pm-stack(aggregator): prevent empty replies when dashboard assembly is slow

Follow-up from triage task 9456 on issue #5082, confirmed against current main 8de9292b77ff3fbfd72153bf0e6ac6758be635cc.

Evidence

On 2026-11-04 at 22:10:34Z, ordinary curl probes against the already-running laptop aggregator at 127.0.0.1:4870 returned curl: (52) Empty reply from server, 0 bytes, after ~12.0s for /board (12.002793s), /api/board.json (12.003588s), /api/overview.json (12.011524s), /frag/overview (12.007072s), and /program (12.004105s). /api/program.json returned 200 / 64,684B in 3.088804s; /api/quota.json returned 200 / 50,995B in 0.018966s. The server process was Bun PID 11116 with cwd /home/user/git/code/pm-stack/aggregator, started 09:39:30 local; this task did not launch it and its loaded SHA is unknown.

Current-main source inspection: pm-stack/aggregator/src/server.ts awaits the full context before returning on /board (1307-1316), /frag/overview (1302-1306), /api/overview.json (1745-1746), and /api/board.json (1748-1761). boardContext serially awaits board refs and items (459-464); overviewContext waits for its slowest facet (701-712). Bun.serve has no explicit idleTimeout (2211-2217), and the source reads have individual timeouts longer than the observed cutoff (for example Kaneo 8s per call, projects 20s, Supabase 60s). Current main has no bounded stale/error response for this assembly. The exact source of the 12s cutoff is not proven; an idle timeout is a plausible contributing cause, not a confirmed root cause.

Scope

Make the affected dashboard assembly routes settle with bytes: return a last-good/stale result when available, otherwise an explicit bounded timeout/error response (including a visible fragment error for /frag/overview) rather than allowing the connection to close silently. Add deterministic tests with slow/stalled stub sources that assert each affected route returns the defined stale/error behavior. Do not treat increasing Bun.serve idleTimeout alone as sufficient; the application still needs a bounded outcome. Preserve fast-path behavior.

Separate symptom

The historical /program HTML size regression (~1.16MB) is not measured by this probe because /program itself currently closes at ~12s; its JSON twin returned normally. Re-measure bytes after the response reliability fix and track HTML slimming separately if still present.

Links: #5082

Acceptance criteria

Under slow dashboard assembly, /board, /api/board.json, /api/overview.json and /frag/overview never return an empty reply: each either serves within a bounded time or returns an explicit error/stale response the client can act on.
The ~12s empty-reply probes recorded in the evidence (2026-11-04 22:10Z against the laptop aggregator) no longer reproduce, or the remaining slow path is named with its bound.
Fast endpoints (/api/quota.json, /api/program.json) do not regress.
Fix is covered by a test or a reproducible probe script showing behaviour before/after.

CODE-1414

Intake review

Related: follow-up from triage task 9456 on issue pm-stack(aggregator): /board, /api/board.json, /api/overview.json + /frag/overview die with empty replies at ~12s; /program HTML regrown to 1.16MB #5082; the aggregator's board/health fixture test failures seen elsewhere today are a separate thread — this issue is specifically the empty-reply-under-slow-assembly behaviour in pm-stack/aggregator/src/server.ts.
Notes: Already well labelled (kind/bug, priority/high, effort/m, queue/queued); this intake adds the missing acceptance criteria only. Evidence is unusually complete: timed curl probes plus the serial-await source locations.
Risks: The obvious fix (parallelise the context assembly) can change ordering/freshness semantics of the board; a timeout-with-explicit-error may be the safer first step. The running server's loaded SHA was unknown at probe time — confirm the repro against current main before designing the fix.
Suggested approach: Reproduce the 12s empty reply against a current-main aggregator, then bound the serial awaits in boardContext/overviewContext (parallel fetch and/or per-section deadlines), verifying with the same probe set.

Intake by fleet monitor (repo-fleet-monitor), 2026-11-04 run.

🤖 Agent note · handed off

Agent: repo queue worker, lane code-b (repo-worker-code)
Session / ID: cron repo-queue-worker-shift-b run 2026-11-05; task 9753
Found via: claimed task 9753 this shift; execution-lane assignment returned opencode
Task: task 9753 for issue pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664 (small_fix). Fresh check: issue pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664 is OPEN, no hard-stop labels (held / user/owned / do-not-merge / do-not-auto-edit). Handed off to the opencode claimant — the claimant launches the bound opencode session; this lane does not work it directly.
Expectation: claimant session drives task 9753 to a terminal state.
Links: task 9753 · issue pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664

🤖 Agent note · handed off

Agent: repo queue worker, lane code-u (repo-worker-code)
Session / ID: cron repo-queue-worker-shift-u run 2026-11-05 21:04 CST; task 9753
Found via: queue claim · task 9753
Task: task 9753 for issue pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664 (small_fix). Fresh check: issue pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664 is OPEN, no hard-stop labels (held / user/owned / do-not-merge / do-not-auto-edit). Execution-lane assignment returned opencode — handed off to the opencode claimant, which launches the bound opencode session; this lane does not work it directly.
Expectation: claimant session drives task 9753 to a terminal state. Attempts 15/16 CAP RISK — claimant should bind before lease expiry.
Links: task 9753 · issue pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664

🤖 Agent note · handed off

Agent: repo queue worker, lane code-d (repo-worker-code)
Session / ID: cron repo-queue-worker-shift-d run 2026-11-05 22:08 CST; task 9753
Found via: claimed task 9753 this shift; execution-lane assignment returned opencode
Task: task 9753 for issue pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664 (small_fix, attempt 16/16). Fresh check via gh: issue pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664 is OPEN, no hard-stop labels (held / user/owned / do-not-merge / do-not-auto-edit). Handed off to the opencode claimant — the claimant launches the bound opencode session; this lane does not work it directly.
Expectation: claimant session drives task 9753 to a terminal state.
Links: task 9753 · issue pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664

Navigation Menu

Repository navigation

pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664

Evidence

Scope

Separate symptom

Acceptance criteria

Intake review

137 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

pm-stack(aggregator): prevent empty replies when dashboard assembly is slow #5664

Description

Evidence

Scope

Separate symptom

Acceptance criteria

Activity

developing-today commented on Nov 4, 2026

dezren39 commented on Nov 4, 2026

Intake review

137 remaining items

dezren39 commented on Nov 6, 2026

dezren39 commented on Nov 6, 2026

dezren39 commented on Nov 6, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions