Skip to content

fix(health): drop stale ollama check, add rerank/embed-mem checks - #15

Merged
SabaTech-dev merged 3 commits into
mainfrom
fix/stack-health-checks
Sep 9, 2026
Merged

fix(health): drop stale ollama check, add rerank/embed-mem checks#15
SabaTech-dev merged 3 commits into
mainfrom
fix/stack-health-checks

Conversation

@SabaTech-dev

Copy link
Copy Markdown
Owner

Qué

  • Elimina el TCP check de ollama :11434 (retirado 2026-08-28) que mantenía /api/health en degraded permanente.
  • Añade checks TCP de llama.cpp-rerank :8005 y llama.cpp-embed-memory :8006, que son los servicios que lo reemplazaron.

Por qué

El build desplegado seguía sondeando ollama muerto. Este cambio ya está corriendo en prod (build 2026-09-04 16:45, reinicio verificado: healthy, 14 checks up) — este PR sincroniza el repo con lo desplegado.

Verificación

  • npx vitest run src/lib/stack-health.test.ts → 8/8
  • npm run build OK (build en prod 04-sep)
  • curl localhost:3000/api/health"status":"healthy"

Nota: tsc --noEmit tiene errores pre-existentes en archivos ajenos a este cambio (agent-os/tasks, agents/status tests); el build pasa por ignoreBuildErrors.

Alfred and others added 3 commits August 26, 2026 04:21
Docker/GHCR require lowercase repository names. github.repository
(SabaTech-dev/Alfred-Mission-Control) broke 'docker buildx build' with
'invalid tag: repository name must be lowercase', failing the SLSA
Build & Attest job on every push to main.

Introduce workflow-level env.IMAGE_NAME (canonical lowercase artifact
name) and use it consistently for: build tags, attestation subject-name
and the verification summary (which also referenced a never-pushed
'sha-' prefixed tag).
…memory TCP checks

Ollama was retired on 2026-08-28; the running build still probed it and
kept the dashboard permanently degraded. This adds TCP checks for the
llama.cpp reranker (:8005) and memory embeddings (:8006) that replaced it.
…ards 64146d33/64df4def/1e20061c)

- Remove browserless :3002 probe (service uninstalled 2026-09-04) — phantom degraded source
- llama.cpp-gpu :8001 (Ornith) reported as held (intentional stop Joker 2026-09-08),
  never down; flips to up automatically if the port returns. Serving prod = 1Cat-vLLM :8009.
- Add held status end-to-end: StackServiceCheck type, summarizeStackHealth (held != degraded),
  formatStackHeartbeat symbols (up ✅ / held ⏸️ / down ❌), dashboard badge (amber) + counter
- Category map: drop browserless, add llama.cpp-embed-memory to llm
- NOTE doc: ollama removed 28-ago, browserless uninstalled 04-sep, :8001 HELD
- Tests: 12 (held semantics + absence guarantees for ollama/browserless)
@SabaTech-dev
SabaTech-dev merged commit 39d908d into main Sep 9, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant