fix: wait for background goroutines during shutdown

[AmanGIT07]

Based on #1747 — merge that first; this base retargets to main after.

Summary

StartServer now waits for the UI server and the audit log listener to
finish shutting down before running deferred cleanup, instead of only
waiting for the connect server.

Changes

• The log listener, UI server, and connect server run under an
  errgroup; StartServer returns g.Wait(), so deferred cleanup
  (database and service Close() calls) runs only after all three
  finish.
• All three receive the group context, so a connect server startup
  failure also winds down the UI server and listener.
• A cancelled context is no longer logged as "log listener failed" on
  normal shutdown.

Technical Details

errgroup.WithContext cancels the group context when any member
returns an error. Only ServeConnect returns a non-nil error, so
g.Wait() returns exactly what StartServer returned before.

Test Plan

• go build, go vet, golangci-lint clean
• Manual: Ctrl-C logs both server "complete" lines before the first
  "cleaning up" line, with no listener warning
• No new unit test: StartServer needs live Postgres and SpiceDB before
  reaching this code; the composed pieces are tested in fix: wait for HTTP server drain during graceful shutdown #1745 and fix: add graceful shutdown to UI server #1747

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
frontier	Ready	Preview, Comment	Jul 14, 2026 8:14pm

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: b615a69a-5e8d-45b9-9dca-799486a2b72b

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coveralls]

Coverage Report for CI Build 29364880379

Coverage decreased (-0.02%) to 44.843%

Details

• Coverage decreased (-0.02%) from the base build.
• Patch coverage: 17 uncovered changes across 1 file (0 of 17 lines covered, 0.0%).
• 89 coverage regressions across 3 files.

Uncovered Changes

File	Changed	Covered	%
cmd/serve.go	17	0	0.0%

Coverage Regressions

89 previously-covered lines in 3 files lost coverage.

File	Lines Losing Coverage	Coverage
core/invitation/service.go	53	48.55%
cmd/serve.go	32	0.0%
internal/store/postgres/invitation_repository.go	4	80.3%

---

Coverage Stats

Relevant Lines:	37625
Covered Lines:	16872
Line Coverage:	44.84%
Coverage Strength:	12.51 hits per line

---

💛 - Coveralls

[rohilsurana]

Verified the whole chain: the listener returns ctx.Err() on cancel, #1747's ServeUI returns on every path, and ServeConnect's drain is bounded, so g.Wait() cannot hang for more than the grace period. One suggestion inline about signal handling during that window.

[rohilsurana]

Now that StartServer actually waits out the drain, a second Ctrl-C during that window does nothing. ctx comes from signal.NotifyContext (line 124), and its stop function only runs via the defer when StartServer returns — so while the drain runs (up to ~10s), further SIGINT/SIGTERM are swallowed by the registration. Before this PR the process exited almost immediately, so it never mattered.

Unregistering the handler once shutdown starts restores default signal handling, giving operators "first signal drains, second signal quits now":

Suggested change

	// Wait for every server to finish draining before the deferred cleanup
	// above closes the database and other dependencies under them.
	return g.Wait()
	// Once shutdown starts, unregister the signal handler so a second
	// SIGINT/SIGTERM gets default handling and can kill a stuck drain
	// instead of being swallowed.
	go func() {
	<-gctx.Done()
	cancelFunc()
	}()
	// Wait for every server to finish draining before the deferred cleanup
	// above closes the database and other dependencies under them.
	return g.Wait()

cancelFunc doubles as the stop function, and by the time gctx is done the context is already cancelled (either by a signal or by a member error), so calling it here only unregisters the handler.

[AmanGIT07]

Good catch — the swallow-window only became real once the drain actually waits. Applied as suggested in 9a90116: the handler unregisters when shutdown starts, so the first signal drains and a second one quits immediately.