Expose operational health and backpressure #25

New issue

Open

opened 2026-07-14 22:32:29 -04:00 by caleb-brown · 0 comments

caleb-brown commented

2026-07-14 22:32:29 -04:00

Owner

What to build

Service operators can monitor Temper's health, capacity, dependencies, and queue behavior without exposing repository content or relying on database inspection.

Acceptance criteria

Structured logs correlate delivery, request, attempt, result, and publication identifiers without raw source, prompts, secrets, or model prose.
Metrics report queue age, state counts, lease recovery, retries, latency, provider usage, invalid output, stale suppression, and publication failures.
Liveness and readiness retain distinct semantics and do not become unavailable merely because Forgejo or the model provider is temporarily unhealthy.
Queue and concurrency limits apply backpressure with explicit Operator-visible degradation rather than unbounded growth.
The console shows dependency health and the appropriate retry, reconnect, or capacity action.
A documented metrics endpoint and diagnostic procedure work in the supported Compose topology.
Load and outage tests verify backpressure, content-safe observability, and recovery.

Blocked by

#18 — Keep feedback correct across retries and revisions
#20 — Diagnose and recover Review Runs

## What to build Service operators can monitor Temper's health, capacity, dependencies, and queue behavior without exposing repository content or relying on database inspection. ## Acceptance criteria - [ ] Structured logs correlate delivery, request, attempt, result, and publication identifiers without raw source, prompts, secrets, or model prose. - [ ] Metrics report queue age, state counts, lease recovery, retries, latency, provider usage, invalid output, stale suppression, and publication failures. - [ ] Liveness and readiness retain distinct semantics and do not become unavailable merely because Forgejo or the model provider is temporarily unhealthy. - [ ] Queue and concurrency limits apply backpressure with explicit Operator-visible degradation rather than unbounded growth. - [ ] The console shows dependency health and the appropriate retry, reconnect, or capacity action. - [ ] A documented metrics endpoint and diagnostic procedure work in the supported Compose topology. - [ ] Load and outage tests verify backpressure, content-safe observability, and recovery. ## Blocked by - [#18](https://git.caleb-brown.dev/caleb-brown/temper/issues/18) — Keep feedback correct across retries and revisions - [#20](https://git.caleb-brown.dev/caleb-brown/temper/issues/20) — Diagnose and recover Review Runs

caleb-brown added the

ready-for-agent

label

2026-07-14 22:32:29 -04:00

caleb-brown referenced this issue

2026-07-14 22:32:31 -04:00

Package a recoverable production release #27

caleb-brown added a new dependency

2026-07-14 22:38:28 -04:00

#18 Keep feedback correct across retries and revisions

caleb-brown added a new dependency

2026-07-14 22:38:29 -04:00