Repository navigation
fix(dev-tools): log codegraph startup sync as skipped when the daemon owns the index - #1340
Merged
Merged
Conversation
… owns the index The every-boot codegraph hook races the editor's MCP server: the server starts within a second or two of the hook, takes the index lock, and the detached `codegraph sync` fails with "database is locked". The log then reported a failure on most boots even though the daemon's file watcher keeps the graph current. After a failed op, the hook now checks daemon.pid for a live process whose command line names codegraph and logs a skip instead. The check runs only after the failure because the daemon usually appears after the hook has started. A missing pid file, or a pid that is running but not codegraph (stale file, recycled pid), is still logged as a failure. The cache dir, binary and log path can be overridden so the new behavioral test can drive the hook with a stub codegraph. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ty lessons Adds eleven memories from recent golem and review cycles (harness resume replaying a fake review, full review before merge, follow-ups before teardown, teardown tripping on scaffolding, merge commits breaking conform CI, readdir-order CI-only failures, refusal tests reaching real gh, mutation scratch target dirs, env scrub under the same uid, evidence pending from published db, heuristic scanners), updates three existing ones, and re-indexes MEMORY.md. Long lines are wrapped to satisfy rumdl MD013. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rve daemon
Review of the previous commit found two ways it logged a real failure as
"skipped":
- Any failure counted while a daemon was alive, which is the normal state
seconds after boot. A corrupt index, ENOSPC, or a CLI change would be
reported as "owns the index and keeps it synced".
- The `*codegraph*` argv match accepted any process that mentions
codegraph, e.g. `less /tmp/codegraph-index.log` on a recycled pid.
The hook now takes the PID from codegraph's own lock refusal ("locked by
another process (PID N)") rather than daemon.pid, so it names the actual
lock holder, and it requires that PID's argv to be a `codegraph ... serve`
process. Reading /proc/PID/cmdline replaces `kill -0`, which also stops
a daemon owned by another user from failing the check with EPERM.
The test now covers a lock held by the serve daemon (skip), a non-lock
error that mentions the daemon's PID, a holder whose argv merely mentions
codegraph, an unrelated holder, and a dead holder (all failures). A second
test covers init completion and the missing-binary no-op. Mutating each
condition in a scratch copy fails the suite.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The fake-daemon fixtures from the previous commit ran `bash -c 'sleep 30; :' ... &` and killed only the bash parents, orphaning each `sleep 30` with the suite's stdout still open. run_unit_tests.sh captures every suite via $(...), so it blocked about 30s on this suite in `just test`, pre-commit, and CI. The fixtures now detach from the suite's stdio and their sleep children are killed before the parents. The captured run drops from 31s to 2s. The daemon match also tightens from `*codegraph*" serve"*` to `*codegraph.js" serve"*`, the argv codegraph spawns its daemon with (direct or via the launcher), so `grep codegraph serve` and similar no longer match. The reader fixture now uses that shape. The hook comment notes that only `sync` can log a skip: a contended `init` and the lock-file race report the lock without a PID. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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.
Summary
On a fresh boot,
/tmp/codegraph-index.logreported:The every-boot hook (
55-codegraph-index.sh) starts a detachedcodegraph sync. The editor's codegraph MCP server starts a second or two later, takes the index lock, and the sync loses. Nothing is actually wrong, because the daemon's file watcher keeps the graph current, but the log reports a failure on most boots.Change
sync/init, the hook readsdaemon.pid(JSON) and checks for a live process whose command line namescodegraph. If one is found, it logssync skipped: codegraph daemon (PID N) owns the index and keeps it synced.CODEGRAPH_CACHE_DIR/CODEGRAPH_BIN/CODEGRAPH_LOGcan now be overridden so the hook can be tested. Defaults are unchanged.The second commit (
chore(memory)) adds the pending.claude/memory/updates: 11 new memories, 3 updated, and the index. Long lines are wrapped to pass rumdl, and I checked that only whitespace changed.Testing
test_codegraph_index_skips_when_daemon_owns_indexruns the real hook with a stubcodegraphthat always fails. It checks three cases: a live daemon (skip), no daemon (failure), and a running pid that isn't codegraph (failure).tests/unit/features/dev-tools.sh: 76/76 pass. shellcheck at warning level and shfmt are clean.sync complete.🤖 Generated with Claude Code