Taskfile.yml

# Taskfile.yml
version: '3'

vars:
  NAMESPACE: ms
  QDRANT_COLLECTION: homel_memory
  POSTGRES_POD:
    sh: kubectl get pod -n {{.NAMESPACE}} -l app=postgres -o jsonpath='{.items[0].metadata.name}' || echo ""
  INGRESS_HTTP_PORT: '30080'
  INGRESS_HTTPS_PORT: '30443'
  BACKUP_DIR: './backups'
  # Tooling image used by every ephemeral ops Job (migrate/backup/restore/export).
  # Local minikube default (loaded by `task build`, no registry pull needed).
  # For GHCR set: ghcr.io/<owner>/<repo>/tooling:latest
  TOOLING_IMAGE: ghcr.io/homel-dev/memory-steward/tooling:latest

tasks:
  default:
    desc: List all tasks
    cmds:
      - task --list

  # ============================================================================
  # 🚀 LIFECYCLE
  # ============================================================================
  up:
    desc: "Full Bootstrap: Bootstrap -> Deploy -> Wait -> Init -> Verify"
    cmds:
      - minikube start --driver=docker --cpus=8 --memory=16384 --gpus all
      - task: k8s:deploy
      - task: ingress:setup
      - task: k8s:wait
      - task: db:init
      - task: verify:health

  down:
    desc: "Teardown: Delete Namespace (DELETES PVCs in the namespace too)"
    cmds:
      - kubectl delete namespace {{.NAMESPACE}} --ignore-not-found

  status:all:
    desc: "Status"
    cmds:
      - kubectl get pods -n {{.NAMESPACE}}
      - kubectl get svc -n {{.NAMESPACE}}

  nuke:
    desc: "🔥 DESTRUCTIVE: Delete this namespace and its Persistent Volumes"
    prompt: "This will destroy all memory and logs in namespace '{{.NAMESPACE}}'. Continue?"
    cmds:
      - kubectl delete namespace {{.NAMESPACE}} --ignore-not-found
      - kubectl delete pvc --all -n {{.NAMESPACE}} --ignore-not-found --force --grace-period=0

  # ============================================================================
  # 🐳 BUILD
  # ============================================================================
  build:
    desc: "Build images into Minikube (incl. tooling)"
    cmds:
      - minikube image build -t homel/embeddings:dev ./components/embeddings
      - minikube image build -t homel/memory-router:dev ./components/memory_router
      - minikube image build -t homel/memory-steward:dev ./components/memory_steward
      - minikube image build -t homel/memory-steward-mcp:dev ./components/memory_steward_mcp
      - minikube image build -t homel/memory-steward-list:dev ./components/memory_steward_list
      - minikube image build -t homel/tooling:dev -f ./components/tooling/Dockerfile .

  # ============================================================================
  # ☸️ KUBERNETES OPERATIONS
  # ============================================================================
  k8s:deploy:
    desc: "Apply manifests"
    cmds:
      - kubectl apply -f k8s/namespace.yaml
      - kubectl apply -f k8s/
      - kubectl apply -f k8s/tooling/
      - kubectl apply -f k8s/backup/
      - kubectl apply -f k8s/vector/
      - kubectl apply -f k8s/grafana/

  k8s:wait:
    desc: "Block until core services are Ready"
    cmds:
      - kubectl rollout status -n {{.NAMESPACE}} statefulset/postgres --timeout=120s
      - kubectl rollout status -n {{.NAMESPACE}} statefulset/qdrant --timeout=120s
      - kubectl rollout status -n {{.NAMESPACE}} deployment/memory-router --timeout=120s
      - kubectl rollout status -n {{.NAMESPACE}} deployment/memory-steward --timeout=120s

  k8s:restart:
    desc: "Restart all core services"
    cmds:
      - kubectl rollout restart -n {{.NAMESPACE}} deployment/memory-router
      - kubectl rollout restart -n {{.NAMESPACE}} deployment/memory-steward
      - kubectl rollout restart -n {{.NAMESPACE}} deployment/memory-steward-mcp
      - kubectl rollout restart -n {{.NAMESPACE}} deployment/memory-steward-list
      - kubectl rollout restart -n {{.NAMESPACE}} deployment/open-webui

  # ============================================================================
  # 💾 DATA & STATE (INIT) -- non-destructive
  # ============================================================================
  db:init:
    desc: "Idempotent init: migrations (tooling Job) + ensure Qdrant collection"
    cmds:
      - task: migrate:up
      - task: db:init-qdrant


  db:init-qdrant:
    internal: true
    cmds:
      - |
        set -e
        kubectl delete job tooling-qdrant-init -n {{.NAMESPACE}} --ignore-not-found >/dev/null 2>&1 || true
        cat <<EOF | kubectl apply -f - >/dev/null
        apiVersion: batch/v1
        kind: Job
        metadata: {name: tooling-qdrant-init, namespace: {{.NAMESPACE}}}
        spec:
          backoffLimit: 0
          ttlSecondsAfterFinished: 180
          template:
            spec:
              serviceAccountName: memory-tooling
              restartPolicy: Never
              containers:
                - name: qdrant-init
                  image: {{.TOOLING_IMAGE}}
                  imagePullPolicy: IfNotPresent
                  command: ["bash","/opt/tooling/init_qdrant.sh","ensure"]
                  env:
                    - {name: QDRANT_URL,        value: "http://qdrant:6333"}
                    - {name: QDRANT_COLLECTION, value: "{{.QDRANT_COLLECTION}}"}
        EOF
        kubectl wait --for=condition=complete job/tooling-qdrant-init -n {{.NAMESPACE}} --timeout=120s \
          || { kubectl logs job/tooling-qdrant-init -n {{.NAMESPACE}}; exit 1; }
        kubectl logs job/tooling-qdrant-init -n {{.NAMESPACE}}

  db:reset:
    desc: "🔥 DESTRUCTIVE: drop & rebuild Postgres schema + Qdrant collection"
    prompt: "This ERASES all Postgres tables and the Qdrant collection. Continue?"
    cmds:
      - |
        set -e
        kubectl delete job tooling-reset -n {{.NAMESPACE}} --ignore-not-found >/dev/null 2>&1 || true
        cat <<EOF | kubectl apply -f - >/dev/null
        apiVersion: batch/v1
        kind: Job
        metadata: {name: tooling-reset, namespace: {{.NAMESPACE}}}
        spec:
          backoffLimit: 0
          ttlSecondsAfterFinished: 180
          template:
            spec:
              serviceAccountName: memory-tooling
              restartPolicy: Never
              containers:
                - name: reset
                  image: {{.TOOLING_IMAGE}}
                  imagePullPolicy: IfNotPresent
                  command: ["sh","-c","psql -v ON_ERROR_STOP=1 -c 'DROP SCHEMA public CASCADE; CREATE SCHEMA public;' && bash /opt/tooling/init_qdrant.sh delete"]
                  env:
                    - {name: PGHOST,            value: "postgres"}
                    - {name: PGUSER,            value: "homel"}
                    - {name: PGPASSWORD,        value: "homel"}
                    - {name: PGDATABASE,        value: "homel"}
                    - {name: QDRANT_URL,        value: "http://qdrant:6333"}
                    - {name: QDRANT_COLLECTION, value: "{{.QDRANT_COLLECTION}}"}
        EOF
        kubectl wait --for=condition=complete job/tooling-reset -n {{.NAMESPACE}} --timeout=180s \
          || { kubectl logs job/tooling-reset -n {{.NAMESPACE}}; exit 1; }
        kubectl logs job/tooling-reset -n {{.NAMESPACE}}
      - task: db:init


  db:shell:
    desc: "Open PSQL shell"
    interactive: true
    cmds:
      - kubectl exec -it -n {{.NAMESPACE}} {{.POSTGRES_POD}} -- psql -U homel -d homel

  # ============================================================================
  # 🧬 MIGRATIONS (ephemeral tooling Job; tracked in schema_migrations)
  # ============================================================================
  migrate:up:
    desc: "Apply pending migrations via an ephemeral tooling Job"
    cmds:
      - |
        set -e
        kubectl delete job tooling-migrate -n {{.NAMESPACE}} --ignore-not-found >/dev/null 2>&1 || true
        cat <<EOF | kubectl apply -f - >/dev/null
        apiVersion: batch/v1
        kind: Job
        metadata: {name: tooling-migrate, namespace: {{.NAMESPACE}}}
        spec:
          backoffLimit: 0
          ttlSecondsAfterFinished: 300
          template:
            spec:
              serviceAccountName: memory-tooling
              restartPolicy: Never
              containers:
                - name: migrate
                  image: {{.TOOLING_IMAGE}}
                  imagePullPolicy: IfNotPresent
                  command: ["bash","/opt/tooling/migrate.sh","up"]
                  env:
                    - {name: PGHOST,     value: "postgres"}
                    - {name: PGUSER,     value: "homel"}
                    - {name: PGPASSWORD, value: "homel"}
                    - {name: PGDATABASE, value: "homel"}
        EOF
        kubectl wait --for=condition=complete job/tooling-migrate -n {{.NAMESPACE}} --timeout=180s \
          || { echo "--- logs ---"; kubectl logs job/tooling-migrate -n {{.NAMESPACE}}; exit 1; }
        kubectl logs job/tooling-migrate -n {{.NAMESPACE}}

  migrate:status:
    desc: "Show applied vs pending migrations (ephemeral tooling Job)"
    cmds:
      - |
        set -e
        kubectl delete job tooling-migstatus -n {{.NAMESPACE}} --ignore-not-found >/dev/null 2>&1 || true
        cat <<EOF | kubectl apply -f - >/dev/null
        apiVersion: batch/v1
        kind: Job
        metadata: {name: tooling-migstatus, namespace: {{.NAMESPACE}}}
        spec:
          backoffLimit: 0
          ttlSecondsAfterFinished: 120
          template:
            spec:
              serviceAccountName: memory-tooling
              restartPolicy: Never
              containers:
                - name: migstatus
                  image: {{.TOOLING_IMAGE}}
                  imagePullPolicy: IfNotPresent
                  command: ["bash","/opt/tooling/migrate.sh","status"]
                  env:
                    - {name: PGHOST,     value: "postgres"}
                    - {name: PGUSER,     value: "homel"}
                    - {name: PGPASSWORD, value: "homel"}
                    - {name: PGDATABASE, value: "homel"}
        EOF
        kubectl wait --for=condition=complete job/tooling-migstatus -n {{.NAMESPACE}} --timeout=120s >/dev/null
        kubectl logs job/tooling-migstatus -n {{.NAMESPACE}}

  # ============================================================================
  # 🗄️ BACKUP (ephemeral tooling Job -> backups PVC)
  # ============================================================================
  backup:
    desc: "Back up Postgres + Qdrant to the backups PVC"
    cmds:
      - |
        set -e
        kubectl delete job tooling-backup -n {{.NAMESPACE}} --ignore-not-found >/dev/null 2>&1 || true
        cat <<EOF | kubectl apply -f - >/dev/null
        apiVersion: batch/v1
        kind: Job
        metadata: {name: tooling-backup, namespace: {{.NAMESPACE}}}
        spec:
          backoffLimit: 0
          ttlSecondsAfterFinished: 600
          template:
            spec:
              serviceAccountName: memory-tooling
              restartPolicy: Never
              containers:
                - name: backup
                  image: {{.TOOLING_IMAGE}}
                  imagePullPolicy: IfNotPresent
                  command: ["bash","/opt/tooling/backup.sh"]
                  env:
                    - {name: PGHOST,            value: "postgres"}
                    - {name: PGUSER,            value: "homel"}
                    - {name: PGPASSWORD,        value: "homel"}
                    - {name: PGDATABASE,        value: "homel"}
                    - {name: QDRANT_URL,        value: "http://qdrant:6333"}
                    - {name: QDRANT_COLLECTION, value: "{{.QDRANT_COLLECTION}}"}
                  volumeMounts: [{name: backups, mountPath: /backups}]
              volumes:
                - name: backups
                  persistentVolumeClaim: {claimName: backups}
        EOF
        kubectl wait --for=condition=complete job/tooling-backup -n {{.NAMESPACE}} --timeout=600s \
          || { kubectl logs job/tooling-backup -n {{.NAMESPACE}}; exit 1; }
        kubectl logs job/tooling-backup -n {{.NAMESPACE}}

  # ============================================================================
  # ♻️ RESTORE (ephemeral tooling Job from backups PVC, then forward-migrate)
  # ============================================================================
  restore:
    desc: "DR restore. Usage: task restore DUMP=<f.dump> [SNAPSHOT=<f.snapshot>]"
    cmds:
      - |
        set -e
        [ -n "{{.DUMP}}" ] || { echo "usage: task restore DUMP=<pgfile.dump> [SNAPSHOT=<file.snapshot>]"; exit 1; }
        kubectl delete job tooling-restore -n {{.NAMESPACE}} --ignore-not-found >/dev/null 2>&1 || true
        cat <<EOF | kubectl apply -f - >/dev/null
        apiVersion: batch/v1
        kind: Job
        metadata: {name: tooling-restore, namespace: {{.NAMESPACE}}}
        spec:
          backoffLimit: 0
          ttlSecondsAfterFinished: 600
          template:
            spec:
              serviceAccountName: memory-tooling
              restartPolicy: Never
              containers:
                - name: restore
                  image: {{.TOOLING_IMAGE}}
                  imagePullPolicy: IfNotPresent
                  command: ["bash","/opt/tooling/restore.sh"]
                  env:
                    - {name: PGHOST,            value: "postgres"}
                    - {name: PGUSER,            value: "homel"}
                    - {name: PGPASSWORD,        value: "homel"}
                    - {name: PGDATABASE,        value: "homel"}
                    - {name: QDRANT_URL,        value: "http://qdrant:6333"}
                    - {name: QDRANT_COLLECTION, value: "{{.QDRANT_COLLECTION}}"}
                    - {name: DUMP_FILE,         value: "{{.DUMP}}"}
                    - {name: SNAPSHOT_FILE,     value: "{{.SNAPSHOT}}"}
                  volumeMounts: [{name: backups, mountPath: /backups}]
              volumes:
                - name: backups
                  persistentVolumeClaim: {claimName: backups}
        EOF
        kubectl wait --for=condition=complete job/tooling-restore -n {{.NAMESPACE}} --timeout=600s \
          || { kubectl logs job/tooling-restore -n {{.NAMESPACE}}; exit 1; }
        kubectl logs job/tooling-restore -n {{.NAMESPACE}}
      - task: migrate:up
      - echo "✅ Restore + forward-migrate complete. Run 'task migrate:status' to confirm head."

  # ============================================================================
  # 📤 EXPORT (ephemeral tooling Job -> backups PVC + stdout)
  # ============================================================================
  export:memory:
    desc: "Export memory dataset. e.g. task export:memory -- --level rich --since 30"
    cmds:
      - |
        set -e
        ARGS="{{.CLI_ARGS}}"; [ -n "$ARGS" ] || ARGS="--level standard"
        kubectl delete job tooling-export -n {{.NAMESPACE}} --ignore-not-found >/dev/null 2>&1 || true
        cat <<EOF | kubectl apply -f - >/dev/null
        apiVersion: batch/v1
        kind: Job
        metadata: {name: tooling-export, namespace: {{.NAMESPACE}}}
        spec:
          backoffLimit: 0
          ttlSecondsAfterFinished: 600
          template:
            spec:
              serviceAccountName: memory-tooling
              restartPolicy: Never
              containers:
                - name: export
                  image: {{.TOOLING_IMAGE}}
                  imagePullPolicy: IfNotPresent
                  command: ["sh","-c","mkdir -p /backups/exports && python3 /opt/tooling/memory_export.py $ARGS | tee /backups/exports/memory-\$(date +%Y%m%d-%H%M%S).json"]
                  env:
                    - {name: POSTGRES_USER,     value: "homel"}
                    - {name: POSTGRES_PASSWORD, value: "homel"}
                    - {name: POSTGRES_DB,       value: "homel"}
                    - {name: EXPORT_PG_HOST,    value: "postgres"}
                    - {name: QDRANT_URL,        value: "http://qdrant:6333"}
                    - {name: QDRANT_COLLECTION, value: "{{.QDRANT_COLLECTION}}"}
                  volumeMounts: [{name: backups, mountPath: /backups}]
              volumes:
                - name: backups
                  persistentVolumeClaim: {claimName: backups}
        EOF
        kubectl wait --for=condition=complete job/tooling-export -n {{.NAMESPACE}} --timeout=600s \
          || { kubectl logs job/tooling-export -n {{.NAMESPACE}}; exit 1; }
        echo "✅ Export written to backups PVC at /backups/exports/. Pull locally with: task export:fetch"

  export:fetch:
    desc: "Copy the newest export from the backups PVC to {{.BACKUP_DIR}}/exports locally"
    cmds:
      - |
        set -e
        mkdir -p {{.BACKUP_DIR}}/exports
        kubectl delete job tooling-fetch -n {{.NAMESPACE}} --ignore-not-found >/dev/null 2>&1 || true
        cat <<EOF | kubectl apply -f - >/dev/null
        apiVersion: batch/v1
        kind: Job
        metadata: {name: tooling-fetch, namespace: {{.NAMESPACE}}}
        spec:
          backoffLimit: 0
          ttlSecondsAfterFinished: 120
          template:
            spec:
              serviceAccountName: memory-tooling
              restartPolicy: Never
              containers:
                - name: fetch
                  image: {{.TOOLING_IMAGE}}
                  imagePullPolicy: IfNotPresent
                  command: ["sh","-c","cat \$(ls -1t /backups/exports/*.json | head -1)"]
                  volumeMounts: [{name: backups, mountPath: /backups}]
              volumes:
                - name: backups
                  persistentVolumeClaim: {claimName: backups}
        EOF
        kubectl wait --for=condition=complete job/tooling-fetch -n {{.NAMESPACE}} --timeout=120s >/dev/null
        OUT="{{.BACKUP_DIR}}/exports/memory-export-$(date +%Y%m%d-%H%M%S).json"
        kubectl logs job/tooling-fetch -n {{.NAMESPACE}} > "$OUT"
        echo "✅ $OUT"

  # ============================================================================
  # ✅ VERIFICATION
  # ============================================================================
  verify:health:
    desc: "Check healthz endpoints"
    cmds:
      - kubectl exec -n {{.NAMESPACE}} deploy/memory-router -- curl -sSf http://localhost:8080/healthz
      - kubectl exec -n {{.NAMESPACE}} deploy/memory-router -- curl -sSf http://memory-steward:8090/healthz

  verify:amp:
    desc: "Smoke-test AMP artifact write + exact Router retrieval (no LLM required)"
    cmds:
      - |
        kubectl exec -i -n {{.NAMESPACE}} deploy/memory-router -- python - <<'PY'
        import hashlib
        import json
        import requests

        project_id = "amp-smoke"
        payload = {"smoke": "amp", "version": 1}
        canonical = json.dumps(payload, sort_keys=True, separators=(",", ":"), ensure_ascii=False)
        content_hash = hashlib.sha256(canonical.encode("utf-8")).hexdigest()

        outcome = requests.post(
            "http://memory-steward:8090/v1/agent/outcomes",
            json={
                "outcome_id": "amp-smoke-v1",
                "project_id": project_id,
                "objective": "AMP deterministic artifact smoke test",
                "result": "artifact generated",
                "admit_knowledge": False,
                "repository_state": {"repository": "amp-smoke", "revision": "smoke-v1"},
                "artifacts": [{
                    "artifact_type": "repository_ir",
                    "schema_version": "1",
                    "producer_type": "analyzer",
                    "producer_name": "amp-smoke",
                    "producer_version": "1",
                    "content_hash": content_hash,
                    "payload": payload,
                    "provenance": [{"kind": "smoke_test"}],
                }],
            },
            timeout=30,
        )
        outcome.raise_for_status()

        retrieved = requests.post(
            "http://localhost:8080/v1/context/retrieve",
            headers={"X-Project-ID": project_id},
            json={
                "artifact_selectors": [{
                    "artifact_type": "repository_ir",
                    "repository": "amp-smoke",
                    "revision": "smoke-v1",
                }]
            },
            timeout=30,
        )
        retrieved.raise_for_status()
        body = retrieved.json()
        artifacts = body.get("agent_reference", [])
        assert artifacts, body
        assert artifacts[0]["payload"] == payload, body
        assert artifacts[0]["content_hash"] == content_hash, body
        print(json.dumps({
            "ok": True,
            "outcome": outcome.json(),
            "context_request_id": body["context_request_id"],
            "artifact_id": artifacts[0]["id"],
        }, indent=2))
        PY

  logs:router:
    desc: "Tail Router Logs"
    cmds:
      - kubectl logs -n {{.NAMESPACE}} -l app=memory-router -f

  logs:steward:
    desc: "Tail Steward Logs"
    cmds:
      - kubectl logs -n {{.NAMESPACE}} -l app=memory-steward -f