edufeed/edufeed-app

Watch

Fork 0

Five DM/inbox/gift-wrap test files fail intermittently under full-suite load (pre-existing on dev) #70

New issue

Open

opened 2026-07-30 21:20:33 +00:00 by laoc · 0 comments

laoc commented

2026-07-30 21:20:33 +00:00

Owner

Five DM/inbox/gift-wrap test files fail intermittently under full-suite parallel load, on dev itself. Found while verifying #64; not caused by that branch — reproduced on unmodified dev.

The files

src/lib/__tests__/dm-recipient-relays.test.js
src/lib/__tests__/gift-wrap-publish.test.js
src/lib/__tests__/services/inbox-prefetch.test.js
src/lib/__tests__/services/inbox-read-tracking.test.js
src/lib/__tests__/services/inbox-service.test.js

Typical failures — all timeouts, not assertion mismatches:

× loads kind 10050 and 10002 for a recipient not yet in the store   5022ms
× publishes only to the relays resolved by the action               5025ms
× creates combined filter array with #p and #P                     30019ms
× extracts address pointer from reaction with a-tag                30023ms
× adds event ID to read set and persists to localStorage           30024ms

src/routes/api/pdf/__tests__/server.test.js and .../pdf-thumbnail/... join the set under heavier load (5031ms, 5002ms).

Evidence it is `dev`, not a branch

Two full-suite runs on a clean detached worktree at dev (aa4a129a):

run	result
1	`470 passed (470)` / `5193 passed (5193)`, exit 0
2	`5 failed \| 465 passed`, `6 failed \| 5187 passed`, exit 1 — exactly the five files above

Same commit, same machine, same command, opposite outcomes.

Why this is worth an issue

It made me misdiagnose my own branch. I ran dev once (green), then my branch (red), and concluded I had introduced a regression. I spent a full bisect on it — stashing, a lazy-import rewrite — before a second dev run showed the same five red. One reading of a flaky suite is not a baseline.
CI will read it as a real failure. pnpm test exits 1. Combined with the known GlobalFAB EnvironmentTeardownError flake (which also exits non-zero while every test passes), the suite currently has two independent ways to report red without a real defect.

Suspected shape

The failures are 5s/30s timeouts concentrated in files that mock relay/pool infrastructure and await network-shaped promises. Timing-sensitive under CPU contention rather than logically wrong — consistent with fake timers, real setTimeout, or unresolved promises that only settle when the event loop is not starved. Not diagnosed.

Not established

Which specific await in each file times out.
Whether one root cause covers all five files or they merely share a trigger.
Whether it reproduces in CI or only on a loaded dev machine.

Five DM/inbox/gift-wrap test files fail intermittently under full-suite parallel load, on `dev` itself. Found while verifying #64; **not caused by that branch** — reproduced on unmodified `dev`. ## The files ``` src/lib/__tests__/dm-recipient-relays.test.js src/lib/__tests__/gift-wrap-publish.test.js src/lib/__tests__/services/inbox-prefetch.test.js src/lib/__tests__/services/inbox-read-tracking.test.js src/lib/__tests__/services/inbox-service.test.js ``` Typical failures — all **timeouts**, not assertion mismatches: ``` × loads kind 10050 and 10002 for a recipient not yet in the store 5022ms × publishes only to the relays resolved by the action 5025ms × creates combined filter array with #p and #P 30019ms × extracts address pointer from reaction with a-tag 30023ms × adds event ID to read set and persists to localStorage 30024ms ``` `src/routes/api/pdf/__tests__/server.test.js` and `.../pdf-thumbnail/...` join the set under heavier load (5031ms, 5002ms). ## Evidence it is `dev`, not a branch Two full-suite runs on a clean detached worktree at `dev` (`aa4a129a`): | run | result | |---|---| | 1 | `470 passed (470)` / `5193 passed (5193)`, exit 0 | | 2 | `5 failed \| 465 passed`, `6 failed \| 5187 passed`, exit 1 — **exactly the five files above** | Same commit, same machine, same command, opposite outcomes. ## Why this is worth an issue 1. **It made me misdiagnose my own branch.** I ran `dev` once (green), then my branch (red), and concluded I had introduced a regression. I spent a full bisect on it — stashing, a lazy-import rewrite — before a second `dev` run showed the same five red. One reading of a flaky suite is not a baseline. 2. **CI will read it as a real failure.** `pnpm test` exits 1. Combined with the known `GlobalFAB` `EnvironmentTeardownError` flake (which also exits non-zero while every test passes), the suite currently has two independent ways to report red without a real defect. ## Suspected shape The failures are 5s/30s timeouts concentrated in files that mock relay/pool infrastructure and await network-shaped promises. Timing-sensitive under CPU contention rather than logically wrong — consistent with fake timers, real `setTimeout`, or unresolved promises that only settle when the event loop is not starved. Not diagnosed. ## Not established - Which specific await in each file times out. - Whether one root cause covers all five files or they merely share a trigger. - Whether it reproduces in CI or only on a loaded dev machine.

laoc referenced this issue

2026-07-31 09:33:09 +00:00

feat(cards): say what the linked material is — type, size, and PDF page count (#57) #75

No labels

bug

discussion

enhancement

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

edufeed/edufeed-app#70

No description provided.

Rows
Columns