[avallete]

TL;DR

The Realtime, pg-meta and Studio catalog test hung on Podman without saying where. It now fails before the test timeout, naming the service and phase that stalled and that service's output.

• Realtime.integration.test.ts › serves Realtime, pg-meta, and Studio against the owned database: on Podman, the test hit the 120s test timeout with only the generator start frame as a clue. The test runs four services in sequence (database, Realtime, pg-meta, Studio), each with a recipe, start, readiness and HTTP step, so a stall anywhere looked the same. The cause is still unknown, so this change is a diagnostic. The scoped body now runs under a 110s guard. Each step records its phase, and each recipe's launch output is collected. A stall fails with CatalogStallError naming the service and phase in progress (including stop and root cleanup) and the last part of that service's output.

ref

spotted in:

• https://github.com/supabase/cli/actions/runs/40530528881/job/122281757547

---

Generated by Claude Code

[avallete]

/ai-review

---

Generated by Claude Code

[github-actions]

Superseded by a newer AI review

🤖 AI Review

Verified all four reported findings and merged the duplicate timeout reports. Confirmed one minor cleanup-diagnostic issue and two diagnostic nits. Updated stale line references to match the checked-out code.

Findings

Severity	Location	Category	Sources	Claim
🟡 MINOR	packages/stack/src/services/Realtime.integration.test.ts:211	test-reliability	claude+codex	The timeout guard waits for scoped finalization, so a hanging teardown can prevent CatalogStallError from reaching the test runner before its timeout.
⚪ NIT	packages/stack/src/services/Realtime.integration.test.ts:209	diagnostics	claude	Some phase labels identify a narrower operation than the work actually running, which can misdirect stall diagnosis.
⚪ NIT	packages/stack/src/services/Realtime.integration.test.ts:43	correctness	claude	Decoding each output chunk independently can corrupt split UTF-8 characters, and trimming by UTF-16 code units can leave a dangling surrogate.

Stats

Claude findings: 3 · Codex findings: 1 · Confirmed: 3 · Refuted: 0 · Uncertain: 0

---

Models: claude-opus-5-5 + gpt-6.1-sol · Trigger: manual · Workflow run

This review runs once per PR. A maintainer can request another with a /ai-review comment.

[github-actions]

🤖 AI Review

Merged four findings into three. Confirmed the phase-label and UTF-8 decoding nits. The cleanup-timeout concern remains uncertain: scope placement and unbounded cleanup are verified, but the pinned Effect implementation is unavailable, and the patched Vitest runner complicates the claimed diagnostic loss.

Findings

Severity	Location	Category	Sources	Claim
🟡 MINOR	packages/stack/src/services/Realtime.integration.test.ts:211	test-diagnostics	claude+codex	Stalled scope cleanup may prevent the 110-second guard from surfacing CatalogStallError before Vitest's 120-second timeout.
⚪ NIT	packages/stack/src/services/Realtime.integration.test.ts:35	test-diagnostics	claude	Filesystem root setup is labeled 'database recipe', so a stall during that setup would identify the wrong phase.
⚪ NIT	packages/stack/src/services/Realtime.integration.test.ts:43	correctness	codex	Decoding each output chunk with a fresh TextDecoder corrupts UTF-8 characters split across chunk boundaries.

Stats

Claude findings: 2 · Codex findings: 2 · Confirmed: 2 · Refuted: 0 · Uncertain: 1

---

Models: claude-opus-5-5 + gpt-6.1-sol · Trigger: auto · Workflow run

This review runs once per PR. A maintainer can request another with a /ai-review comment.

[github-actions]

Flaky test check: ✅ no flaky tests

Commit e9d829dcbef2 · focused ×10 (5 executions each), 1 test · integration ×10, 5635 tests · unit ×10, 8394 tests · workflow run

_{Each execution is a separate Vitest process; Where lists failing runs, as run.iteration for suites executed more than once per run. The full data is in the flaky-check-report artifact.}