: str
-
-    @classmethod
-    async def gather(cls) -> Self:
-        return cls(friendly_name=await get_friendly_name())
-
-
-async def _gather_iface_map() -> dict[str, str] | None:
-    proc = await anyio.run_process(
-        ["networksetup", "-listallhardwareports"], check=False
-    )
-    if proc.returncode != 0:
-        return None
-
-    ports: dict[str, str] = {}
-    port = ""
-    for line in proc.stdout.decode("utf-8").split("\n"):
-        if line.startswith("Hardware Port:"):
-            port = line.split(": ")[1]
-        elif line.startswith("Device:"):
-            ports[port] = line.split(": ")[1]
-            port = ""
-    if "" in ports:
-        del ports[""]
-    return ports
-
-
-GatheredInfo = (
-    MacmonMetrics
-    | MemoryUsage
-    | NodeNetworkInterfaces
-    | MacThunderboltIdentifiers
-    | MacThunderboltConnections
-    | NodeConfig
-    | MiscData
-    | StaticNodeInformation
-)
-
-
-@dataclass
-class InfoGatherer:
-    info_sender: Sender[GatheredInfo]
-    interface_watcher_interval: float | None = 10
-    misc_poll_interval: float | None = 60
-    system_profiler_interval: float | None = 5 if IS_DARWIN else None
-    memory_poll_rate: float | None = None if IS_DARWIN else 1
-    macmon_interval: float | None = 1 if IS_DARWIN else None
-    _tg: TaskGroup = field(init=False, default_factory=create_task_group)
-
-    async def run(self):
-        async with self._tg as tg:
-            if IS_DARWIN:
-                if (macmon_path := shutil.which("macmon")) is not None:
-                    tg.start_soon(self._monitor_macmon, macmon_path)
-                tg.start_soon(self._monitor_system_profiler_thunderbolt_data)
-            tg.start_soon(self._watch_system_info)
-            tg.start_soon(self._monitor_memory_usage)
-            tg.start_soon(self._monitor_misc)
-
-            nc = await NodeConfig.gather()
-            if nc is not None:
-                await self.info_sender.send(nc)
-            sni = await StaticNodeInformation.gather()
-            await self.info_sender.send(sni)
-
-    def shutdown(self):
-        self._tg.cancel_scope.cancel()
-
-    async def _monitor_misc(self):
-        if self.misc_poll_interval is None:
-            return
-        prev = await MiscData.gather()
-        await self.info_sender.send(prev)
-        while True:
-            curr = await MiscData.gather()
-            if prev != curr:
-                prev = curr
-                await self.info_sender.send(curr)
-            await anyio.sleep(self.misc_poll_interval)
-
-    async def _monitor_system_profiler_thunderbolt_data(self):
-        if self.system_profiler_interval is None:
-            return
-        iface_map = await _gather_iface_map()
-        if iface_map is None:
-            return
-
-        old_idents = []
-        while True:
-            data = await ThunderboltConnectivity.gather()
-            assert data is not None
-
-            idents = [it for i in data if (it := i.ident(iface_map)) is not None]
-            if idents != old_idents:
-                await self.info_sender.send(MacThunderboltIdentifiers(idents=idents))
-            old_idents = idents
-
-            conns = [it for i in data if (it := i.conn()) is not None]
-            await self.info_sender.send(MacThunderboltConnections(conns=conns))
-
-            await anyio.sleep(self.system_profiler_interval)
-
-    async def _monitor_memory_usage(self):
-        override_memory_env = os.getenv("OVERRIDE_MEMORY_MB")
-        override_memory: int | None = (
-            Memory.from_mb(int(override_memory_env)).in_bytes
-            if override_memory_env
-            else None
-        )
-        if self.memory_poll_rate is None:
-            return
-        while True:
-            await self.info_sender.send(
-                MemoryUsage.from_psutil(override_memory=override_memory)
-            )
-            await anyio.sleep(self.memory_poll_rate)
-
-    async def _watch_system_info(self):
-        if self.interface_watcher_interval is None:
-            return
-        old_nics = []
-        while True:
-            nics = get_network_interfaces()
-            if nics != old_nics:
-                old_nics = nics
-                await self.info_sender.send(NodeNetworkInterfaces(ifaces=nics))
-            await anyio.sleep(self.interface_watcher_interval)
-
-    async def _monitor_macmon(self, macmon_path: str):
-        if self.macmon_interval is None:
-            return
-        # macmon pipe --interval [interval in ms]
-        try:
-            async with await open_process(
-                [macmon_path, "pipe", "--interval", str(self.macmon_interval * 1000)]
-            ) as p:
-                if not p.stdout:
-                    logger.critical("MacMon closed stdout")
-                    return
-                async for text in TextReceiveStream(
-                    BufferedByteReceiveStream(p.stdout)
-                ):
-                    await self.info_sender.send(MacmonMetrics.from_raw_json(text))
-        except CalledProcessError as e:
-            stderr_msg = "no stderr"
-            stderr_output = cast(bytes | str | None, e.stderr)
-            if stderr_output is not None:
-                stderr_msg = (
-                    stderr_output.decode()
-                    if isinstance(stderr_output, bytes)
-                    else str(stderr_output)
-                )
-            logger.warning(
-                f"MacMon failed with return code {e.returncode}: {stderr_msg}"
-            )

src/exo/utils/info_gatherer/macmon.py

@@ -1,70 +0,0 @@
-from typing import Self
-
-from pydantic import BaseModel
-
-from exo.shared.types.profiling import MemoryUsage, SystemPerformanceProfile
-from exo.utils.pydantic_ext import TaggedModel
-
-
-class _TempMetrics(BaseModel, extra="ignore"):
-    """Temperature-related metrics returned by macmon."""
-
-    cpu_temp_avg: float
-    gpu_temp_avg: float
-
-
-class _MemoryMetrics(BaseModel, extra="ignore"):
-    """Memory-related metrics returned by macmon."""
-
-    ram_total: int
-    ram_usage: int
-    swap_total: int
-    swap_usage: int
-
-
-class RawMacmonMetrics(BaseModel, extra="ignore"):
-    """Complete set of metrics returned by macmon.
-
-    Unknown fields are ignored for forward-compatibility.
-    """
-
-    timestamp: str  # ignored
-    temp: _TempMetrics
-    memory: _MemoryMetrics
-    ecpu_usage: tuple[int, float]  # freq mhz, usage %
-    pcpu_usage: tuple[int, float]  # freq mhz, usage %
-    gpu_usage: tuple[int, float]  # freq mhz, usage %
-    all_power: float
-    ane_power: float
-    cpu_power: float
-    gpu_power: float
-    gpu_ram_power: float
-    ram_power: float
-    sys_power: float
-
-
-class MacmonMetrics(TaggedModel):
-    system_profile: SystemPerformanceProfile
-    memory: MemoryUsage
-
-    @classmethod
-    def from_raw(cls, raw: RawMacmonMetrics) -> Self:
-        return cls(
-            system_profile=SystemPerformanceProfile(
-                gpu_usage=raw.gpu_usage[1],
-                temp=raw.temp.gpu_temp_avg,
-                sys_power=raw.sys_power,
-                pcpu_usage=raw.pcpu_usage[1],
-                ecpu_usage=raw.ecpu_usage[1],
-            ),
-            memory=MemoryUsage.from_bytes(
-                ram_total=raw.memory.ram_total,
-                ram_available=(raw.memory.ram_total - raw.memory.ram_usage),
-                swap_total=raw.memory.swap_total,
-                swap_available=(raw.memory.swap_total - raw.memory.swap_usage),
-            ),
-        )
-
-    @classmethod
-    def from_raw_json(cls, json: str) -> Self:
-        return cls.from_raw(RawMacmonMetrics.model_validate_json(json))

src/exo/utils/info_gatherer/tests/test_tb_parsing.py

@@ -1,24 +0,0 @@
-import sys
-
-import pytest
-
-from exo.shared.types.thunderbolt import (
-    ThunderboltConnectivity,
-)
-from exo.utils.info_gatherer.info_gatherer import (
-    _gather_iface_map,  # pyright: ignore[reportPrivateUsage]
-)
-
-
-@pytest.mark.anyio
-@pytest.mark.skipif(
-    sys.platform != "darwin", reason="Thunderbolt info can only be gathered on macos"
-)
-async def test_tb_parsing():
-    data = await ThunderboltConnectivity.gather()
-    ifaces = await _gather_iface_map()
-    assert ifaces
-    assert data
-    for datum in data:
-        datum.ident(ifaces)
-        datum.conn()

src/exo/utils/pydantic_ext.py

@@ -19,20 +19,11 @@ class CamelCaseModel(BaseModel):
         alias_generator=to_camel,
         validate_by_name=True,
         extra="forbid",
+        # I want to reenable this ASAP, but it's causing an issue with TaskStatus
         strict=True,
     )
 
 
-class FrozenModel(BaseModel):
-    model_config = ConfigDict(
-        alias_generator=to_camel,
-        validate_by_name=True,
-        extra="forbid",
-        strict=True,
-        frozen=True,
-    )
-
-
 class TaggedModel(CamelCaseModel):
     @model_serializer(mode="wrap")
     def _serialize(self, handler: SerializerFunctionWrapHandler):

src/exo/utils/tests/testing_mp.py

@@ -28,8 +28,9 @@ def bar(send: MpSender[str]):
     send.close()
 
 
+# not async, just want the fail_after
 @pytest.mark.anyio
-async def test_channel_ipc():
+async def test_channel_setup():
     with fail_after(0.5):
         s, r = mp_channel[str]()
         p1 = mp.Process(target=foo, args=(r,))

src/exo/worker/engines/mlx/utils_mlx.py

@@ -2,7 +2,9 @@ import json
 import os
 import resource
 import sys
+import threading
 import time
+from collections.abc import Callable
 from pathlib import Path
 from typing import Any, cast
 
@@ -82,6 +84,45 @@ def get_weights_size(model_shard_meta: ShardMetadata) -> Memory:
     )
 
 
+class ModelLoadingTimeoutError(Exception):
+    pass
+
+
+TimeoutCallback = Callable[[], None]
+
+
+def eval_with_timeout(
+    mlx_item: Any,  # pyright: ignore[reportAny]
+    timeout_seconds: float = 60.0,
+    on_timeout: TimeoutCallback | None = None,
+) -> None:
+    """Evaluate MLX item with a hard timeout.
+
+    If on_timeout callback is provided, it will be called before terminating
+    the process. This allows the runner to send a failure event before exit.
+    """
+    completed = threading.Event()
+
+    def watchdog() -> None:
+        if not completed.wait(timeout=timeout_seconds):
+            logger.error(
+                f"mlx_item evaluation timed out after {timeout_seconds:.0f}s. "
+                "This may indicate an issue with FAST_SYNCH and tensor parallel sharding. "
+                "Terminating process."
+            )
+            if on_timeout is not None:
+                on_timeout()
+            os._exit(1)
+
+    watchdog_thread = threading.Thread(target=watchdog, daemon=True)
+    watchdog_thread.start()
+
+    try:
+        mx.eval(mlx_item)  # pyright: ignore[reportAny]
+    finally:
+        completed.set()
+
+
 def mx_barrier(group: Group | None = None):
     mx.eval(
         mx.distributed.all_sum(
@@ -145,26 +186,20 @@ def mlx_distributed_init(
                 group = mx.distributed.init(backend="ring", strict=True)
 
             case MlxJacclInstance(
-                jaccl_devices=jaccl_devices, jaccl_coordinators=jaccl_coordinators
+                ibv_devices=ibv_devices, jaccl_coordinators=jaccl_coordinators
             ):
-                assert all(
-                    jaccl_devices[i][i] is None for i in range(len(jaccl_devices))
-                )
                 # Use RDMA connectivity matrix
                 coordination_file = (
                     f"./hosts_{bound_instance.instance.instance_id}_{rank}.json"
                 )
-                jaccl_devices_json = json.dumps(jaccl_devices)
+                ibv_devices_json = json.dumps(ibv_devices)
 
                 with open(coordination_file, "w") as f:
-                    _ = f.write(jaccl_devices_json)
+                    _ = f.write(ibv_devices_json)
 
                 jaccl_coordinator = jaccl_coordinators[bound_instance.bound_node_id]
 
-                # TODO: update once upstream fixes
-                logger.info(
-                    f"rank {rank} MLX_IBV_DEVICES: {coordination_file} with devices: {jaccl_devices_json}"
-                )
+                logger.info(f"rank {rank} MLX_IBV_DEVICES: {ibv_devices_json}")
                 logger.info(f"rank {rank} MLX_JACCL_COORDINATOR: {jaccl_coordinator}")
                 os.environ["MLX_IBV_DEVICES"] = coordination_file
                 os.environ["MLX_RANK"] = str(rank)
@@ -194,7 +229,9 @@ def initialize_mlx(
 
 
 def load_mlx_items(
-    bound_instance: BoundInstance, group: Group | None
+    bound_instance: BoundInstance,
+    group: Group | None,
+    on_timeout: TimeoutCallback | None = None,
 ) -> tuple[Model, TokenizerWrapper]:
     if group is None:
         logger.info(f"Single device used for {bound_instance.instance}")
@@ -208,7 +245,9 @@ def load_mlx_items(
     else:
         logger.info("Starting distributed init")
         start_time = time.perf_counter()
-        model, tokenizer = shard_and_load(bound_instance.bound_shard, group=group)
+        model, tokenizer = shard_and_load(
+            bound_instance.bound_shard, group=group, on_timeout=on_timeout
+        )
         end_time = time.perf_counter()
         logger.info(
             f"Time taken to shard and load model: {(end_time - start_time):.2f}s"
@@ -222,6 +261,7 @@ def load_mlx_items(
 def shard_and_load(
     shard_metadata: ShardMetadata,
     group: Group,
+    on_timeout: TimeoutCallback | None = None,
 ) -> tuple[nn.Module, TokenizerWrapper]:
     model_path = build_model_path(shard_metadata.model_meta.model_id)
 
@@ -258,7 +298,15 @@ def shard_and_load(
             logger.info(f"loading model from {model_path} with pipeline parallelism")
             model = pipeline_auto_parallel(model, group, shard_metadata)
 
-    mx.eval(model.parameters())
+    # Estimate timeout based on model size
+    base_timeout = float(os.environ.get("EXO_MODEL_LOAD_TIMEOUT", "60"))
+    model_size_gb = get_weights_size(shard_metadata).in_bytes / (1024**3)
+    timeout_seconds = base_timeout + model_size_gb / 5
+    logger.info(
+        f"Evaluating model parameters with timeout of {timeout_seconds:.0f}s "
+        f"(model size: {model_size_gb:.1f}GB)"
+    )
+    eval_with_timeout(model.parameters(), timeout_seconds, on_timeout)
 
     # TODO: Do we need this?
     mx.eval(model)

src/exo/worker/main.py

@@ -16,7 +16,10 @@ from exo.shared.types.events import (
     ForwarderEvent,
     IndexedEvent,
     NodeDownloadProgress,
-    NodeGatheredInfo,
+    NodeIdentityMeasured,
+    NodeMemoryMeasured,
+    NodeNetworkMeasured,
+    NodeSystemMeasured,
     TaskCreated,
     TaskStatusUpdated,
     TopologyEdgeCreated,
@@ -24,6 +27,11 @@ from exo.shared.types.events import (
 )
 from exo.shared.types.models import ModelId
 from exo.shared.types.multiaddr import Multiaddr
+from exo.shared.types.profiling import (
+    MemoryPerformanceProfile,
+    NetworkInterfaceInfo,
+    SystemPerformanceProfile,
+)
 from exo.shared.types.state import State
 from exo.shared.types.tasks import (
     CreateRunner,
@@ -32,7 +40,7 @@ from exo.shared.types.tasks import (
     Task,
     TaskStatus,
 )
-from exo.shared.types.topology import Connection, SocketConnection
+from exo.shared.types.topology import Connection
 from exo.shared.types.worker.downloads import (
     DownloadCompleted,
     DownloadOngoing,
@@ -43,14 +51,20 @@ from exo.shared.types.worker.runners import RunnerId
 from exo.shared.types.worker.shards import ShardMetadata
 from exo.utils.channels import Receiver, Sender, channel
 from exo.utils.event_buffer import OrderedBuffer
-from exo.utils.info_gatherer.info_gatherer import GatheredInfo, InfoGatherer
-from exo.utils.info_gatherer.net_profile import check_reachable
 from exo.worker.download.download_utils import (
     map_repo_download_progress_to_download_progress_data,
 )
 from exo.worker.download.shard_downloader import RepoDownloadProgress, ShardDownloader
 from exo.worker.plan import plan
 from exo.worker.runner.runner_supervisor import RunnerSupervisor
+from exo.worker.utils import (
+    IdentityMetrics,
+    start_polling_identity_metrics,
+    start_polling_memory_metrics,
+    start_polling_network_metrics,
+    start_polling_system_metrics,
+)
+from exo.worker.utils.net_profile import check_reachable
 
 
 class Worker:
@@ -84,7 +98,7 @@ class Worker:
         self.state: State = State()
         self.download_status: dict[ModelId, DownloadProgress] = {}
         self.runners: dict[RunnerId, RunnerSupervisor] = {}
-        self._tg: TaskGroup = create_task_group()
+        self._tg: TaskGroup | None = None
 
         self._nack_cancel_scope: CancelScope | None = None
         self._nack_attempts: int = 0
@@ -96,13 +110,51 @@ class Worker:
     async def run(self):
         logger.info("Starting Worker")
 
-        info_send, info_recv = channel[GatheredInfo]()
-        info_gatherer: InfoGatherer = InfoGatherer(info_send)
+        async def identity_callback(identity: IdentityMetrics) -> None:
+            await self.event_sender.send(
+                NodeIdentityMeasured(
+                    node_id=self.node_id,
+                    model_id=identity.model_id,
+                    chip_id=identity.chip_id,
+                    friendly_name=identity.friendly_name,
+                    when=str(datetime.now(tz=timezone.utc)),
+                ),
+            )
 
-        async with self._tg as tg:
-            tg.start_soon(info_gatherer.run)
-            tg.start_soon(self._forward_info, info_recv)
+        async def system_callback(system: SystemPerformanceProfile) -> None:
+            await self.event_sender.send(
+                NodeSystemMeasured(
+                    node_id=self.node_id,
+                    system=system,
+                    when=str(datetime.now(tz=timezone.utc)),
+                ),
+            )
+
+        async def network_callback(interfaces: list[NetworkInterfaceInfo]) -> None:
+            await self.event_sender.send(
+                NodeNetworkMeasured(
+                    node_id=self.node_id,
+                    network_interfaces=interfaces,
+                    when=str(datetime.now(tz=timezone.utc)),
+                ),
+            )
+
+        async def memory_callback(memory: MemoryPerformanceProfile) -> None:
+            await self.event_sender.send(
+                NodeMemoryMeasured(
+                    node_id=self.node_id,
+                    memory=memory,
+                    when=str(datetime.now(tz=timezone.utc)),
+                )
+            )
+
+        async with create_task_group() as tg:
+            self._tg = tg
             tg.start_soon(self.plan_step)
+            tg.start_soon(start_polling_identity_metrics, identity_callback)
+            tg.start_soon(start_polling_system_metrics, system_callback)
+            tg.start_soon(start_polling_network_metrics, network_callback)
+            tg.start_soon(start_polling_memory_metrics, memory_callback)
             tg.start_soon(self._emit_existing_download_progress)
             tg.start_soon(self._connection_message_event_writer)
             tg.start_soon(self._resend_out_for_delivery)
@@ -116,17 +168,6 @@ class Worker:
         for runner in self.runners.values():
             runner.shutdown()
 
-    async def _forward_info(self, recv: Receiver[GatheredInfo]):
-        with recv as info_stream:
-            async for info in info_stream:
-                await self.event_sender.send(
-                    NodeGatheredInfo(
-                        node_id=self.node_id,
-                        when=str(datetime.now(tz=timezone.utc)),
-                        info=info,
-                    )
-                )
-
     async def _event_applier(self):
         with self.global_event_receiver as events:
             async for f_event in events:
@@ -146,6 +187,7 @@ class Worker:
                     self._nack_cancel_scope is None
                     or self._nack_cancel_scope.cancel_called
                 ):
+                    assert self._tg
                     # Request the next index.
                     self._tg.start_soon(
                         self._nack_request, self.state.last_event_applied_idx + 1
@@ -236,7 +278,8 @@ class Worker:
                     await self.runners[self._task_to_runner_id(task)].start_task(task)
 
     def shutdown(self):
-        self._tg.cancel_scope.cancel()
+        if self._tg:
+            self._tg.cancel_scope.cancel()
 
     def _task_to_runner_id(self, task: Task):
         instance = self.state.instances[task.instance_id]
@@ -253,28 +296,24 @@ class Worker:
         match msg.connection_type:
             case ConnectionMessageType.Connected:
                 return TopologyEdgeCreated(
-                    conn=Connection(
-                        source=self.node_id,
-                        sink=msg.node_id,
-                        edge=SocketConnection(
-                            sink_multiaddr=Multiaddr(
-                                address=f"/ip4/{msg.remote_ipv4}/tcp/{msg.remote_tcp_port}"
-                            ),
+                    edge=Connection(
+                        local_node_id=self.node_id,
+                        send_back_node_id=msg.node_id,
+                        send_back_multiaddr=Multiaddr(
+                            address=f"/ip4/{msg.remote_ipv4}/tcp/{msg.remote_tcp_port}"
                         ),
-                    ),
+                    )
                 )
 
             case ConnectionMessageType.Disconnected:
                 return TopologyEdgeDeleted(
-                    conn=Connection(
-                        source=self.node_id,
-                        sink=msg.node_id,
-                        edge=SocketConnection(
-                            sink_multiaddr=Multiaddr(
-                                address=f"/ip4/{msg.remote_ipv4}/tcp/{msg.remote_tcp_port}"
-                            ),
+                    edge=Connection(
+                        local_node_id=self.node_id,
+                        send_back_node_id=msg.node_id,
+                        send_back_multiaddr=Multiaddr(
+                            address=f"/ip4/{msg.remote_ipv4}/tcp/{msg.remote_tcp_port}"
                         ),
-                    ),
+                    )
                 )
 
     async def _nack_request(self, since_idx: int) -> None:
@@ -323,6 +362,7 @@ class Worker:
             event_sender=self.event_sender.clone(),
         )
         self.runners[task.bound_instance.bound_runner_id] = runner
+        assert self._tg
         self._tg.start_soon(runner.run)
         return runner
 
@@ -385,6 +425,7 @@ class Worker:
                 last_progress_time = current_time()
 
         self.shard_downloader.on_progress(download_progress_callback)
+        assert self._tg
         self._tg.start_soon(self.shard_downloader.ensure_shard, task.shard_metadata)
 
     async def _forward_events(self) -> None:
@@ -405,14 +446,9 @@ class Worker:
 
     async def _poll_connection_updates(self):
         while True:
-            edges = set(
-                conn.edge for conn in self.state.topology.out_edges(self.node_id)
-            )
-            conns = await check_reachable(
-                self.state.topology,
-                self.node_id,
-                self.state.node_profiles,
-            )
+            # TODO: EdgeDeleted
+            edges = set(self.state.topology.list_connections())
+            conns = await check_reachable(self.state.topology, self.node_id)
             for nid in conns:
                 for ip in conns[nid]:
                     if "127.0.0.1" in ip or "localhost" in ip:
@@ -420,33 +456,26 @@ class Worker:
                             f"Loopback connection should not happen: {ip=} for {nid=}"
                         )
 
-                    edge = SocketConnection(
+                    edge = Connection(
+                        local_node_id=self.node_id,
+                        send_back_node_id=nid,
                         # nonsense multiaddr
-                        sink_multiaddr=Multiaddr(address=f"/ip4/{ip}/tcp/52415")
+                        send_back_multiaddr=Multiaddr(address=f"/ip4/{ip}/tcp/52415")
                         if "." in ip
                         # nonsense multiaddr
                         else Multiaddr(address=f"/ip6/{ip}/tcp/52415"),
                     )
                     if edge not in edges:
                         logger.debug(f"ping discovered {edge=}")
-                        await self.event_sender.send(
-                            TopologyEdgeCreated(
-                                conn=Connection(
-                                    source=self.node_id, sink=nid, edge=edge
-                                )
-                            )
-                        )
+                        await self.event_sender.send(TopologyEdgeCreated(edge=edge))
 
-            for conn in self.state.topology.out_edges(self.node_id):
-                if not isinstance(conn.edge, SocketConnection):
-                    continue
+            for nid, conn in self.state.topology.out_edges(self.node_id):
                 if (
-                    conn.sink not in conns
-                    or conn.edge.sink_multiaddr.ip_address
-                    not in conns.get(conn.source, set())
+                    nid not in conns
+                    or conn.send_back_multiaddr.ip_address not in conns.get(nid, set())
                 ):
                     logger.debug(f"ping failed to discover {conn=}")
-                    await self.event_sender.send(TopologyEdgeDeleted(conn=conn))
+                    await self.event_sender.send(TopologyEdgeDeleted(edge=conn))
 
             await anyio.sleep(10)
 

src/exo/worker/runner/bootstrap.py

@@ -17,15 +17,23 @@ def entrypoint(
     task_receiver: MpReceiver[Task],
     _logger: "loguru.Logger",
 ) -> None:
-    if (
-        isinstance(bound_instance.instance, MlxJacclInstance)
-        and len(bound_instance.instance.jaccl_devices) >= 2
+    fast_synch_override = os.environ.get("EXO_FAST_SYNCH")
+    if fast_synch_override == "on" or (
+        fast_synch_override != "off"
+        and (
+            isinstance(bound_instance.instance, MlxJacclInstance)
+            and len(bound_instance.instance.ibv_devices) >= 2
+        )
     ):
         os.environ["MLX_METAL_FAST_SYNCH"] = "1"
+    else:
+        os.environ["MLX_METAL_FAST_SYNCH"] = "0"
 
     global logger
     logger = _logger
 
+    logger.info(f"Fast synch flag: {os.environ['MLX_METAL_FAST_SYNCH']}")
+
     # Import main after setting global logger - this lets us just import logger from this module
     try:
         from exo.worker.runner.runner import main

src/exo/worker/runner/runner.py

@@ -1,6 +1,8 @@
 import time
 from collections.abc import Generator
+from contextlib import contextmanager
 from functools import cache
+from typing import cast
 
 import mlx.core as mx
 from mlx_lm.models.gpt_oss import Model as GptOssModel
@@ -13,6 +15,7 @@ from openai_harmony import (  # pyright: ignore[reportMissingTypeStubs]
 
 from exo.shared.types.api import ChatCompletionMessageText
 from exo.shared.types.chunks import TokenChunk
+from exo.shared.types.common import CommandId
 from exo.shared.types.events import (
     ChunkGenerated,
     Event,
@@ -20,6 +23,7 @@ from exo.shared.types.events import (
     TaskAcknowledged,
     TaskStatusUpdated,
 )
+from exo.shared.types.models import ModelId
 from exo.shared.types.tasks import (
     ChatCompletion,
     ConnectToGroup,
@@ -48,6 +52,7 @@ from exo.shared.types.worker.runners import (
     RunnerWarmingUp,
 )
 from exo.utils.channels import MpReceiver, MpSender
+from exo.worker.engines.mlx import Model
 from exo.worker.engines.mlx.generator.generate import mlx_generate, warmup_inference
 from exo.worker.engines.mlx.utils_mlx import (
     initialize_mlx,
@@ -57,6 +62,33 @@ from exo.worker.engines.mlx.utils_mlx import (
 from exo.worker.runner.bootstrap import logger
 
 
+@contextmanager
+def send_error_chunk_on_exception(
+    event_sender: MpSender[Event],
+    command_id: CommandId,
+    model_id: ModelId,
+    device_rank: int,
+):
+    try:
+        yield
+    except Exception as e:
+        logger.error(e)
+        if device_rank == 0:
+            event_sender.send(
+                ChunkGenerated(
+                    command_id=command_id,
+                    chunk=TokenChunk(
+                        idx=0,
+                        model=model_id,
+                        text="",
+                        token_id=0,
+                        finish_reason="error",
+                        error_message=str(e),
+                    ),
+                )
+            )
+
+
 def main(
     bound_instance: BoundInstance,
     event_sender: MpSender[Event],
@@ -118,7 +150,20 @@ def main(
                         )
                     )
 
-                    model, tokenizer = load_mlx_items(bound_instance, group)
+                    def on_model_load_timeout() -> None:
+                        event_sender.send(
+                            RunnerStatusUpdated(
+                                runner_id=runner_id,
+                                runner_status=RunnerFailed(
+                                    error_message="Model loading timed out"
+                                ),
+                            )
+                        )
+                        time.sleep(0.5)
+
+                    model, tokenizer = load_mlx_items(
+                        bound_instance, group, on_timeout=on_model_load_timeout
+                    )
 
                     current_status = RunnerLoaded()
                     logger.info("runner loaded")
@@ -135,7 +180,7 @@ def main(
 
                     logger.info(f"warming up inference for instance: {instance}")
                     toks = warmup_inference(
-                        model=model,
+                        model=cast(Model, model),
                         tokenizer=tokenizer,
                         # kv_prefix_cache=kv_prefix_cache,  # supply for warmup-time prefix caching
                     )
@@ -148,8 +193,6 @@ def main(
                 case ChatCompletion(task_params=task_params, command_id=command_id) if (
                     isinstance(current_status, RunnerReady)
                 ):
-                    assert model
-                    assert tokenizer
                     logger.info(f"received chat request: {str(task)[:500]}")
                     current_status = RunnerRunning()
                     logger.info("runner running")
@@ -158,41 +201,47 @@ def main(
                             runner_id=runner_id, runner_status=current_status
                         )
                     )
-                    assert task_params.messages[0].content is not None
-                    _check_for_debug_prompts(task_params.messages[0].content)
+                    with send_error_chunk_on_exception(
+                        event_sender,
+                        command_id,
+                        shard_metadata.model_meta.model_id,
+                        shard_metadata.device_rank,
+                    ):
+                        assert model
+                        assert tokenizer
+                        assert task_params.messages[0].content is not None
+                        _check_for_debug_prompts(task_params.messages[0].content)
 
-                    # Generate responses using the actual MLX generation
-                    mlx_generator = mlx_generate(
-                        model=model,
-                        tokenizer=tokenizer,
-                        task=task_params,
-                    )
+                        # Generate responses using the actual MLX generation
+                        mlx_generator = mlx_generate(
+                            model=cast(Model, model),
+                            tokenizer=tokenizer,
+                            task=task_params,
+                        )
 
-                    # GPT-OSS specific parsing to match other model formats.
-                    if isinstance(model, GptOssModel):
-                        mlx_generator = parse_gpt_oss(mlx_generator)
+                        # GPT-OSS specific parsing to match other model formats.
+                        if isinstance(model, GptOssModel):
+                            mlx_generator = parse_gpt_oss(mlx_generator)
 
-                    # TODO: Add tool call parser here
+                        # TODO: Add tool call parser here
 
-                    for response in mlx_generator:
-                        match response:
-                            case GenerationResponse():
-                                if shard_metadata.device_rank == 0:
-                                    event_sender.send(
-                                        ChunkGenerated(
-                                            command_id=command_id,
-                                            chunk=TokenChunk(
-                                                idx=response.token,
-                                                model=shard_metadata.model_meta.model_id,
-                                                text=response.text,
-                                                token_id=response.token,
-                                                finish_reason=response.finish_reason,
-                                                stats=response.stats,
-                                            ),
+                        for response in mlx_generator:
+                            match response:
+                                case GenerationResponse():
+                                    if shard_metadata.device_rank == 0:
+                                        event_sender.send(
+                                            ChunkGenerated(
+                                                command_id=command_id,
+                                                chunk=TokenChunk(
+                                                    idx=response.token,
+                                                    model=shard_metadata.model_meta.model_id,
+                                                    text=response.text,
+                                                    token_id=response.token,
+                                                    finish_reason=response.finish_reason,
+                                                    stats=response.stats,
+                                                ),
+                                            )
                                         )
-                                    )
-                                # case TokenizedResponse():
-                                # TODO: something here ig
 
                     current_status = RunnerReady()
                     logger.info("runner ready")

src/exo/worker/tests/unittests/test_runner/test_error_handling.py

@@ -0,0 +1,50 @@
+# pyright: reportAny=false
+from unittest.mock import MagicMock
+
+from exo.shared.types.chunks import TokenChunk
+from exo.shared.types.common import CommandId
+from exo.shared.types.events import ChunkGenerated
+from exo.worker.runner.runner import send_error_chunk_on_exception
+from exo.worker.tests.constants import MODEL_A_ID
+
+
+def test_send_error_chunk_on_exception_no_error() -> None:
+    event_sender = MagicMock()
+    command_id = CommandId()
+
+    with send_error_chunk_on_exception(
+        event_sender, command_id, MODEL_A_ID, device_rank=0
+    ):
+        _ = 1 + 1
+
+    event_sender.send.assert_not_called()
+
+
+def test_send_error_chunk_on_exception_catches_error() -> None:
+    event_sender = MagicMock()
+    command_id = CommandId()
+
+    with send_error_chunk_on_exception(
+        event_sender, command_id, MODEL_A_ID, device_rank=0
+    ):
+        raise ValueError("test error")
+
+    event_sender.send.assert_called_once()
+    call_args = event_sender.send.call_args[0][0]
+    assert isinstance(call_args, ChunkGenerated)
+    assert call_args.command_id == command_id
+    assert isinstance(call_args.chunk, TokenChunk)
+    assert call_args.chunk.finish_reason == "error"
+    assert call_args.chunk.error_message == "test error"
+
+
+def test_send_error_chunk_on_exception_skips_non_rank_zero() -> None:
+    event_sender = MagicMock()
+    command_id = CommandId()
+
+    with send_error_chunk_on_exception(
+        event_sender, command_id, MODEL_A_ID, device_rank=1
+    ):
+        raise ValueError("test error")
+
+    event_sender.send.assert_not_called()

src/exo/worker/utils/__init__.py

@@ -0,0 +1,15 @@
+from .profile import (
+    IdentityMetrics,
+    start_polling_identity_metrics,
+    start_polling_memory_metrics,
+    start_polling_network_metrics,
+    start_polling_system_metrics,
+)
+
+__all__ = [
+    "IdentityMetrics",
+    "start_polling_identity_metrics",
+    "start_polling_memory_metrics",
+    "start_polling_network_metrics",
+    "start_polling_system_metrics",
+]

src/exo/worker/utils/macmon.py

@@ -0,0 +1,103 @@
+import platform
+import shutil
+from subprocess import CalledProcessError
+from typing import cast
+
+from anyio import run_process
+from pydantic import BaseModel, ConfigDict, ValidationError
+
+
+class MacMonError(Exception):
+    """Exception raised for errors in the MacMon functions."""
+
+
+def _get_binary_path() -> str:
+    """
+    Get the path to the macmon binary.
+
+    Raises:
+        MacMonError: If the binary doesn't exist or can't be made executable.
+    """
+    # Check for macOS with ARM chip
+    system = platform.system().lower()
+    machine = platform.machine().lower()
+
+    if system != "darwin" or not (
+        "arm" in machine or "m1" in machine or "m2" in machine
+    ):
+        raise MacMonError("MacMon only supports macOS with Apple Silicon (ARM) chips")
+
+    path = shutil.which("macmon")
+
+    if path is None:
+        raise MacMonError("MacMon not found in PATH")
+
+    return path
+
+
+class TempMetrics(BaseModel):
+    """Temperature-related metrics returned by macmon."""
+
+    cpu_temp_avg: float
+    gpu_temp_avg: float
+
+    model_config = ConfigDict(extra="ignore")
+
+
+class Metrics(BaseModel):
+    """Complete set of metrics returned by macmon.
+
+    Unknown fields are ignored for forward-compatibility.
+    """
+
+    all_power: float
+    ane_power: float
+    cpu_power: float
+    ecpu_usage: tuple[int, float]
+    gpu_power: float
+    gpu_ram_power: float
+    gpu_usage: tuple[int, float]
+    pcpu_usage: tuple[int, float]
+    ram_power: float
+    sys_power: float
+    temp: TempMetrics
+    timestamp: str
+
+    model_config = ConfigDict(extra="ignore")
+
+
+async def get_metrics_async() -> Metrics:
+    """
+    Asynchronously run the binary and return the metrics as a Python dictionary.
+
+    Args:
+        binary_path: Optional path to the binary. If not provided, will use the bundled binary.
+
+    Returns:
+        A mapping containing system metrics.
+
+    Raises:
+        MacMonError: If there's an error running the binary.
+    """
+    path = _get_binary_path()
+
+    try:
+        # TODO: Keep Macmon running in the background?
+        result = await run_process([path, "pipe", "-s", "1"])
+
+        return Metrics.model_validate_json(result.stdout.decode().strip())
+
+    except ValidationError as e:
+        raise MacMonError(f"Error parsing JSON output: {e}") from e
+    except CalledProcessError as e:
+        stderr_msg = "no stderr"
+        stderr_output = cast(bytes | str | None, e.stderr)
+        if stderr_output is not None:
+            stderr_msg = (
+                stderr_output.decode()
+                if isinstance(stderr_output, bytes)
+                else str(stderr_output)
+            )
+        raise MacMonError(
+            f"MacMon failed with return code {e.returncode}: {stderr_msg}"
+        ) from e

src/exo/worker/utils/net_profile.py

@@ -1,5 +1,3 @@
-from collections.abc import Mapping
-
 import anyio
 import httpx
 from anyio import create_task_group
@@ -7,7 +5,6 @@ from loguru import logger
 
 from exo.shared.topology import Topology
 from exo.shared.types.common import NodeId
-from exo.shared.types.profiling import NodePerformanceProfile
 
 REACHABILITY_ATTEMPTS = 3
 
@@ -76,9 +73,7 @@ async def check_reachability(
 
 
 async def check_reachable(
-    topology: Topology,
-    self_node_id: NodeId,
-    node_profiles: Mapping[NodeId, NodePerformanceProfile],
+    topology: Topology, self_node_id: NodeId
 ) -> dict[NodeId, set[str]]:
     """Check which nodes are reachable and return their IPs."""
 
@@ -96,16 +91,16 @@ async def check_reachable(
         httpx.AsyncClient(timeout=timeout, limits=limits) as client,
         create_task_group() as tg,
     ):
-        for node_id in topology.list_nodes():
-            if node_id not in node_profiles:
+        for node in topology.list_nodes():
+            if not node.node_profile:
                 continue
-            if node_id == self_node_id:
+            if node.node_id == self_node_id:
                 continue
-            for iface in node_profiles[node_id].network_interfaces:
+            for iface in node.node_profile.network_interfaces:
                 tg.start_soon(
                     check_reachability,
                     iface.ip_address,
-                    node_id,
+                    node.node_id,
                     reachable,
                     client,
                 )

src/exo/worker/utils/profile.py

@@ -0,0 +1,147 @@
+import asyncio
+import os
+import platform
+from dataclasses import dataclass
+from typing import Any, Callable, Coroutine
+
+import anyio
+from loguru import logger
+
+from exo.shared.types.memory import Memory
+from exo.shared.types.profiling import (
+    MemoryPerformanceProfile,
+    NetworkInterfaceInfo,
+    SystemPerformanceProfile,
+)
+
+from .macmon import (
+    MacMonError,
+    Metrics,
+)
+from .macmon import (
+    get_metrics_async as macmon_get_metrics_async,
+)
+from .system_info import (
+    get_friendly_name,
+    get_model_and_chip,
+    get_network_interfaces,
+)
+
+
+@dataclass(frozen=True)
+class IdentityMetrics:
+    model_id: str
+    chip_id: str
+    friendly_name: str
+
+
+async def get_metrics_async() -> Metrics | None:
+    """Return detailed Metrics on macOS or a minimal fallback elsewhere."""
+
+    if platform.system().lower() == "darwin":
+        return await macmon_get_metrics_async()
+
+
+def get_memory_profile() -> MemoryPerformanceProfile:
+    """Construct a MemoryPerformanceProfile using psutil"""
+    override_memory_env = os.getenv("OVERRIDE_MEMORY_MB")
+    override_memory: int | None = (
+        Memory.from_mb(int(override_memory_env)).in_bytes
+        if override_memory_env
+        else None
+    )
+
+    return MemoryPerformanceProfile.from_psutil(override_memory=override_memory)
+
+
+async def start_polling_memory_metrics(
+    callback: Callable[[MemoryPerformanceProfile], Coroutine[Any, Any, None]],
+    *,
+    poll_interval_s: float = 0.5,
+) -> None:
+    """Continuously poll and emit memory-only metrics at a faster cadence.
+
+    Parameters
+    - callback: coroutine called with a fresh MemoryPerformanceProfile each tick
+    - poll_interval_s: interval between polls
+    """
+    while True:
+        try:
+            mem = get_memory_profile()
+            await callback(mem)
+        except MacMonError as e:
+            logger.opt(exception=e).error("Memory Monitor encountered error")
+        finally:
+            await anyio.sleep(poll_interval_s)
+
+
+async def start_polling_identity_metrics(
+    callback: Callable[[IdentityMetrics], Coroutine[Any, Any, None]],
+    *,
+    poll_interval_s: float = 30.0,
+) -> None:
+    """Continuously poll and emit identity metrics at 30s intervals."""
+    while True:
+        try:
+            model_id, chip_id = await get_model_and_chip()
+            friendly_name = await get_friendly_name()
+            await callback(
+                IdentityMetrics(
+                    model_id=model_id,
+                    chip_id=chip_id,
+                    friendly_name=friendly_name,
+                )
+            )
+        except Exception as e:
+            logger.opt(exception=e).error("Failed to emit identity metrics")
+        finally:
+            await anyio.sleep(poll_interval_s)
+
+
+async def start_polling_system_metrics(
+    callback: Callable[[SystemPerformanceProfile], Coroutine[Any, Any, None]],
+    *,
+    poll_interval_s: float = 1.0,
+) -> None:
+    """Continuously poll and emit system metrics (GPU, temp, power) at 1s intervals."""
+    while True:
+        try:
+            metrics = await get_metrics_async()
+            if metrics is None:
+                return
+
+            await callback(
+                SystemPerformanceProfile(
+                    gpu_usage=metrics.gpu_usage[1],
+                    temp=metrics.temp.gpu_temp_avg,
+                    sys_power=metrics.sys_power,
+                    pcpu_usage=metrics.pcpu_usage[1],
+                    ecpu_usage=metrics.ecpu_usage[1],
+                    ane_power=metrics.ane_power,
+                )
+            )
+        except asyncio.TimeoutError:
+            logger.warning(
+                "[system_monitor] Operation timed out after 30s, skipping this cycle."
+            )
+        except MacMonError as e:
+            logger.opt(exception=e).error("System Monitor encountered error")
+            return
+        finally:
+            await anyio.sleep(poll_interval_s)
+
+
+async def start_polling_network_metrics(
+    callback: Callable[[list[NetworkInterfaceInfo]], Coroutine[Any, Any, None]],
+    *,
+    poll_interval_s: float = 30.0,
+) -> None:
+    """Continuously poll and emit network interface info at 30s intervals."""
+    while True:
+        try:
+            network_interfaces = get_network_interfaces()
+            await callback(network_interfaces)
+        except Exception as e:
+            logger.opt(exception=e).error("Network Monitor encountered error")
+        finally:
+            await anyio.sleep(poll_interval_s)

src/exo/worker/utils/tests/test_macmon.py

@@ -0,0 +1,77 @@
+"""Tests for macmon error handling.
+
+These tests verify that MacMon errors are handled gracefully without
+crashing the application or spamming logs.
+"""
+
+import platform
+from subprocess import CalledProcessError
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from exo.worker.utils.macmon import MacMonError, get_metrics_async
+
+
+@pytest.mark.skipif(
+    platform.system().lower() != "darwin" or "arm" not in platform.machine().lower(),
+    reason="MacMon only supports macOS with Apple Silicon",
+)
+class TestMacMonErrorHandling:
+    """Test MacMon error handling."""
+
+    async def test_called_process_error_wrapped_as_macmon_error(self) -> None:
+        """CalledProcessError should be wrapped as MacMonError."""
+        mock_error = CalledProcessError(
+            returncode=1,
+            cmd=["macmon", "pipe", "-s", "1"],
+            stderr=b"some error message",
+        )
+
+        with (
+            patch(
+                "exo.worker.utils.macmon.shutil.which", return_value="/usr/bin/macmon"
+            ),
+            patch(
+                "exo.worker.utils.macmon.run_process", new_callable=AsyncMock
+            ) as mock_run,
+        ):
+            mock_run.side_effect = mock_error
+
+            with pytest.raises(MacMonError) as exc_info:
+                await get_metrics_async()
+
+            assert "MacMon failed with return code 1" in str(exc_info.value)
+            assert "some error message" in str(exc_info.value)
+
+    async def test_called_process_error_with_no_stderr(self) -> None:
+        """CalledProcessError with no stderr should be handled gracefully."""
+        mock_error = CalledProcessError(
+            returncode=1,
+            cmd=["macmon", "pipe", "-s", "1"],
+            stderr=None,
+        )
+
+        with (
+            patch(
+                "exo.worker.utils.macmon.shutil.which", return_value="/usr/bin/macmon"
+            ),
+            patch(
+                "exo.worker.utils.macmon.run_process", new_callable=AsyncMock
+            ) as mock_run,
+        ):
+            mock_run.side_effect = mock_error
+
+            with pytest.raises(MacMonError) as exc_info:
+                await get_metrics_async()
+
+            assert "MacMon failed with return code 1" in str(exc_info.value)
+            assert "no stderr" in str(exc_info.value)
+
+    async def test_macmon_not_found_raises_macmon_error(self) -> None:
+        """When macmon is not found in PATH, MacMonError should be raised."""
+        with patch("exo.worker.utils.macmon.shutil.which", return_value=None):
+            with pytest.raises(MacMonError) as exc_info:
+                await get_metrics_async()
+
+            assert "MacMon not found in PATH" in str(exc_info.value)

tests/headless_runner.py

@@ -34,8 +34,7 @@ from exo.shared.types.worker.instances import (
 )
 from exo.shared.types.worker.runners import RunnerId, ShardAssignments
 from exo.shared.types.worker.shards import PipelineShardMetadata, TensorShardMetadata
-from exo.utils.channels import MpReceiver, MpSender, channel, mp_channel
-from exo.utils.info_gatherer.info_gatherer import GatheredInfo, InfoGatherer
+from exo.utils.channels import MpReceiver, MpSender, mp_channel
 from exo.worker.download.impl_shard_downloader import (
     build_full_shard,
     exo_shard_downloader,
@@ -66,7 +65,6 @@ async def main():
     app = FastAPI()
     app.post("/ring")(ring_backend)
     app.post("/jaccl")(jaccl_backend)
-    app.post("/tb_detection")(tb_detection)
     shutdown = anyio.Event()
     await serve(
         app,  # type: ignore
@@ -78,15 +76,6 @@ async def main():
     shutdown.set()
 
 
-async def tb_detection():
-    send, recv = channel[GatheredInfo]()
-    ig = InfoGatherer(send)
-    with anyio.move_on_after(1):
-        await ig._monitor_system_profiler()  # pyright: ignore[reportPrivateUsage]
-    with recv:
-        return recv.collect()
-
-
 async def assert_downloads():
     sd = exo_shard_downloader()
     # await sd.ensure_shard(await build_full_shard(MODEL_CARDS["qwen3-0.6b"].model_id))
@@ -220,16 +209,16 @@ async def jaccl_backend(test: Tests):
             break
     else:
         raise ValueError(f"{weird_hn} not in {test.devs}")
-    return await execute_test(test, jaccl_instance(test, iid), hn)
+    return await execute_test(test, jaccl_instance(test, iid, hn), hn)
 
 
-def jaccl_instance(test: Tests, iid: InstanceId):
+def jaccl_instance(test: Tests, iid: InstanceId, hn: str):
     meta = MODEL_CARDS[test.model_id].metadata
     world_size = len(test.devs)
 
     return MlxJacclInstance(
         instance_id=iid,
-        jaccl_devices=[[None, "rdma_en3"], ["rdma_en3", None]],
+        ibv_devices=[[None, "rdma_en3"], ["rdma_en3", None]],
         # rank 0 is always coordinator
         jaccl_coordinators={
             NodeId(host[0]): test.devs[0][1] + ":52416" for host in test.devs