Files
LocalAI/core/services/messaging/subjects.go
Ettore Di Giacinto 6b63b47f61 feat(distributed): support multiple replicas of one model on the same node (#9583)
* feat(distributed): support multiple replicas of one model on the same node

The distributed scheduler implicitly assumed `(node_id, model_name)` was
unique, but the schema didn't enforce it and the worker keyed all gRPC
processes by model name alone. With `MinReplicas=2` against a single
worker, the reconciler "scaled up" every 30s but the registry never
advanced past 1 row — the worker re-loaded the model in-place every tick
until VRAM fragmented and the gRPC process died.

This change introduces multi-replica-per-node as a first-class concept,
with capacity-aware scheduling, a circuit breaker, and VRAM
soft-reservation. Operators can declare per-node capacity via the worker
flag `--max-replicas-per-model` (mirrored as auto-label
`node.replica-slots=N`) or override per-node from the UI.

* Schema: BackendNode gains MaxReplicasPerModel (default 1) and
  ReservedVRAM. NodeModel gains ReplicaIndex (composite with node_id +
  model_name). ModelSchedulingConfig gains UnsatisfiableUntil/Ticks for
  the reconciler circuit breaker.

* Registry: replica_index threaded through SetNodeModel, RemoveNodeModel,
  IncrementInFlight, DecrementInFlight, TouchNodeModel, GetNodeModel,
  SetNodeModelLoadInfo and the InFlightTrackingClient. New helpers:
  CountReplicasOnNode, NextFreeReplicaIndex (with ErrNoFreeSlot),
  RemoveAllNodeModelReplicas, FindNodesWithFreeSlot,
  ClusterCapacityForModel, ReserveVRAM/ReleaseVRAM (atomic UPDATE with
  ErrInsufficientVRAM), and the unsatisfiable-flag CRUD.

* Worker: processKey now `<modelID>#<replicaIndex>` so concurrent loads
  of the same model land on distinct ports. Adds CLI flag
  --max-replicas-per-model (env LOCALAI_MAX_REPLICAS_PER_MODEL, default 1)
  and emits the auto-label.

* Router: scheduleNewModel filters candidates by free slot, allocates the
  replica index, and soft-reserves VRAM before installing the backend.
  evictLRUAndFreeNode now deletes the targeted row by ID instead of all
  replicas of the model on the node — fixes a latent bug where evicting
  one replica orphaned its siblings.

* Reconciler: caps scale-up at ClusterCapacityForModel so a misconfig
  (MinReplicas > capacity) doesn't loop forever. After 3 consecutive
  ticks of capacity==0 it sets UnsatisfiableUntil for a 5m cooldown and
  emits a warning. ClearAllUnsatisfiable fires from Register,
  ApproveNode, SetNodeLabel(s), RemoveNodeLabel and
  UpdateMaxReplicasPerModel so a new node joining or label changes wake
  the reconciler immediately. scaleDownIdle removes highest-replica-index
  first to keep slots compact.

* Heartbeat resets reserved_vram to 0 — worker is the source of truth
  for actual free VRAM; the reservation is only for the in-tick race
  window between two scheduling decisions.

* Probe path (reconciler.probeLoadedModels and health.doCheckAll) now
  pass the row's replica_index to RemoveNodeModel so an unreachable
  replica doesn't orphan healthy siblings.

* Admin override: PUT /api/nodes/:id/max-replicas-per-model sets a
  sticky override (preserved across worker re-registration). DELETE
  clears the override so the worker's flag applies again on next
  register. Required because Kong defaults the worker flag to 1, so
  every worker restart would have silently reverted the UI value.

* React UI: always-visible slot badge on the node row (muted at default
  1, accented when >1); inline editor in the expanded drawer with
  pencil-to-edit, Save/Cancel, Esc/Enter, "(override)" indicator when
  the value is admin-set, and a "Reset" button to hand control back to
  the worker. Soft confirm when shrinking the cap below the count of
  loaded replicas. Scheduling rules table gets an "Unsatisfiable until
  HH:MM" status badge surfacing the cooldown.

* node.replica-slots filtered out of the labels strip on the row to
  avoid duplicating the slot badge.

23 new Ginkgo specs (registry, reconciler, inflight, health) cover:
multi-replica row independence, RemoveNodeModel of one replica
preserving siblings, NextFreeReplicaIndex slot allocation including
ErrNoFreeSlot, capacity-gated scale-up with circuit breaker tripping
and recovery on Register, scheduleDownIdle ordering, ClusterCapacity
math, ReserveVRAM admission gating, Heartbeat reset, override survival
across worker re-registration, and ResetMaxReplicasPerModel handing
control back. Plus 8 stdlib tests for the worker processKey / CLI /
auto-label.

Closes the flap reproduced on Qwen3.6-35B against the nvidia-thor
worker (single 128 GiB node, MinReplicas=2): the reconciler now caps
the scale-up at the cluster's actual capacity instead of looping.

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: claude-code:opus-4-7 [Read] [Edit] [Bash] [Skill:critique] [Skill:audit] [Skill:polish] [Skill:golang-testing]

* refactor(react-ui/nodes): tighten capacity editor copy + adopt ActionMenu for row actions

* Capacity editor hint trimmed from operator-doc-style ("Sourced from
  the worker's `--max-replicas-per-model` flag. Changing it here makes it
  a sticky admin override that survives worker restarts." → "Saved
  values stick across worker restarts.") and the override-state copy
  similarly compressed. The full mechanic is no longer needed in the UI
  — the override pill carries the meaning and the docs cover the rest.

* Node row actions migrated from an inline cluster of icon buttons
  (Drain / Resume / Trash) to the kebab ActionMenu used by /manage for
  per-row model actions, so dense Nodes tables stay clean. Approve
  stays as a prominent primary button — it's a stateful admission gate,
  not a routine action, and elevating it matches how /manage surfaces
  install-time decisions outside the menu.

* The expanded drawer's Labels section now filters node.replica-slots
  out of the editable label list. The label is owned by the Capacity
  editor above; surfacing it again as an editable label invited
  confusion (the Capacity save would clobber any direct edit).

Both backend and agent workers benefit — they share the row rendering
path, so the action menu and label filter apply to both.

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: claude-code:opus-4-7 [Edit] [chrome-devtools-mcp] [Skill:critique] [Skill:audit] [Skill:polish]

* fix(react-ui/nodes): suppress slot badge on agent workers

Agent workers don't load models, so the per-node replica capacity is
inapplicable to them. Showing "1× slots" on agent rows was a tiny
inconsistency from the unified rendering path — gate the badge on
node_type !== 'agent' so it only appears on backend workers.

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: claude-code:opus-4-7 [Edit] [chrome-devtools-mcp]

* refactor(react-ui/nodes): distill expanded drawer + restyle scheduling form

The expanded node drawer used to stack five panels — slot badge,
filled capacity box, Loaded Models h4+empty-state, Installed Backends
h4+empty-state, Labels h4+chips+form — making routine inspections feel
like a control panel. The scheduling rule form wrapped its mode toggle
as two 50%-width filled buttons that competed visually with the actual
primary action.

* Drawer: collapse three rarely-touched config zones (Capacity,
  Backends, Labels) into one `<details>` "Manage" disclosure (closed by
  default) with small uppercase eyebrow labels for each zone instead of
  parallel h4 sub-headings. Loaded Models stays as the at-a-glance
  headline with a single-line empty hint instead of a boxed empty state.
  CapacityEditor renders flat (no filled background) — the Manage
  disclosure provides framing.

* Scheduling form: replace the chunky 50%-width button-tabs with the
  project's existing `.segmented` control (icon + label, sized to
  content). Mode hint becomes a single tied line below. Fields stack
  vertically with helper text under inputs and a hairline divider above
  the right-aligned Save / Cancel.

The empty drawer collapses from ~5 stacked sections (~280px tall) to
two lines (~80px). The scheduling form now reads as a designed dialog
instead of raw building blocks. Both surfaces now match the typographic
density and weight of the rest of the admin pages.

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: claude-code:opus-4-7 [Edit] [chrome-devtools-mcp] [Skill:distill] [Skill:audit] [Skill:polish]

* feat(react-ui/nodes): replace scheduling form's model picker with searchable combobox

The native <select> made operators scroll through every gallery entry to
find a model name. The project already has SearchableModelSelect (used
in Studio/Talk/etc.) which combines free-text search with the gallery
list and accepts typed model names that aren't installed yet — useful
for pre-staging a scheduling rule before the node it'll run on has
finished bootstrapping.

Also drops the now-unused useModels import (the combobox manages the
gallery hook internally).

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: claude-code:opus-4-7 [Edit]

* refactor(react-ui/nodes): consolidate key/value chip editor + add replica preset chips

The Nodes page was rendering the same key=value chip pattern in two
places with subtly different markup: the Labels editor in the expanded
drawer and (post-distill) the Node Selector input in the scheduling
form. The form's input was also a comma-separated string that operators
were getting wrong.

* Extract <KeyValueChips> as a fully controlled chip-builder. Parent
  owns the map and decides what onAdd/onRemove does — form state for the
  scheduling form, API calls for the live drawer Labels editor. Same
  visuals everywhere; one component to change when polish needs apply.

* Replace the comma-separated Node Selector text input with KeyValueChips.
  Operators were copying syntax from docs and missing commas; the chip
  vocabulary makes the key=value structure self-documenting.

* Add <ReplicaInput>: numeric input + quick-pick preset chips for Min/Max
  replicas. Picked over a slider because replica counts are exact specs
  derived from VRAM math (operator decision, not a fuzzy estimate). The
  chips give one-click access to common values (1/2/3/4 for Min,
  0=no-limit/2/4/8 for Max) without the slider's special-value problem
  (MaxReplicas=0 is categorical, not a position on a continuum).

* Drop the now-unused labelInputs state in the Nodes page (the inline
  label editor's per-node draft state lived there and is now owned by
  KeyValueChips).

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: claude-code:opus-4-7 [Edit] [Skill:distill]

* test: fix CI fallout from multi-replica refactor (e2e/distributed + playwright)

Two breakages caught by CI that didn't surface in the local run:

* tests/e2e/distributed/*.go — multiple files used the pre-PR2 registry
  signatures for SetNodeModel / IncrementInFlight / DecrementInFlight /
  RemoveNodeModel / TouchNodeModel / GetNodeModel / SetNodeModelLoadInfo
  and one stale adapter.InstallBackend call in node_lifecycle_test.go.
  All updated to pass replicaIndex=0 — these tests don't exercise
  multi-replica behavior, they just need to compile against the new
  signatures. The chip-builder tests in core/services/nodes/ already
  cover the multi-replica logic.

* core/http/react-ui/e2e/nodes-per-node-backend-actions.spec.js — the
  drawer's distill refactor moved Backends inside a "Manage" <details>
  disclosure that's collapsed by default. The test helper expanded the
  node row but never opened Manage, so the per-node backend table was
  never in the DOM. Helper now clicks `.node-manage > summary` after
  expanding the row.

All 100 playwright tests pass locally; tests/e2e/distributed compiles
clean.

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: claude-code:opus-4-7 [Edit] [Bash]

---------

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2026-04-27 21:20:05 +02:00

282 lines
11 KiB
Go

package messaging
import "strings"
// sanitizeSubjectToken replaces NATS-reserved characters in a subject token.
// NATS uses '.' as hierarchy delimiter and '*'/'>' as wildcards.
func sanitizeSubjectToken(s string) string {
r := strings.NewReplacer(".", "-", "*", "-", ">", "-", " ", "-", "\t", "-", "\n", "-")
return r.Replace(s)
}
// NATS subject constants for the distributed architecture.
// Following the notetaker pattern: <entity>.<action>
// Job Distribution (Queue Groups — load-balanced, one consumer gets each message)
const (
SubjectJobsNew = "jobs.new"
SubjectMCPCIJobsNew = "jobs.mcp-ci.new"
SubjectAgentExecute = "agent.execute"
QueueWorkers = "workers"
)
// Status Updates (Pub/Sub — all subscribers get every message, for SSE bridging)
// These use parameterized subjects: e.g. SubjectAgentEvents("myagent", "user1")
const (
subjectAgentEventsPrefix = "agent."
subjectJobProgressPrefix = "jobs."
subjectFineTunePrefix = "finetune."
subjectGalleryPrefix = "gallery."
)
// SubjectAgentEvents returns the NATS subject for agent SSE events.
func SubjectAgentEvents(agentName, userID string) string {
if userID == "" {
userID = "anonymous"
}
return subjectAgentEventsPrefix + sanitizeSubjectToken(agentName) + ".events." + sanitizeSubjectToken(userID)
}
// SubjectJobProgress returns the NATS subject for job progress updates.
func SubjectJobProgress(jobID string) string {
return subjectJobProgressPrefix + sanitizeSubjectToken(jobID) + ".progress"
}
// SubjectJobResult returns the NATS subject for the final job result (terminal state).
func SubjectJobResult(jobID string) string {
return subjectJobProgressPrefix + sanitizeSubjectToken(jobID) + ".result"
}
// MCP Tool Execution (Request-Reply via NATS — load-balanced across agent workers)
const (
SubjectMCPToolExecute = "mcp.tools.execute"
SubjectMCPDiscovery = "mcp.discovery"
QueueAgentWorkers = "agent-workers"
)
// SubjectFineTuneProgress returns the NATS subject for fine-tune progress.
func SubjectFineTuneProgress(jobID string) string {
return subjectFineTunePrefix + sanitizeSubjectToken(jobID) + ".progress"
}
// SubjectGalleryProgress returns the NATS subject for gallery download progress.
func SubjectGalleryProgress(opID string) string {
return subjectGalleryPrefix + sanitizeSubjectToken(opID) + ".progress"
}
// Control Signals (Pub/Sub — targeted cancellation)
const (
subjectJobCancelPrefix = "jobs."
subjectAgentCancelPrefix = "agent."
subjectFineTuneCancelPrefix = "finetune."
subjectGalleryCancelPrefix = "gallery."
)
// Wildcard subjects for NATS subscriptions that match all IDs.
const (
SubjectJobCancelWildcard = "jobs.*.cancel"
SubjectJobResultWildcard = "jobs.*.result"
SubjectJobProgressWildcard = "jobs.*.progress"
SubjectAgentCancelWildcard = "agent.*.cancel"
SubjectGalleryCancelWildcard = "gallery.*.cancel"
SubjectGalleryProgressWildcard = "gallery.*.progress"
)
// SubjectJobCancel returns the NATS subject to cancel a running job.
func SubjectJobCancel(jobID string) string {
return subjectJobCancelPrefix + sanitizeSubjectToken(jobID) + ".cancel"
}
// SubjectAgentCancel returns the NATS subject to cancel agent execution.
func SubjectAgentCancel(agentID string) string {
return subjectAgentCancelPrefix + sanitizeSubjectToken(agentID) + ".cancel"
}
// SubjectFineTuneCancel returns the NATS subject to stop fine-tuning.
func SubjectFineTuneCancel(jobID string) string {
return subjectFineTuneCancelPrefix + sanitizeSubjectToken(jobID) + ".cancel"
}
// SubjectGalleryCancel returns the NATS subject to cancel a gallery download.
func SubjectGalleryCancel(opID string) string {
return subjectGalleryCancelPrefix + sanitizeSubjectToken(opID) + ".cancel"
}
// Node Backend Lifecycle (Pub/Sub — targeted to specific nodes)
//
// These subjects control the backend *process* lifecycle on a serve-backend node,
// mirroring how the local ModelLoader uses startProcess() / deleteProcess().
//
// Model loading (LoadModel gRPC) is done via direct gRPC calls to the node's
// address — no NATS needed for that, same as local mode.
const (
subjectNodePrefix = "nodes."
)
// SubjectNodeBackendInstall tells a worker node to install a backend and start its gRPC process.
// Uses NATS request-reply: the SmartRouter sends the request, the worker installs
// the backend from gallery (if not already installed), starts the gRPC process,
// and replies when ready.
func SubjectNodeBackendInstall(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".backend.install"
}
// BackendInstallRequest is the payload for a backend.install NATS request.
type BackendInstallRequest struct {
Backend string `json:"backend"`
ModelID string `json:"model_id,omitempty"`
BackendGalleries string `json:"backend_galleries,omitempty"`
// URI is set for external installs (OCI image, URL, or path). When non-empty
// the worker routes to InstallExternalBackend instead of the gallery lookup.
URI string `json:"uri,omitempty"`
Name string `json:"name,omitempty"`
Alias string `json:"alias,omitempty"`
// ReplicaIndex selects which slot on the worker this load occupies, so two
// concurrent backend.install requests for the same model land on distinct
// gRPC processes and ports. Workers older than this field treat it as 0
// (single-replica behavior — no collision because the controller never
// asks for replica > 0 on a node whose MaxReplicasPerModel is 1).
ReplicaIndex int32 `json:"replica_index,omitempty"`
}
// BackendInstallReply is the response from a backend.install NATS request.
type BackendInstallReply struct {
Success bool `json:"success"`
Address string `json:"address,omitempty"` // gRPC address of the backend process (host:port)
Error string `json:"error,omitempty"`
}
// SubjectNodeBackendList queries a worker node for its installed backends.
// Uses NATS request-reply.
func SubjectNodeBackendList(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".backend.list"
}
// BackendListRequest is the payload for a backend.list NATS request.
type BackendListRequest struct{}
// BackendListReply is the response from a backend.list NATS request.
type BackendListReply struct {
Backends []NodeBackendInfo `json:"backends"`
Error string `json:"error,omitempty"`
}
// NodeBackendInfo describes a backend installed on a worker node.
type NodeBackendInfo struct {
Name string `json:"name"`
IsSystem bool `json:"is_system"`
IsMeta bool `json:"is_meta"`
InstalledAt string `json:"installed_at,omitempty"`
GalleryURL string `json:"gallery_url,omitempty"`
// Version, URI and Digest enable cluster-wide upgrade detection —
// without them, the frontend cannot tell whether the installed OCI
// image matches the gallery entry, and upgrades silently never surface.
Version string `json:"version,omitempty"`
URI string `json:"uri,omitempty"`
Digest string `json:"digest,omitempty"`
}
// SubjectNodeBackendStop tells a worker node to stop its gRPC backend process.
// Equivalent to the local deleteProcess(). The node will:
// 1. Best-effort Free() via gRPC
// 2. Kill the backend process
// 3. Can be restarted via another backend.start event.
func SubjectNodeBackendStop(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".backend.stop"
}
// SubjectNodeBackendDelete tells a worker node to delete a backend (stop + remove files).
// Uses NATS request-reply.
func SubjectNodeBackendDelete(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".backend.delete"
}
// BackendDeleteRequest is the payload for a backend.delete NATS request.
type BackendDeleteRequest struct {
Backend string `json:"backend"`
}
// BackendDeleteReply is the response from a backend.delete NATS request.
type BackendDeleteReply struct {
Success bool `json:"success"`
Error string `json:"error,omitempty"`
}
// SubjectNodeModelUnload tells a worker node to unload a model (gRPC Free) without killing the backend.
// Uses NATS request-reply.
func SubjectNodeModelUnload(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".model.unload"
}
// ModelUnloadRequest is the payload for a model.unload NATS request.
type ModelUnloadRequest struct {
ModelName string `json:"model_name"`
Address string `json:"address,omitempty"` // gRPC address of the backend process to unload from
}
// ModelUnloadReply is the response from a model.unload NATS request.
type ModelUnloadReply struct {
Success bool `json:"success"`
Error string `json:"error,omitempty"`
}
// SubjectNodeModelDelete tells a worker node to delete model files from disk.
// Uses NATS request-reply.
func SubjectNodeModelDelete(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".model.delete"
}
// ModelDeleteRequest is the payload for a model.delete NATS request.
type ModelDeleteRequest struct {
ModelName string `json:"model_name"`
}
// ModelDeleteReply is the response from a model.delete NATS request.
type ModelDeleteReply struct {
Success bool `json:"success"`
Error string `json:"error,omitempty"`
}
// SubjectNodeStop tells a serve-backend node to shut down entirely
// (deregister + exit). The node will not restart the backend process.
func SubjectNodeStop(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".stop"
}
// File Staging (Request-Reply — targeted to specific nodes)
// These subjects use request-reply for synchronous file operations.
// SubjectNodeFilesEnsure tells a serve-backend node to download an S3 key to its local cache.
// Reply: {local_path, error}
func SubjectNodeFilesEnsure(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".files.ensure"
}
// SubjectNodeFilesStage tells a serve-backend node to upload a local file to S3.
// Reply: {key, error}
func SubjectNodeFilesStage(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".files.stage"
}
// SubjectNodeFilesTemp tells a serve-backend node to allocate a temp file.
// Reply: {local_path, error}
func SubjectNodeFilesTemp(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".files.temp"
}
// SubjectNodeFilesListDir tells a serve-backend node to list files in a directory.
// Reply: {files: [...], error}
func SubjectNodeFilesListDir(nodeID string) string {
return subjectNodePrefix + sanitizeSubjectToken(nodeID) + ".files.listdir"
}
// Cache Invalidation (Pub/Sub — broadcast to all instances)
const (
SubjectCacheInvalidateSkills = "cache.invalidate.skills"
)
// SubjectCacheInvalidateCollection returns the NATS subject for collection cache invalidation.
func SubjectCacheInvalidateCollection(name string) string {
return "cache.invalidate.collections." + sanitizeSubjectToken(name)
}