e54dbe9d5d3739e0d368c4a5adb2c1cd5f908ac4
2
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 not
|
e54dbe9d5d |
fix(observability): stop Tempo evicting its only ingester under load (refs #156)
`verify-tracing` flaked on run 722: `FAIL — no single trace spanned ['bff',
'projection-api']`, green on a plain re-run of the same commit. Not a broken trace
chain — Tempo could not ingest:
removing distributor_pool failing healthcheck addr=127.0.0.1:9095
reason="rpc error: code = DeadlineExceeded"
pusher failed to consume trace data err="context canceled" (x18)
Root cause is the mechanism of the data loss, not whatever caused the stall. Tempo
runs single-binary, so distributor and ingester are the *same process* and the
distributor's ingester pool holds exactly one, in-process, member. dskit still
health-checks it over loopback gRPC with a 1s deadline; on the shared runner a
transient stall blows that, the only ingester is dropped from the pool, and every
push fails until the next 15s check interval — spans silently lost. With one
in-process ingester the check can never route around a failure, so it can only ever
discard data.
Fix: `ingester_client.pool_config.healthcheckenabled: false`. This addresses both
candidate triggers (GC pressure near `mem_limit`, CPU contention) at the point where
they turn into lost data, so `mem_limit: 400m` stays untouched — raising it would
risk reintroducing the verify-e2e OOM of #144 on a memory-tight runner.
Also print `tempo_distributor_ingester_clients` on the check's failure path: a
recurrence then names Tempo-dropped-spans instead of costing another container-log
dive, since from the check's side that is indistinguishable from missing
instrumentation.
Verified against the built image: config parses (`-config.verify`), the effective
`/status/config` reports `healthcheckenabled: false`, and the diagnostic reads the
metric off a live Tempo.
|
||
|
not
|
88338396f6 |
feat(obs): distributed traces across the .NET services (S-16b, closes #123) (#126)
## What & why S-16b, second of the S-16 split, on top of the #125 backplane. The five .NET services now emit OpenTelemetry traces so a request is **one connected trace** across them. - Each host wires `AddOpenTelemetry().WithTracing(...)` with `AddAspNetCoreInstrumentation` (incoming) + `AddHttpClientInstrumentation` (outgoing) + `AddOtlpExporter` to **Tempo**. - Because every cross-service call already goes through a typed `HttpClient` (§8 boundaries), the W3C `traceparent` propagates with no manual code — bff → domain → acl → openzaak and bff → projection-api stitch into a single trace. - Service name + OTLP endpoint come from `OTEL_*` env set per app service in compose. `/health` is filtered out so liveness polls don't flood the traces. No new ADR — ADR-0023 already records the stack + the two documented gaps (browser-side tracing is out of scope, so the trace begins at the BFF; the async Flowable-poll boundary is a separate trace). Closes #123 ## Definition of Done - [x] Failing test committed first (`verify-tracing` fails with no instrumentation). - [x] Implementation makes it pass — **validated locally end to end**: a real connected trace spanning `bff` + `projection-api` was found in Tempo (BFF→projection→db + Tempo subset, no OpenZaak/egress). - [x] Conventional Commits referencing the issue (`refs #123`). - [ ] CI green — awaiting Gitea Actions (verify-tracing added to verify-stack after verify-bff). - [x] `docker compose up` health unaffected — services boot healthy even when Tempo is unreachable (exporter no-ops; verified). - [x] Docs — demo-script + BACKLOG. - [x] ADR — none needed (covered by ADR-0023). ## Notes for reviewers - **Per-service wiring, no shared lib:** the block is duplicated across the five hosts by design — services don't share code across boundaries here (§8), same as the duplicated typed clients. - **Packages:** OpenTelemetry.Extensions.Hosting / Instrumentation.AspNetCore / Instrumentation.Http / Exporter.OpenTelemetryProtocol, all 1.17.0, pinned per-csproj (no central props file). - **The check** generates anonymous BFF→projection traffic (no auth, no OpenZaak), then queries Tempo (TraceQL search → fetch trace → assert both service.names present) from a python:3-slim container in-network — same idiom as run-projection-check.sh. - **Next:** #124 (S-16c) adds `/metrics` + Prometheus scrape targets + golden-signal Grafana dashboards. Reviewed-on: #126 |