fix(remote): stop CI teardown reporting itself as a broken connection

[veksen]

Goal

Make CI run logs trustworthy: the final lines should reflect the verdict, not a scary false warning. Fixes Query-Doctor/Site#3580.

What

Before: every CI run that reached teardown logged [main] API connection broken during CI run: Error: RPC session was shut down by disposing the main stub right next to the verdict, reading as if the run was cut off mid-flight. After: an intentional teardown logs nothing; a genuine mid-run transport break still logs the broken-connection warning.

The run in the issue had completed normally — ingest succeeded, the baseline was fetched, and the exit 1 came from a correct verdict (2 untriaged regressions). The connection was never broken; the message fired during our own teardown.

How

connect() in src/remote/api-client.ts wires onRpcBroken → triggerBroken → onBroken. triggerBroken guards on a broken flag, but dispose() disposed the stub without ever setting it. Disposing the main stub makes capnweb fire onRpcBroken, so the intentional teardown reported itself as broken.

Fix: set broken = true at the top of dispose(), before disposing the stubs, so triggerBroken becomes a no-op for an intentional teardown. connectWithReconnect is unaffected — its onBroken calls dispose() after the fact, and triggerBroken already returns early when broken is set.

Read src/remote/api-client.ts first (the one-line guard + comment), then the test.

Tests

New src/remote/api-client.test.ts drives ApiClient.connect end-to-end over an in-process ws relay that answers the real auth handshake — the only setup where capnweb actually fires onRpcBroken on dispose (an in-memory MessagePort session does not reproduce it). Two cases:

• an intentional dispose() does not call onBroken;
• a real transport drop (server terminates the socket) still calls onBroken.

Confirmed the dispose test fails without the guard and passes with it. Full suite green (285 passed), tsc --noEmit clean, build clean.

🤖 Generated with Claude Code

[github-actions]

Query Doctor Analysis

3 queries analyzed

0 regressed · 0 improved · 0 new · 0 removed

2 pre-existing issues

• SELECT "guests"."id", "guests"."session_id", "guests"."username", "guests"."avatar_path", "guests"."color", "guests"."side", "guests"."audio_recording_path", "guests"."audio_recording_public", "gue...
  index assets(event_id, inserted_at desc)
  cost 31,003,449 → 1,498 (100% reduction)
• SELECT * FROM guest_ip_addresses WHERE ip_address = '127.0.0.1';
  index guest_ip_addresses(ip_address)
  cost 154,402 → 8 (100% reduction)

_{Using assumed statistics (10000000 rows/table). For better results, sync production stats.}

_{More detail → get_ci_run({ runId: "019f6d3a-5668-7841-8383-6d80dc1c0eb4" }) · view run · docs}