chore: 本地测试快照(#7 自底向上分解 + _parse_task 健壮化 + handoff 透传 + 本地部署清单)

仅供导出到 xiaohei/Agentswarm 的本地测试镜像快照,非主仓 PR。
2026-06-17 12:58:28 +08:00
parent f9df550ac9
commit 4d77b9263e
2 changed files with 108 additions and 0 deletions
@@ -0,0 +1,104 @@
+# Local (Docker Desktop k8s) orchestrator deploy — adapted from orchestrator-deployment.yaml.
+# Differences vs prod: local-built arm64 image (imagePullPolicy: Never), host Docker redis via the
+# kind gateway, OpenAI base = api.heicode.cc, NO azkv (model key from an optional local Secret).
+apiVersion: v1
+kind: Service
+metadata:
+  name: orchestrator-service
+  namespace: swarm-system
+  labels: { app: orchestrator }
+spec:
+  type: ClusterIP
+  ports:
+    - port: 8000
+      targetPort: 8000
+      name: http
+  selector: { app: orchestrator }
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: orchestrator
+  namespace: swarm-system
+  labels: { app: orchestrator }
+spec:
+  replicas: 1
+  selector:
+    matchLabels: { app: orchestrator }
+  template:
+    metadata:
+      labels: { app: orchestrator }
+    spec:
+      serviceAccountName: swarm-orchestrator
+      containers:
+        - name: orchestrator
+          image: swarm-orchestrator:local
+          imagePullPolicy: Never           # use the node-loaded local image; never pull
+          ports:
+            - containerPort: 8000
+              name: http
+          env:
+            # Reuse the existing in-cluster redis (Service redis-service in swarm-system).
+            - name: REDIS_HOST
+              value: "redis-service"
+            - name: REDIS_PORT
+              value: "6379"
+            - name: REDIS_DB
+              value: "0"
+            - name: LOG_LEVEL
+              value: "INFO"
+            # Swarm launches agent pods in-cluster using the local agent image.
+            - name: AGENT_LAUNCH_BACKEND
+              value: "kubernetes"
+            - name: AGENT_POD_IMAGE
+              value: "swarm-agent:local"
+            - name: AGENT_POD_NAMESPACE
+              value: "swarm-system"
+            - name: ORCHESTRATOR_PUBLIC_URL
+              value: "ws://orchestrator-service.swarm-system.svc.cluster.local:8000"
+            # User-supplied LLM gateway (OpenAI-compatible).
+            - name: AGENT_OPENAI_API_BASE
+              value: "https://api.heicode.cc/v1"
+            - name: OPENAI_API_BASE
+              value: "https://api.heicode.cc/v1"
+            # Default model id for launched agents (benchmark target; overridable per create request).
+            - name: OPENAI_MODEL
+              value: "qwen3.7-max"
+            # Agent pool window (local test): at least 16, up to 64 per run.
+            - name: AGENT_LAUNCH_MIN_POOL
+              value: "16"
+            - name: AGENT_LAUNCH_POOL_SIZE
+              value: "16"
+            - name: AGENT_LAUNCH_MAX_POOL
+              value: "64"
+            - name: MAX_AGENTS_PER_USER
+              value: "64"
+            # agent_swarm#7 bottom-up decomposition: how many subtasks one run may spawn.
+            - name: AGENT_PROPOSAL_BUDGET
+              value: "12"
+            # Local: no Azure workload identity. Model key comes from an OPTIONAL local Secret
+            # (swarm-model-key/OPENAI_API_KEY). resolve_model_key() falls back to this OPENAI_API_KEY.
+            # Create it to enable real LLM calls:
+            #   kubectl -n swarm-system create secret generic swarm-model-key \
+            #     --from-literal=OPENAI_API_KEY=sk-xxxx
+            - name: OPENAI_API_KEY
+              valueFrom:
+                secretKeyRef:
+                  name: swarm-model-key
+                  key: OPENAI_API_KEY
+                  optional: true
+          resources:
+            requests: { memory: "256Mi", cpu: "200m" }
+            limits: { memory: "512Mi", cpu: "500m" }
+          livenessProbe:
+            httpGet: { path: /health, port: 8000 }
+            initialDelaySeconds: 20
+            periodSeconds: 10
+            timeoutSeconds: 5
+            failureThreshold: 3
+          readinessProbe:
+            httpGet: { path: /health, port: 8000 }
+            initialDelaySeconds: 8
+            periodSeconds: 5
+            timeoutSeconds: 3
+            failureThreshold: 3
@@ -492,6 +492,10 @@ def plan_launch_specs(
            env["HEICODE_USER_ID"] = user_id
        if git_env:
            env.update(git_env)
+        # LOCAL TEST ONLY (not for upstream): propagate the agent-side subtask-handoff switch so
+        # launched agents decompose + hand off subtasks to peers (else each agent does its task alone).
+        if os.getenv("ENABLE_SUBTASK_HANDOFF", "false").lower() in {"1", "true", "yes"}:
+            env["ENABLE_SUBTASK_HANDOFF"] = "true"
        specs.append(AgentLaunchSpec(agent_id=env["AGENT_ID"], capabilities=cap_csv, env=env))
    return specs