test: make changelog contract robust to rollover; bump ubuntu restart budget - #309
Merged
StefanSteiner merged 2 commits intoSep 7, 2026
Conversation
Fix A: smoke_demo_and_changelog_contract asserted KV/export/routing claims only against the `## [Unreleased]` slice of the embedded mcp CHANGELOG. PR tableau#307 (the rc.3 changelog rollover) moved every bullet into `## [1.0.0-rc.3]`, emptying Unreleased and failing the test. Because tableau#307 was docs-only, CI's `paths-ignore: **/*.md` skipped the Rust suite, so the break surfaced only on a later code PR. Assert instead against the current release window (`## [Unreleased]` + the most-recent dated section) via a new `current_release_window` helper, so a future rollover can't silently rebreak it. The window is bounded to the newest dated section so a stale token in an ancient entry can't satisfy a deleted-claim check. Fix B: engine_recovers_after_hyperd_killed and hyperd_monitor_detects_killed_hyperd_and_restarts time out on ubuntu-latest under CI load — a 5s monitor tick plus cold hyperd spawn against a 12s readiness budget. Replace the three magic `12`s with a named RESTART_READINESS_BUDGET_SECS = 45 and document the derivation. Budget bump only; the macOS-only ignore is unchanged, so Linux/Windows crash-recovery coverage from tableau#279/tableau#286 stays. See tableau#305.
The constant's only use sites are the three `#[cfg(unix)]` restart tests and their Unix-only helpers, so on Windows it compiled but was never referenced and `clippy -D warnings` (windows-latest) rejected it as dead code. Gate the definition with `#[cfg(unix)]` so it exists exactly where it's used — the honest fix, not `#[allow(dead_code)]`.
StefanSteiner
added a commit
to StefanSteiner/hyper-api-rust
that referenced
this pull request
Sep 7, 2026
…query The two Linux-only daemon restart failures in tableau#310 are a test-harness bug, not a regression in the restart path. `find_hyperd_pid_for_endpoint` selected the socket owner with lsof -nP -t -c hyperd <socket path> `lsof` ORs its list-selection options unless `-a` is given, so that query returns every process whose command begins with `hyperd` *in addition to* the socket's owner, and `-t` prints PIDs lowest-first. Whenever more than one hyperd is alive -- routine in CI, where each test starts its own daemon while the previous test's engine is still shutting down -- the helper returns the oldest hyperd rather than this endpoint's, and the test kills a process it does not own. Its own hyperd stays up, so phase 1 of `wait_for_live_hyperd_after_kill` spins until the budget expires: a deterministic full-budget timeout, which is why raising 12s to 45s in tableau#309 did not help. Over TCP the equivalent selection (`-iTCP:<port> -sTCP:LISTEN`) names exactly one listener, which is why these tests were green before the transport moved to a socket path shared across restarts. Identify the engine from hyperd's own lock file instead. hyperd writes `<socket path>.pid` (bare decimal PID, O_EXCL) before it binds and unlinks it on clean exit, so it names the one process serving this endpoint. A leftover file can name a dead or recycled PID, so confirm the PID still is a hyperd before returning it. Also make the readiness helper report which phase expired and what it last observed. A bare "didn't restart in time" is what kept this undiagnosed across two CI runs. Refs tableau#310
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two main-side test fixes that currently make
main's full test suite red. Both are test-only (hyperdb-mcp/tests/); no product code, versions, release-please files, or workflows are touched.Fix A —
smoke_demo_and_changelog_contractregression from #307smoke_demo_and_changelog_contractasserted the release's KV/export/routing claims only against the## [Unreleased]slice of the embedded mcpCHANGELOG. PR #307 (the rc.3 changelog rollover) moved every bullet out of## [Unreleased]into## [1.0.0-rc.3], emptying the Unreleased slice and failing the token/heading/hyper-format exportassertions.This is a real regression from #307, surfaced only later: #307 was docs-only, so CI's
paths-ignore: **/*.mdskipped the Rust suite for it, and the break didn't appear until a code PR ran the tests.Approach — assert against the current release window (union of
## [Unreleased]+ the most-recent dated section), not the whole changelog. A newcurrent_release_windowhelper concatenates the## [Unreleased]slice with the single newest dated## [x.y.z]section. Before a release the claims live under Unreleased; after a rollover they live under the newest dated section; the window spans both, so a futuredocs:rollover can't silently rebreak the contract. I deliberately did not search the whole changelog: tokens likedoctor/health portcould plausibly appear in older historical entries (## [0.5.0],## [0.1.1]), which would let a deleted claim still pass and weaken the check. Bounding to the newest dated section preserves the "these claims are documented for the current release candidate" intent while excluding stale history. Every existing assertion's intent is preserved; no check was weakened or removed.Red-before-green (
cargo test -p hyperdb-mcp --test readme_tests):must contain at least one bullet under ### added/fixed/changed,does not account for "doctor"/"resolved_database"/"resource_busy"/"health port"/"engine_busy"/"bar_orientation"/"label_values"/"show_legend"/"y_scale").test result: ok. 9 passed; 0 failed.Fix B — ubuntu restart tests time out under CI load
engine_recovers_after_hyperd_killedandhyperd_monitor_detects_killed_hyperd_and_restartstime out onubuntu-latestunder CI load: the liveness monitor polls on a 5 s tick (daemon::run::HYPERD_POLL_INTERVAL), so a kill costs up to ~5 s just to be noticed, then recovery must drop the dead process, cold-spawn a freshhyperd, rebind, and republish STATUS. The old 12 s budget (~5 s detection + ~7 s slack) is too tight when a shared runner's cold spawn eats several seconds. They pass 4/4 locally and the third restart test passes on the same CI run, so this is a false timeout, and it predates the #304 IPC work.This is a budget bump, not a coverage change. The three magic
12s are replaced with a documented named constantRESTART_READINESS_BUDGET_SECS = 45(~40 s headroom past the detection tick, still bounding a genuinely wedged daemon). The macOS-only#[cfg_attr(target_os = "macos", ignore)]is unchanged — it is not extended to Linux, so the Linux crash-recovery coverage added by #279 (and depended on by #286's fix) stays intact on Linux/Windows.Cross-references issue #305 (the standing daemon-test timing decision for the macOS-ignored set); this timing fragility is now evidence the budget question spans Linux too. Related flakes: #300, #288.
Local
--ignoredrun (macOS host, these are macOS-ignored):16 s locally sits well under 45 s; the extra headroom absorbs a loaded runner's slower detection + cold spawn without ever masking a real hang.
Gate (all green, real output)
cargo fmt --all -- --check→ exit 0cargo clippy --workspace --all-targets --all-features -- -D warnings→ exit 0, 0 warningscargo test -p hyperdb-mcp→ 663 passed; 0 failed; 17 ignored (the previously-failing contract test now passes)RUSTDOCFLAGS="-D warnings" cargo doc --no-deps→ exit 0, 0 warningsnpx markdownlint-cli2→ 0 issues in 68 filesNotes
paths-ignore: **/*.md, which is how docs: roll over per-crate changelogs for 1.0.0-rc.3 #307 broke a Rust contract test undetected. Fix A mitigates the symptom for changelog rollovers, but the general gap (a.mdchange breaking a Rust test) remains — left for a separate decision (e.g. exempt CHANGELOG paths from the ignore, or run the contract test on docs changes). No workflow files touched here.