Skip to content

fix(rpc): configure ClickHouse verification timeouts - #93

Draft
notwedtm wants to merge 1 commit into
mainfrom
notwedtm/clickhouse-verification-timeouts
Draft

notwedtm wants to merge 1 commit into
mainfrom
notwedtm/clickhouse-verification-timeouts

Conversation

@notwedtm

@notwedtm notwedtm commented Sep 17, 2026

Copy link
Copy Markdown
Contributor
AI description, click for details

ClickHouse cancellation preflight and runtime termination probes use a fixed one-second deadline, which can reject slow responses and prevent RPC startup. Add independent startup and runtime CLI/environment settings, both defaulting to 10,000 ms. No Linear ticket was supplied.

Startup applies per macro-resolution, discovery, and capability query, including retries. Runtime applies per abandoned-query batch. Both propagate through primary, shard, cache, cloned and background clients. Client deadlines preserve milliseconds; both server execution limits round up to whole seconds. Positive values are required; CLI overrides env. Normal query/connect timeouts, polling/backoff delays, permit retention and two complete absence observations remain unchanged. Diagnostics report effective budgets and phase without SDK/server payloads. Docs explain capacity retention and why the five-second unconfirmed warning can occur after a slow probe finishes.

Validation: formatting, workspace tests, all-feature RPC tests (593 passed), workspace and all-feature Clippy with warnings denied, isolated HTTP log test, and real ClickHouse cache integration passed. Delayed HTTP fixtures cover each startup phase/retry, independent budgets, rounding, runtime timeout/evidence reset, admission recovery, and warning evaluation after five seconds. Basic k6 passed against the candidate and a disposable local ClickHouse fixture: 2,334 successful requests in 10 seconds with one VU, zero HTTP/RPC errors, 25 signatures per response, p95 4.27 ms. This is a smoke check, not AP performance evidence.

Rollout: Binaries-ci run 35241515451 successfully built b5915631740ae9a0b0b983354dca43d397ab4f5f as v0.6.1-verification.b5915631. Per the approved rollout update, hold staging and update existing AP infrastructure PR rpcpool/terraform#3455 after artifact verification. No v0.6.0 artifact replacement; EU/US production upgrades are outside scope. The published AMD64 archive checksum and Cosign signature pass; its startup log reports the exact source SHA and both 10000 ms defaults. The published binary passed another local k6 smoke: 3,231 successful nonempty requests, zero errors. Existing AP PR #3455 now selects this artifact; Atlantis and policy checks pass with one AP job update and no staging/EU/US changes. Production apply awaits renewed PR approval (REVIEW_REQUIRED); live AP verification follows apply. All source CI checks pass.

🤖 Generated with AI

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