Files
LocalAI/tests/e2e/distributed/agent_native_executor_test.go
T
Ettore Di Giacinto da8cf9c9fd feat(distributed): carry an agent cancel on the worker's own tunnel
agent.<name>.cancel was the last family on a message bus, and the only
reason an agent worker dialled one. Its subscriber is the worker running
the execution, and a worker has no database, so the family could not move
to the PostgreSQL fan-out carrier: a cancel published there would reach no
worker while reporting that it had been sent.

It is a control verb now. An agent worker mounts workerctl.PathAgentCancel
on the loopback control plane behind its tunnel and applies the cancel to
the same registry the executor registers a run on. The frontend issues it
through nodes.AgentControlClient.CancelAgentRun.

That call is a FAN-OUT and not a pick, because nothing records which worker
holds a given execution: the claim row names the claiming replica, and it
is deleted when the run ends. Every agent worker a live replica can reach
is asked over its own tunnel, relayed by the peer mesh when a peer holds
it, and each worker answers only for itself.

The answers stay apart, which is why this family was held back. A cancel a
worker made is nil. A cancel some worker could not be asked is
ErrAgentCancelUndelivered, which is neither a refusal nor a missing run. A
cancel every reachable worker declined to own is ErrAgentRunNotOnAnyWorker.
A deployment with no agent worker is ErrNoAgentWorker. Neither new sentinel
wraps ErrWorkerUnroutable and neither is a worker answer, so nothing is
reaped, demoted or evicted because of a cancel.

A worker in the ABSENT CONNECTION condition, one whose tunnel was lost
inside the reconnect grace, counts as undelivered. It is not retried in the
call and not queued: a retry would spend a budget the caller did not
choose, and a queue would need durable state whose only consumer is a run
whose control stream went with the tunnel. A worker whose departure has
outlived the grace is the one routing fact a caller may act on and is
excluded, or a single retired agent node would make every cancel
undelivered for ever.

The fan-out reads a different node set from the pick. A draining worker
takes no new work but is still finishing what it holds, so it is offered
the cancel; a pending one is refused by the tunnel route on every dial and
is not.

With that, nothing in LocalAI connects to NATS. The agent worker's dial,
its credential ladder and its refresh loop are gone, and so is the
frontend's cancel carrier. LOCALAI_NATS_URL is accepted and ignored
everywhere, and distributed mode no longer requires it.

Assisted-by: Claude Opus 5 [claude-code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2026-09-27 03:05:13 +00:00

1281 lines
41 KiB
Go

package distributed_test
import (
"context"
"encoding/json"
"fmt"
"io"
"net"
"net/http"
"sync"
"sync/atomic"
"time"
"github.com/mudler/LocalAI/core/services/agents"
"github.com/mudler/LocalAI/core/services/jobs"
"github.com/mudler/LocalAI/core/services/messaging"
"github.com/mudler/LocalAI/core/services/nodes"
"github.com/mudler/cogito"
openai "github.com/sashabaranov/go-openai"
. "github.com/onsi/ginkgo/v2"
. "github.com/onsi/gomega"
pgdriver "gorm.io/driver/postgres"
"gorm.io/gorm"
"gorm.io/gorm/logger"
)
// mockLLM is a test LLM that returns a fixed response.
type mockLLM struct {
response string
toolCall *openai.ToolCall // if set, first call returns a tool call
callCount atomic.Int32
}
func (m *mockLLM) Ask(ctx context.Context, f cogito.Fragment) (cogito.Fragment, error) {
m.callCount.Add(1)
result := f.AddMessage(cogito.AssistantMessageRole, m.response)
return result, nil
}
func (m *mockLLM) CreateChatCompletion(ctx context.Context, req openai.ChatCompletionRequest) (cogito.LLMReply, cogito.LLMUsage, error) {
m.callCount.Add(1)
msg := openai.ChatCompletionMessage{
Role: "assistant",
Content: m.response,
}
if m.toolCall != nil && m.callCount.Load() == 1 {
// First call: return tool call
msg.Content = ""
msg.ToolCalls = []openai.ToolCall{*m.toolCall}
}
return cogito.LLMReply{
ChatCompletionResponse: openai.ChatCompletionResponse{
Choices: []openai.ChatCompletionChoice{{Message: msg}},
},
}, cogito.LLMUsage{}, nil
}
// mockSSEWriter collects SSE events for testing.
type mockSSEWriter struct {
mu sync.Mutex
events []mockSSEEvent
}
type mockSSEEvent struct {
Event string
Data any
}
func (w *mockSSEWriter) SendEvent(event string, data any) {
w.mu.Lock()
defer w.mu.Unlock()
w.events = append(w.events, mockSSEEvent{Event: event, Data: data})
}
func (w *mockSSEWriter) getEvents() []mockSSEEvent {
w.mu.Lock()
defer w.mu.Unlock()
cp := make([]mockSSEEvent, len(w.events))
copy(cp, w.events)
return cp
}
// mockSSEPool always returns the same writer.
type mockSSEPool struct {
writer *mockSSEWriter
}
func (p *mockSSEPool) GetWriter(key string) agents.SSEWriter {
return p.writer
}
// mockLLMWithCapture is a test LLM that captures the messages it receives.
type mockLLMWithCapture struct {
response string
captureMessages func([]openai.ChatCompletionMessage)
}
func (m *mockLLMWithCapture) Ask(ctx context.Context, f cogito.Fragment) (cogito.Fragment, error) {
return f.AddMessage(cogito.AssistantMessageRole, m.response), nil
}
func (m *mockLLMWithCapture) CreateChatCompletion(ctx context.Context, req openai.ChatCompletionRequest) (cogito.LLMReply, cogito.LLMUsage, error) {
if m.captureMessages != nil {
m.captureMessages(req.Messages)
}
msg := openai.ChatCompletionMessage{
Role: "assistant",
Content: m.response,
}
return cogito.LLMReply{
ChatCompletionResponse: openai.ChatCompletionResponse{
Choices: []openai.ChatCompletionChoice{{Message: msg}},
},
}, cogito.LLMUsage{}, nil
}
// staticSkillProviderTest provides skills for testing.
type staticSkillProviderTest struct {
skills []agents.SkillInfo
}
func (p *staticSkillProviderTest) ListSkills() ([]agents.SkillInfo, error) {
return p.skills, nil
}
// startAgentMockLLMServer starts a mock OpenAI-compatible HTTP server that returns
// the given response text. Supports both streaming and non-streaming requests.
func startAgentMockLLMServer(responseText string) (string, func()) {
GinkgoHelper()
handler := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.URL.Path != "/v1/chat/completions" {
http.NotFound(w, r)
return
}
var req struct {
Stream bool `json:"stream"`
Model string `json:"model"`
}
data, _ := io.ReadAll(r.Body)
json.Unmarshal(data, &req)
if req.Stream {
w.Header().Set("Content-Type", "text/event-stream")
flusher, _ := w.(http.Flusher)
chunk := map[string]any{
"id": "chatcmpl-mock", "model": req.Model,
"choices": []map[string]any{{
"index": 0,
"delta": map[string]any{"role": "assistant", "content": responseText},
}},
}
d, _ := json.Marshal(chunk)
fmt.Fprintf(w, "data: %s\n\n", d)
if flusher != nil {
flusher.Flush()
}
done := map[string]any{
"id": "chatcmpl-mock-done", "model": req.Model,
"choices": []map[string]any{{
"index": 0, "delta": map[string]any{}, "finish_reason": "stop",
}},
}
d, _ = json.Marshal(done)
fmt.Fprintf(w, "data: %s\n\n", d)
fmt.Fprintf(w, "data: [DONE]\n\n")
if flusher != nil {
flusher.Flush()
}
} else {
w.Header().Set("Content-Type", "application/json")
resp := map[string]any{
"id": "chatcmpl-mock", "model": req.Model,
"choices": []map[string]any{{
"index": 0,
"message": map[string]any{"role": "assistant", "content": responseText},
"finish_reason": "stop",
}},
}
json.NewEncoder(w).Encode(resp)
}
})
listener, err := net.Listen("tcp", "127.0.0.1:0")
Expect(err).ToNot(HaveOccurred())
httpServer := &http.Server{Handler: handler}
go httpServer.Serve(listener)
url := fmt.Sprintf("http://%s", listener.Addr().String())
shutdown := func() {
ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
defer cancel()
httpServer.Shutdown(ctx)
}
return url, shutdown
}
// mockConfigProvider returns a fixed config.
type mockConfigProvider struct {
configs map[string]*agents.AgentConfig
}
func (p *mockConfigProvider) GetAgentConfig(userID, name string) (*agents.AgentConfig, error) {
key := name
if userID != "" {
key = userID + ":" + name
}
if cfg, ok := p.configs[key]; ok {
return cfg, nil
}
if cfg, ok := p.configs[name]; ok {
return cfg, nil
}
return nil, fmt.Errorf("agent not found: %s", name)
}
var _ = Describe("Native Agent Executor", Label("Distributed", "AgentNative"), func() {
var (
infra *TestInfra
db *gorm.DB
store *agents.AgentStore
)
BeforeEach(func() {
infra = SetupInfra("localai_agent_native_test")
var err error
db, err = gorm.Open(pgdriver.Open(infra.PGURL), &gorm.Config{
Logger: logger.Default.LogMode(logger.Silent),
})
Expect(err).ToNot(HaveOccurred())
store, err = agents.NewAgentStore(db)
Expect(err).ToNot(HaveOccurred())
// Also migrate BackendNode for registration tests
Expect(db.AutoMigrate(&nodes.BackendNode{})).To(Succeed())
})
Context("ExecuteChat", func() {
It("should execute a chat and deliver response via callbacks", func() {
llm := &mockLLM{response: "Hello! I'm an AI assistant."}
cfg := &agents.AgentConfig{
Name: "test-agent",
Model: "test-model",
SystemPrompt: "You are helpful.",
}
var gotStatus []string
var gotMessage string
var gotSender string
var mu sync.Mutex
cb := agents.Callbacks{
OnStatus: func(s string) {
mu.Lock()
gotStatus = append(gotStatus, s)
mu.Unlock()
},
OnMessage: func(sender, content, id string) {
mu.Lock()
gotSender = sender
gotMessage = content
mu.Unlock()
},
}
response, err := agents.ExecuteChatWithLLM(infra.Ctx, llm, cfg, "Hi there", cb)
Expect(err).ToNot(HaveOccurred())
Expect(response).To(ContainSubstring("AI assistant"))
mu.Lock()
defer mu.Unlock()
Expect(gotStatus).To(ContainElement("processing"))
Expect(gotStatus).To(ContainElement("completed"))
Expect(gotSender).To(Equal("agent"))
Expect(gotMessage).To(ContainSubstring("AI assistant"))
})
It("should return error when no model is configured", func() {
cfg := &agents.AgentConfig{Name: "no-model"}
_, err := agents.ExecuteChat(infra.Ctx, "http://localhost", "", cfg, "hello", agents.Callbacks{})
Expect(err).To(HaveOccurred())
Expect(err.Error()).To(ContainSubstring("no model configured"))
})
It("should respect context cancellation", func() {
// Create a slow mock LLM
slowLLM := &mockLLM{response: "slow response"}
cfg := &agents.AgentConfig{
Name: "cancel-test",
Model: "test",
}
cancelCtx, cancel := context.WithCancel(infra.Ctx)
cancel() // cancel immediately
_, err := agents.ExecuteChatWithLLM(cancelCtx, slowLLM, cfg, "hello", agents.Callbacks{})
// Cancelled context should produce an error
Expect(err).To(HaveOccurred())
Expect(err.Error()).To(Or(
ContainSubstring("context canceled"),
ContainSubstring("context deadline exceeded"),
))
})
})
Context("WorkerExecutor", func() {
It("should dispatch chat via NATS and receive response", func() {
bridge := agents.NewEventBridge(infra.NC, nil, "test-instance", nil)
configs := &mockConfigProvider{configs: map[string]*agents.AgentConfig{
"test-agent": {
Name: "test-agent",
Model: "test-model",
SystemPrompt: "Be helpful.",
},
}}
// Subscribe to agent events to capture the response
var receivedEvents []agents.AgentEvent
var eventMu sync.Mutex
sub, err := infra.NC.Subscribe(messaging.SubjectAgentEvents("test-agent", "user1"), func(data []byte) {
var evt agents.AgentEvent
if json.Unmarshal(data, &evt) == nil {
eventMu.Lock()
receivedEvents = append(receivedEvents, evt)
eventMu.Unlock()
}
})
Expect(err).ToNot(HaveOccurred())
defer sub.Unsubscribe()
// The executor is handed one run by the frontend replica that
// claimed it, over the control verb, and publishes everything it
// produces onto the writer that replica is reading. Here that
// writer is the bus itself, so the subscription above still sees
// the same events on the same subject.
executor := agents.NewWorkerExecutor(bridge, configs, "http://localhost:8080", "test-key")
reply, err := executor.Execute(infra.Ctx, mustAgentJSON(agents.AgentChatEvent{
AgentName: "test-agent",
UserID: "user1",
Message: "Hello",
MessageID: "msg-test-001",
Role: "user",
}), infra.NC)
Expect(err).ToNot(HaveOccurred(), "an error here is this worker failing to SERVE the verb, not the run failing")
Expect(string(reply)).To(ContainSubstring("status"))
// The run reaches no real LLM, so what it produces is a status the
// frontend can show. What matters is that it travelled.
Eventually(func() int {
eventMu.Lock()
defer eventMu.Unlock()
return len(receivedEvents)
}, "5s").Should(BeNumerically(">=", 1))
eventMu.Lock()
hasStatus := false
for _, evt := range receivedEvents {
if evt.EventType == "json_message_status" {
hasStatus = true
}
}
eventMu.Unlock()
Expect(hasStatus).To(BeTrue(), "a status event should reach the subscriber")
})
It("should handle cancellation via EventBridge", func() {
// The local half: a run held by THIS process is cancelled without
// anything leaving it. The cancel of a run held by an agent WORKER
// travels over that worker's tunnel and is driven end to end, on a
// real one, in agent_distributed_test.go.
bridge := agents.NewEventBridge(infra.NC, nil, "cancel-test", nil)
var cancelled atomic.Bool
bridge.RegisterCancel("test-msg-id", func() {
cancelled.Store(true)
})
Expect(bridge.CancelExecution(infra.Ctx, "test-agent", "user1", "test-msg-id")).To(Succeed())
Eventually(func() bool { return cancelled.Load() }, "5s").Should(BeTrue())
})
It("should execute agent chat from enriched payload without ConfigProvider", func() {
bridge := agents.NewEventBridge(infra.NC, nil, "enriched-test", nil)
// Executor with NO ConfigProvider (simulating DB-free worker)
executor := agents.NewWorkerExecutor(bridge, nil, "http://localhost:8080", "test-key")
// Subscribe to events to verify processing
var receivedEvents []agents.AgentEvent
var eventMu sync.Mutex
sub, err := infra.NC.Subscribe(messaging.SubjectAgentEvents("enriched-agent", "user1"), func(data []byte) {
var evt agents.AgentEvent
if json.Unmarshal(data, &evt) == nil {
eventMu.Lock()
receivedEvents = append(receivedEvents, evt)
eventMu.Unlock()
}
})
Expect(err).ToNot(HaveOccurred())
defer sub.Unsubscribe()
FlushNATS(infra.NC)
// An enriched AgentChatEvent with embedded Config, exactly the body
// the claim row carries.
evt := agents.AgentChatEvent{
AgentName: "enriched-agent",
UserID: "user1",
Message: "Hello from enriched payload",
MessageID: "msg-enriched-001",
Role: "user",
Config: &agents.AgentConfig{
Name: "enriched-agent",
Model: "test-model",
SystemPrompt: "Be helpful.",
},
}
_, err = executor.Execute(infra.Ctx, mustAgentJSON(evt), infra.NC)
Expect(err).ToNot(HaveOccurred())
// The executor should process this even without a ConfigProvider.
// It will fail at ExecuteChat (no real LLM), but it should at least
// publish a processing status event before failing.
Eventually(func() int {
eventMu.Lock()
defer eventMu.Unlock()
return len(receivedEvents)
}, "5s").Should(BeNumerically(">=", 1))
// Verify we got status events (processing or error — both prove the
// dispatcher accepted the enriched event without a ConfigProvider)
eventMu.Lock()
hasEvent := len(receivedEvents) > 0
eventMu.Unlock()
Expect(hasEvent).To(BeTrue(), "executor should process enriched event without ConfigProvider")
})
It("should marshal/unmarshal matching LocalAGI format", func() {
cfg := agents.AgentConfig{
Name: "test",
Model: "llama3",
SystemPrompt: "Be helpful",
MCPServers: []agents.MCPServer{{URL: "http://mcp.example.com", Token: "tok"}},
Actions: []agents.ActionsConfig{{Type: "search", Config: "{}"}},
EnableSkills: true,
MaxAttempts: 3,
}
data, err := json.Marshal(cfg)
Expect(err).ToNot(HaveOccurred())
// Verify key field names match LocalAGI
var raw map[string]any
Expect(json.Unmarshal(data, &raw)).To(Succeed())
Expect(raw).To(HaveKey("name"))
Expect(raw).To(HaveKey("model"))
Expect(raw).To(HaveKey("system_prompt"))
Expect(raw).To(HaveKey("mcp_servers"))
Expect(raw).To(HaveKey("actions"))
Expect(raw).To(HaveKey("enable_skills"))
Expect(raw).To(HaveKey("max_attempts"))
// Round-trip
var cfg2 agents.AgentConfig
Expect(json.Unmarshal(data, &cfg2)).To(Succeed())
Expect(cfg2.Name).To(Equal("test"))
Expect(cfg2.Model).To(Equal("llama3"))
Expect(cfg2.MCPServers).To(HaveLen(1))
Expect(cfg2.EnableSkills).To(BeTrue())
})
It("should survive PostgreSQL round-trip", func() {
cfg := agents.AgentConfig{
Name: "db-test",
Model: "qwen",
SystemPrompt: "Hello world",
EnableSkills: true,
SelectedSkills: []string{"skill-a", "skill-b"},
}
configJSON, err := json.Marshal(cfg)
Expect(err).ToNot(HaveOccurred())
// Save to DB
Expect(store.SaveConfig(&agents.AgentConfigRecord{
UserID: "u1",
Name: cfg.Name,
ConfigJSON: string(configJSON),
Status: "active",
})).To(Succeed())
// Load from DB
rec, err := store.GetConfig("u1", "db-test")
Expect(err).ToNot(HaveOccurred())
var loaded agents.AgentConfig
Expect(agents.ParseConfigJSON(rec.ConfigJSON, &loaded)).To(Succeed())
Expect(loaded.Name).To(Equal("db-test"))
Expect(loaded.Model).To(Equal("qwen"))
Expect(loaded.SelectedSkills).To(ConsistOf("skill-a", "skill-b"))
})
})
Context("Node Registration with NodeType", func() {
It("should store node_type for agent workers", func() {
registry, err := nodes.NewNodeRegistry(db)
Expect(err).ToNot(HaveOccurred())
node := &nodes.BackendNode{
Name: "agent-worker-1",
NodeType: nodes.NodeTypeAgent,
Status: "healthy",
}
Expect(registry.Register(context.Background(), node, true)).To(Succeed())
// Verify node type is stored
loaded, err := registry.Get(context.Background(), node.ID)
Expect(err).ToNot(HaveOccurred())
Expect(loaded.NodeType).To(Equal(nodes.NodeTypeAgent))
Expect(loaded.Name).To(Equal("agent-worker-1"))
})
It("should list both backend and agent workers", func() {
registry, err := nodes.NewNodeRegistry(db)
Expect(err).ToNot(HaveOccurred())
backend := &nodes.BackendNode{
Name: "backend-1",
NodeType: nodes.NodeTypeBackend,
Address: "localhost:50051",
}
agent := &nodes.BackendNode{
Name: "agent-1",
NodeType: nodes.NodeTypeAgent,
}
Expect(registry.Register(context.Background(), backend, true)).To(Succeed())
Expect(registry.Register(context.Background(), agent, true)).To(Succeed())
allNodes, err := registry.List(context.Background())
Expect(err).ToNot(HaveOccurred())
var backendCount, agentCount int
for _, n := range allNodes {
switch n.NodeType {
case nodes.NodeTypeBackend:
backendCount++
case nodes.NodeTypeAgent:
agentCount++
}
}
Expect(backendCount).To(Equal(1))
Expect(agentCount).To(Equal(1))
})
})
Context("Full Distributed Chat Flow", func() {
It("should dispatch chat via NATS, execute, and publish response via EventBridge", func() {
bridge := agents.NewEventBridge(infra.NC, nil, "flow-test", nil)
// Store agent config in PostgreSQL
cfg := agents.AgentConfig{
Name: "flow-agent",
Model: "test-model",
SystemPrompt: "You are a test agent.",
}
configJSON, err := json.Marshal(cfg)
Expect(err).ToNot(HaveOccurred())
Expect(store.SaveConfig(&agents.AgentConfigRecord{
UserID: "user1",
Name: "flow-agent",
ConfigJSON: string(configJSON),
Status: "active",
})).To(Succeed())
// Subscribe to events to capture the full flow
var receivedEvents []agents.AgentEvent
var eventMu sync.Mutex
sub, err := infra.NC.Subscribe(messaging.SubjectAgentEvents("flow-agent", "user1"), func(data []byte) {
var evt agents.AgentEvent
if json.Unmarshal(data, &evt) == nil {
eventMu.Lock()
receivedEvents = append(receivedEvents, evt)
eventMu.Unlock()
}
})
Expect(err).ToNot(HaveOccurred())
defer sub.Unsubscribe()
configs := &mockConfigProvider{configs: map[string]*agents.AgentConfig{
"flow-agent": &cfg,
}}
executor := agents.NewWorkerExecutor(bridge, configs, "http://localhost:8080", "test-key")
_, err = executor.Execute(infra.Ctx, mustAgentJSON(agents.AgentChatEvent{
AgentName: "flow-agent",
UserID: "user1",
Message: "Hello flow test",
MessageID: "msg-flow-001",
Role: "user",
}), infra.NC)
Expect(err).ToNot(HaveOccurred())
Eventually(func() int {
eventMu.Lock()
defer eventMu.Unlock()
return len(receivedEvents)
}, "5s").Should(BeNumerically(">=", 1))
eventMu.Lock()
var hasProcessing bool
for _, evt := range receivedEvents {
if evt.EventType == "json_message_status" {
hasProcessing = true
}
}
eventMu.Unlock()
Expect(hasProcessing).To(BeTrue(), "expected a status event")
})
})
Context("Background Agent Execution", func() {
It("should use inner monologue template with permanent goal for system role", func() {
// ExecuteBackgroundRunWithLLM should substitute {{.Goal}} in the inner monologue template
cfg := &agents.AgentConfig{
Name: "bg-agent",
Model: "test-model",
PermanentGoal: "Monitor system health and report issues",
InnerMonologueTemplate: "Your goal is: {{.Goal}}. What should you do next?",
SystemPrompt: "You are an autonomous agent.",
}
llm := &mockLLM{response: "I will monitor system health."}
cb := agents.Callbacks{}
response, err := agents.ExecuteBackgroundRunWithLLM(infra.Ctx, llm, cfg, cb)
// ExecuteBackgroundRunWithLLM should complete without panicking.
// The mock LLM always succeeds, so we expect no error.
Expect(err).ToNot(HaveOccurred())
Expect(len(response)).To(BeNumerically(">=", 0))
})
It("should use default inner monologue when template is empty", func() {
cfg := &agents.AgentConfig{
Name: "bg-default",
Model: "test-model",
PermanentGoal: "Keep things running",
SystemPrompt: "You are helpful.",
}
// ExecuteBackgroundRun with empty template should use DefaultInnerMonologueTemplate
Expect(agents.DefaultInnerMonologueTemplate).To(ContainSubstring("{{.Goal}}"))
// The function should not panic with empty template
_, _ = agents.ExecuteBackgroundRun(infra.Ctx, "http://localhost:8080", "key", cfg, agents.Callbacks{})
})
It("should dispatch background run via NATS with system role", func() {
bridge := agents.NewEventBridge(infra.NC, nil, "bg-test", nil)
cfg := agents.AgentConfig{
Name: "bg-agent",
Model: "test-model",
PermanentGoal: "Monitor system health",
SystemPrompt: "You are a monitoring agent.",
}
configs := &mockConfigProvider{configs: map[string]*agents.AgentConfig{
"bg-agent": &cfg,
}}
// Subscribe to capture events
var receivedEvents []agents.AgentEvent
var eventMu sync.Mutex
sub, err := infra.NC.Subscribe(messaging.SubjectAgentEvents("bg-agent", "system"), func(data []byte) {
var evt agents.AgentEvent
if json.Unmarshal(data, &evt) == nil {
eventMu.Lock()
receivedEvents = append(receivedEvents, evt)
eventMu.Unlock()
}
})
Expect(err).ToNot(HaveOccurred())
defer sub.Unsubscribe()
executor := agents.NewWorkerExecutor(bridge, configs, "http://localhost:8080", "test-key")
// A background/system role run, the shape the scheduler enqueues.
evt := agents.AgentChatEvent{
AgentName: "bg-agent",
UserID: "system",
Message: "",
MessageID: "bg-1",
Role: "system",
}
_, err = executor.Execute(infra.Ctx, mustAgentJSON(evt), infra.NC)
Expect(err).ToNot(HaveOccurred())
// Should receive at least a processing status
Eventually(func() int {
eventMu.Lock()
defer eventMu.Unlock()
return len(receivedEvents)
}, "10s").Should(BeNumerically(">=", 1))
})
})
Context("Skills Injection", func() {
It("should render skills prompt with default template", func() {
skills := []agents.SkillInfo{
{Name: "web_search", Description: "Search the web for information"},
{Name: "code_review", Description: "Review code for bugs"},
}
result := agents.RenderSkillsPrompt(skills, "")
Expect(result).To(ContainSubstring("web_search"))
Expect(result).To(ContainSubstring("code_review"))
Expect(result).To(ContainSubstring("available_skills"))
})
It("should render skills with custom template", func() {
skills := []agents.SkillInfo{
{Name: "tool1", Description: "First tool"},
}
custom := "Tools: {{range .Skills}}{{.Name}} - {{.Description}}; {{end}}"
result := agents.RenderSkillsPrompt(skills, custom)
Expect(result).To(ContainSubstring("tool1 - First tool"))
})
It("should filter skills by selected_skills", func() {
all := []agents.SkillInfo{
{Name: "a", Description: "skill a"},
{Name: "b", Description: "skill b"},
{Name: "c", Description: "skill c"},
}
filtered := agents.FilterSkills(all, []string{"a", "c"})
Expect(filtered).To(HaveLen(2))
names := []string{filtered[0].Name, filtered[1].Name}
Expect(names).To(ConsistOf("a", "c"))
})
It("should return all skills when selected_skills is empty", func() {
all := []agents.SkillInfo{
{Name: "x", Description: "skill x"},
{Name: "y", Description: "skill y"},
}
filtered := agents.FilterSkills(all, nil)
Expect(filtered).To(HaveLen(2))
})
It("should render full content when Content field is set", func() {
skills := []agents.SkillInfo{
{Name: "search", Description: "Search the web", Content: "You are a web search skill. Given a query, search the web and return results."},
}
result := agents.RenderSkillsPrompt(skills, "")
Expect(result).To(ContainSubstring("<content>"))
Expect(result).To(ContainSubstring("You are a web search skill"))
Expect(result).NotTo(ContainSubstring("<description>"))
})
It("should fall back to description when Content is empty", func() {
skills := []agents.SkillInfo{
{Name: "search", Description: "Search the web"},
}
result := agents.RenderSkillsPrompt(skills, "")
Expect(result).To(ContainSubstring("<description>"))
Expect(result).To(ContainSubstring("Search the web"))
Expect(result).NotTo(ContainSubstring("<content>"))
})
})
Context("Agent Scheduler", func() {
It("should detect due standalone agents and publish events", func() {
// Create an agent with standalone_job=true and periodic_runs=1s
cfg := agents.AgentConfig{
Name: "cron-agent",
Model: "test-model",
StandaloneJob: true,
PeriodicRuns: "1s",
SystemPrompt: "You are autonomous.",
}
configJSON, _ := json.Marshal(cfg)
Expect(store.SaveConfig(&agents.AgentConfigRecord{
UserID: "u1",
Name: "cron-agent",
ConfigJSON: string(configJSON),
Status: "active",
// LastRunAt is nil — never run, so it's due immediately
})).To(Succeed())
// The scheduler writes a claim row rather than publishing. A
// publish onto a queue group nobody had joined succeeded and the
// background run simply never happened, with nothing recording that
// it had been due.
Expect(jobs.MigrateClaims(infra.Ctx, db)).To(Succeed())
scheduler := agents.NewAgentScheduler(db, store)
schedCtx, schedCancel := context.WithCancel(infra.Ctx)
defer schedCancel()
go scheduler.Start(schedCtx)
var rows []jobs.WorkClaim
Eventually(func() int {
rows = nil
Expect(db.Where("kind = ?", string(jobs.ClaimKindAgentRun)).Find(&rows).Error).To(Succeed())
return len(rows)
}, "20s").Should(BeNumerically(">=", 1))
var evt agents.AgentChatEvent
Expect(json.Unmarshal(rows[0].Payload, &evt)).To(Succeed())
Expect(evt.AgentName).To(Equal("cron-agent"))
Expect(evt.UserID).To(Equal("u1"))
Expect(evt.Role).To(Equal("system"))
// Verify enriched payload: config should be embedded in the event
Expect(evt.Config).ToNot(BeNil(), "scheduler should embed config in the event")
Expect(evt.Config.Model).To(Equal("test-model"))
Expect(evt.Config.StandaloneJob).To(BeTrue())
})
It("should skip agents without standalone_job", func() {
cfg := agents.AgentConfig{
Name: "no-cron-agent",
Model: "test-model",
StandaloneJob: false,
}
configJSON, _ := json.Marshal(cfg)
Expect(store.SaveConfig(&agents.AgentConfigRecord{
UserID: "u1",
Name: "no-cron-agent",
ConfigJSON: string(configJSON),
Status: "active",
})).To(Succeed())
Expect(agents.IsDueExported(nil, 10*time.Minute)).To(BeTrue(), "nil lastRun should be due")
// But the scheduler skips non-standalone agents
// (tested implicitly via the scheduler — it won't publish for this agent)
})
It("should not re-run before interval elapses", func() {
now := time.Now()
Expect(agents.IsDueExported(&now, 10*time.Minute)).To(BeFalse(), "just ran should not be due")
past := time.Now().Add(-11 * time.Minute)
Expect(agents.IsDueExported(&past, 10*time.Minute)).To(BeTrue(), "11m ago with 10m interval should be due")
})
})
Context("Agent Chat with Tool Calls and Skills", func() {
It("should execute agent chat where LLM calls request_skill tool", func() {
// Configure mockLLM:
// - First call: return tool_call for request_skill with skill_name="web_search"
// - Second call: return final response incorporating skill content
llm := &mockLLM{
response: "Based on the web search skill, here is the information you need.",
toolCall: &openai.ToolCall{
ID: "call_001",
Type: openai.ToolTypeFunction,
Function: openai.FunctionCall{
Name: "request_skill",
Arguments: `{"skill_name":"web_search"}`,
},
},
}
skills := []agents.SkillInfo{
{Name: "web_search", Description: "Search the web for information", Content: "You are a web search skill. Query external sources and return results."},
{Name: "code_review", Description: "Review code for bugs"},
}
cfg := &agents.AgentConfig{
Name: "skill-tool-agent",
Model: "test-model",
SystemPrompt: "You are helpful.",
EnableSkills: true,
SkillsMode: agents.SkillsModeTools,
SelectedSkills: []string{"web_search"},
MaxIterations: 1,
}
var statuses []string
var finalMessage string
var toolCalls []string
var toolResults []string
var mu sync.Mutex
cb := agents.Callbacks{
OnStatus: func(s string) {
mu.Lock()
statuses = append(statuses, s)
mu.Unlock()
},
OnMessage: func(sender, content, id string) {
mu.Lock()
if sender == "agent" {
finalMessage = content
}
mu.Unlock()
},
OnToolCall: func(name, args string) {
mu.Lock()
toolCalls = append(toolCalls, name)
mu.Unlock()
},
OnToolResult: func(name, result string) {
mu.Lock()
toolResults = append(toolResults, name+": "+result)
mu.Unlock()
},
}
opts := agents.ExecuteChatOpts{
SkillProvider: &staticSkillProviderTest{skills: skills},
}
response, err := agents.ExecuteChatWithLLM(infra.Ctx, llm, cfg, "Find me info about Go", cb, opts)
Expect(err).ToNot(HaveOccurred())
Expect(response).To(ContainSubstring("web search skill"))
mu.Lock()
defer mu.Unlock()
Expect(statuses).To(ContainElement("processing"))
Expect(statuses).To(ContainElement("completed"))
Expect(finalMessage).To(ContainSubstring("web search skill"))
// Verify the tool was called
Expect(toolCalls).To(ContainElement("request_skill"))
// Verify tool result contains skill content
Expect(toolResults).To(HaveLen(1))
Expect(toolResults[0]).To(ContainSubstring("web_search"))
})
It("should inject skills into system prompt in prompt mode", func() {
// In prompt mode, skills are injected into the system prompt.
// The LLM should see the skill content in the system prompt.
var receivedMessages []openai.ChatCompletionMessage
llm := &mockLLMWithCapture{
response: "I see the skill content in my prompt.",
captureMessages: func(msgs []openai.ChatCompletionMessage) {
receivedMessages = msgs
},
}
skills := []agents.SkillInfo{
{Name: "data_analysis", Description: "Analyze data", Content: "You are a data analysis skill. Use pandas and numpy."},
}
cfg := &agents.AgentConfig{
Name: "prompt-skill-agent",
Model: "test-model",
SystemPrompt: "You are helpful.",
EnableSkills: true,
SkillsMode: agents.SkillsModePrompt,
}
opts := agents.ExecuteChatOpts{
SkillProvider: &staticSkillProviderTest{skills: skills},
}
_, err := agents.ExecuteChatWithLLM(infra.Ctx, llm, cfg, "Analyze this data", agents.Callbacks{}, opts)
Expect(err).ToNot(HaveOccurred())
// Verify system prompt contains skill content
Expect(receivedMessages).ToNot(BeEmpty())
systemMsg := receivedMessages[0]
Expect(systemMsg.Role).To(Equal("system"))
Expect(systemMsg.Content).To(ContainSubstring("data_analysis"))
Expect(systemMsg.Content).To(ContainSubstring("pandas and numpy"))
})
})
Context("Full Distributed Agent Execution via NATS with Mock LLM Server", func() {
It("should dispatch chat via NATS, execute on worker with mock LLM, and publish response via EventBridge", func() {
// Start mock LLM HTTP server
llmURL, llmShutdown := startAgentMockLLMServer("I found the answer using my skills.")
defer llmShutdown()
bridge := agents.NewEventBridge(infra.NC, nil, "full-e2e-test", nil)
// Subscribe to agent events
var receivedEvents []agents.AgentEvent
var eventMu sync.Mutex
sub, err := infra.NC.Subscribe(messaging.SubjectAgentEvents("e2e-agent", "user1"), func(data []byte) {
var evt agents.AgentEvent
if json.Unmarshal(data, &evt) == nil {
eventMu.Lock()
receivedEvents = append(receivedEvents, evt)
eventMu.Unlock()
}
})
Expect(err).ToNot(HaveOccurred())
defer sub.Unsubscribe()
// Point the executor at our mock LLM server
executor := agents.NewWorkerExecutor(bridge, nil, llmURL, "test-key")
FlushNATS(infra.NC)
// The enriched event with skills, as the claim row carries it
evt := agents.AgentChatEvent{
AgentName: "e2e-agent",
UserID: "user1",
Message: "Hello, use your skills",
MessageID: "msg-e2e-001",
Role: "user",
Config: &agents.AgentConfig{
Name: "e2e-agent",
Model: "test-model",
SystemPrompt: "You are helpful.",
EnableSkills: true,
SkillsMode: agents.SkillsModePrompt,
},
Skills: []agents.SkillInfo{
{Name: "search", Description: "Search the web", Content: "Full search skill content here"},
},
}
reply, err := executor.Execute(infra.Ctx, mustAgentJSON(evt), infra.NC)
Expect(err).ToNot(HaveOccurred())
// A run that reached a conclusion is this worker's own answer, and
// it travels on the response body rather than only as a publish.
Expect(string(reply)).To(ContainSubstring(`"status":"completed"`))
// Waited for by CONTENT and not by count. A run with streaming on
// publishes an unbounded number of stream_event lines before its
// terminal status, so "at least three events have arrived" is
// reached long before the completed status is, and the assertions
// below then read a snapshot that cannot contain it yet. That is a
// race in the spec rather than in the executor, and it fails about
// one run in five.
seen := func() (agentMessage, completed bool) {
eventMu.Lock()
defer eventMu.Unlock()
for _, evt := range receivedEvents {
if evt.EventType == "json_message" && evt.Sender == "agent" {
agentMessage = true
Expect(evt.Content).To(ContainSubstring("found the answer"))
}
if evt.EventType == "json_message_status" && evt.Metadata != "" {
var meta map[string]string
// A metadata blob this spec cannot read is not a
// completed status, and saying so beats failing the
// whole run on one malformed event.
if err := json.Unmarshal([]byte(evt.Metadata), &meta); err == nil && meta["status"] == "completed" {
completed = true
}
}
}
return agentMessage, completed
}
Eventually(func() bool { _, completed := seen(); return completed }, "15s").
Should(BeTrue(), "should receive completed status via EventBridge")
agentMessage, _ := seen()
Expect(agentMessage).To(BeTrue(), "should receive agent response message via EventBridge")
})
It("should execute background agent run via NATS dispatcher with mock LLM", func() {
// Start mock LLM HTTP server
llmURL, llmShutdown := startAgentMockLLMServer("All systems operational. No issues detected.")
defer llmShutdown()
bridge := agents.NewEventBridge(infra.NC, nil, "bg-e2e-test", nil)
// Subscribe to agent events
var receivedEvents []agents.AgentEvent
var eventMu sync.Mutex
sub, err := infra.NC.Subscribe(messaging.SubjectAgentEvents("bg-e2e-agent", "system"), func(data []byte) {
var evt agents.AgentEvent
if json.Unmarshal(data, &evt) == nil {
eventMu.Lock()
receivedEvents = append(receivedEvents, evt)
eventMu.Unlock()
}
})
Expect(err).ToNot(HaveOccurred())
defer sub.Unsubscribe()
executor := agents.NewWorkerExecutor(bridge, nil, llmURL, "test-key")
FlushNATS(infra.NC)
// A system-role run (background/autonomous)
evt := agents.AgentChatEvent{
AgentName: "bg-e2e-agent",
UserID: "system",
Message: "",
MessageID: "msg-bg-e2e-001",
Role: "system",
Config: &agents.AgentConfig{
Name: "bg-e2e-agent",
Model: "test-model",
PermanentGoal: "Monitor all services and report status",
SystemPrompt: "You are an autonomous monitoring agent.",
InnerMonologueTemplate: "Your goal is: {{.Goal}}. What should you do?",
},
}
reply, err := executor.Execute(infra.Ctx, mustAgentJSON(evt), infra.NC)
Expect(err).ToNot(HaveOccurred())
Expect(string(reply)).To(ContainSubstring(`"status":"completed"`))
// Wait for agent response
Eventually(func() int {
eventMu.Lock()
defer eventMu.Unlock()
return len(receivedEvents)
}, "15s").Should(BeNumerically(">=", 2))
eventMu.Lock()
defer eventMu.Unlock()
var hasAgentMessage, hasCompleted bool
for _, evt := range receivedEvents {
if evt.EventType == "json_message" && evt.Sender == "agent" {
hasAgentMessage = true
Expect(evt.Content).To(ContainSubstring("systems operational"))
}
if evt.EventType == "json_message_status" && evt.Metadata != "" {
var meta map[string]string
json.Unmarshal([]byte(evt.Metadata), &meta)
if meta["status"] == "completed" {
hasCompleted = true
}
}
}
Expect(hasAgentMessage).To(BeTrue(), "background agent should produce a response via EventBridge")
Expect(hasCompleted).To(BeTrue(), "background agent should complete")
})
})
Context("Background Agent Execution with Mock LLM", func() {
It("should execute background run with mock LLM and verify goal in prompt", func() {
var receivedMessages []openai.ChatCompletionMessage
llm := &mockLLMWithCapture{
response: "System health is nominal. All services running.",
captureMessages: func(msgs []openai.ChatCompletionMessage) {
receivedMessages = msgs
},
}
cfg := &agents.AgentConfig{
Name: "bg-llm-agent",
Model: "test-model",
PermanentGoal: "Monitor system health and report issues",
SystemPrompt: "You are an autonomous monitoring agent.",
}
var statuses []string
var gotMessage string
var mu sync.Mutex
cb := agents.Callbacks{
OnStatus: func(s string) {
mu.Lock()
statuses = append(statuses, s)
mu.Unlock()
},
OnMessage: func(sender, content, id string) {
mu.Lock()
if sender == "agent" {
gotMessage = content
}
mu.Unlock()
},
}
response, err := agents.ExecuteBackgroundRunWithLLM(infra.Ctx, llm, cfg, cb)
Expect(err).ToNot(HaveOccurred())
Expect(response).To(ContainSubstring("System health"))
mu.Lock()
defer mu.Unlock()
Expect(statuses).To(ContainElement("processing"))
Expect(statuses).To(ContainElement("completed"))
Expect(gotMessage).To(ContainSubstring("System health"))
// Verify the inner monologue template was substituted with the goal
Expect(receivedMessages).To(HaveLen(2)) // system + user (inner monologue)
userMsg := receivedMessages[1]
Expect(userMsg.Role).To(Equal("user"))
Expect(userMsg.Content).To(ContainSubstring("Monitor system health and report issues"))
})
It("should execute background run with skills", func() {
var receivedMessages []openai.ChatCompletionMessage
llm := &mockLLMWithCapture{
response: "Executed monitoring skill successfully.",
captureMessages: func(msgs []openai.ChatCompletionMessage) {
receivedMessages = msgs
},
}
cfg := &agents.AgentConfig{
Name: "bg-skill-agent",
Model: "test-model",
PermanentGoal: "Check service status",
SystemPrompt: "You are a monitoring agent.",
EnableSkills: true,
SkillsMode: agents.SkillsModePrompt,
}
skills := []agents.SkillInfo{
{Name: "health_check", Description: "Check service health", Content: "Run health checks on all services."},
}
opts := agents.ExecuteChatOpts{
SkillProvider: &staticSkillProviderTest{skills: skills},
}
response, err := agents.ExecuteBackgroundRunWithLLM(infra.Ctx, llm, cfg, agents.Callbacks{}, opts)
Expect(err).ToNot(HaveOccurred())
Expect(response).To(ContainSubstring("monitoring skill"))
// Verify skills were injected into system prompt
Expect(receivedMessages).ToNot(BeEmpty())
systemMsg := receivedMessages[0]
Expect(systemMsg.Content).To(ContainSubstring("health_check"))
Expect(systemMsg.Content).To(ContainSubstring("Run health checks"))
})
})
Context("Config Metadata", func() {
It("should return all expected sections", func() {
meta := agents.DefaultConfigMeta()
Expect(meta.Fields).ToNot(BeEmpty())
sections := map[string]bool{}
for _, f := range meta.Fields {
sections[f.Tags.Section] = true
}
Expect(sections).To(HaveKey("BasicInfo"))
Expect(sections).To(HaveKey("ModelSettings"))
Expect(sections).To(HaveKey("MemorySettings"))
Expect(sections).To(HaveKey("PromptsGoals"))
Expect(sections).To(HaveKey("AdvancedSettings"))
Expect(sections).To(HaveKey("MCP"))
})
It("should include key fields", func() {
meta := agents.DefaultConfigMeta()
fieldNames := map[string]bool{}
for _, f := range meta.Fields {
fieldNames[f.Name] = true
}
Expect(fieldNames).To(HaveKey("name"))
Expect(fieldNames).To(HaveKey("model"))
Expect(fieldNames).To(HaveKey("system_prompt"))
Expect(fieldNames).To(HaveKey("enable_kb"))
Expect(fieldNames).To(HaveKey("kb_mode"))
Expect(fieldNames).To(HaveKey("enable_skills"))
Expect(fieldNames).To(HaveKey("mcp_stdio_servers"))
Expect(fieldNames).To(HaveKey("permanent_goal"))
})
})
})
// mustAgentJSON encodes one agent chat event as the control verb's request
// body, which is byte for byte what the claim row carries.
func mustAgentJSON(evt agents.AgentChatEvent) json.RawMessage {
GinkgoHelper()
raw, err := json.Marshal(evt)
Expect(err).ToNot(HaveOccurred())
return raw
}