[objectstack-fleet]

Requested by Irene · project thread
Fixes #2357
Clause-②: no
Metadata-protocol: none

维护者速读

• 改了什么:test/flow-campaign-enrollment.test.ts 里 17 个会建 world() 的用例,每个都显式带 30 秒预算(WORLD_BUDGET_MS),不再吃 vitest 默认的 5 秒。只改测试,不发布任何东西。
• 为什么改:每次「添加营销成员」都会把营销员看得到的所有合格人员一起入组,而每个 world() 多加 2 个线索、2 个联系人,所以文件越往后入组人数越多(7 → 51)。最后那个线索用例本机全量跑要 4.9 秒,main CI 上两次超过 5 秒变红。
• 风险与回滚:风险很低,只放宽了单个用例的超时,断言一个没动;回滚就是 revert 这一个 commit。
• 先看:test/flow-campaign-enrollment.test.ts 第 46–58 行(预算常量和测量说明)。
• 两分钟验证:npx vitest run test/flow-campaign-enrollment.test.ts,预期 30 个用例全过,最慢的约 2.5 秒。
• 需要你做的:无,席位落地。

What changes

• test/flow-campaign-enrollment.test.ts: one constant, WORLD_BUDGET_MS = 30_000, with the measurement and the reason beside it, passed as the per-case timeout (the trailing argument, this repo's per-case convention) to every case that seeds a world(): 13 it and 4 it.each call sites, 17 in all. The cases that seed nothing (the structural reads and the two "names no campaign" runs) keep the default.
• .changeset/2357-campaign-enrollment-test-budget.md: an empty-frontmatter changeset, since nothing ships.
• No retry, no skip, no global testTimeout. The size matches ENROLL_BUDGET_MS in test/campaign-enrollment-doors.test.ts, which is 30 s for the same growth.

Why the late cases are slower than the early ones. The mechanism assumption holds: this is a slow case under load, not a hang. The cases are not equal, though. An Enroll Members run enrols every eligible person the marketer can see, not only its own world's people. Each world() adds two eligible leads (status: 'new') and two eligible contacts (department: 'engineering'), all owned by the marketer, and every member is inserted through a per-person subflow. So the run's work grows with each case before it in the file. A probe that counted each run's member rows (a scratch copy of the file, not committed) read 7, 9, 11, 15, 17 members for the leads cases, 16 to 24 for the contacts cases, and then 46 (the #2241 contacts case) and 51 (the #2241 leads case at :438, the last enrolling run in the file). Duration tracks that count. The file's own header already says the filter "rightly enrols the eligible ones among them too", so the growth is how the fixture works, not a defect. It is bounded by the number of cases in this file, because every file boots its own stack.

Tests

Per-case durations, in ms, on base f660fd8 (the change only adds timeouts, so the durations are the same on this head). "Full" is one full local vitest run --coverage (CI's pnpm run test:coverage) on a 4-core container: 239 files, 4592 passed, 1 skipped, 0 failed, about 17 minutes.

Case (world()-seeding only)	Alone	Full run
enrols the eligible leads in the chosen status	507	799
never enrols an opted-out, converted or off-status lead…	518	1032
tops up rather than double-enrolling on a re-run	541	1198
does not treat an enrolment in ANOTHER campaign as a duplicate	970	1651
enrols into a campaign still in planning	780	1672
contacts: enrols the eligible contacts in the chosen department	689	1456
contacts: never enrols an opted-out or off-segment contact…	878	1478
contacts: tops up rather than double-enrolling a contact	881	1488
contacts: does not treat a LEAD enrolment as a duplicate	1031	2247
contacts: picking contacts enrols no leads	1189	2299
contacts: picking leads enrols no contacts	1381	2662
#2121 completed / aborted × leads / contacts (refused, 4 cases)	87–153	165–212
#2121 control: in_progress leads	1739	4003
#2121 control: in_progress contacts	2079	4294
#2121 control: planning leads	2205	4272
#2241 Enroll = leads / contacts shows only its criterion	92 / 67	205 / 129
#2241 a contacts run with no lead status submits…	2061	4947
#2241 a leads run with no department submits…	2528	4906
#2241 control: contacts / leads run without its criterion	154 / 164	250 / 122

The local full run passed, but the two #2241 submit cases finished 53 ms and 94 ms under the 5 s default. On main CI the leads case went past 5000 ms on 0f17c6c and on 173b488 (job 122642586187: "Test timed out in 5000ms", test/flow-campaign-enrollment.test.ts:424:3 on that tree). 30 s is about 6× the slowest full-run reading and 12× the slowest alone.

Acceptance notes

• CI per-case durations: NOT MEASURED. CI's default vitest reporter prints one total per file and names a case only when it fails. So the CI reading is the two 5000 ms timeouts above. This PR's own CI run gives the file total.
• A different fix would retire each world's eligible people after its case, so every run enrols about 7 members and the file stops growing. That changes the fixture's lifecycle, which this card did not ask for, so it is noted here and not done.

Verification

• pnpm verify (validate, typecheck, lint --strict, hygiene, token ratchet, build, vitest run) on this branch, exit 0: Test Files 239 passed (239), Tests 4592 passed, 1 skipped (4593), 824 s. The changeset gained its two claim lines during that run, so pnpm hygiene and pnpm hygiene:tokens were re-run on head cb2c9c3: "source hygiene clean", "source token ratchet clean", both exit 0.
• The isolated run on this head: npx vitest run test/flow-campaign-enrollment.test.ts, 30 passed.

---

Generated by Claude Code

🤖 Generated with Claude Code

https://claude.ai/code/session_01J8VVAjZErQs1eFzRrtWuDe

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated
hotcrm	Ignored		Oct 28, 2026 6:27am UTC

[objectstack-fleet]

席位核对

Verdict: ACCEPT · repo:hotcrm seat 1 · huangyiirene · session session_01J8VVAnPkgK6CWrJa6t7Fk5 · 2026-10-28T06:30Z · head cb2c9c3
Implemented-by: card thread session session_01J8VVAjZErQs1eFzRrtWuDe (report 6462342379 on #2357). Reviewed-by: the seat, against the 2-file diff and hotcrm origin/main f660fd8.

• Shape:
  • Draft, base main, assignee huangyiirene, and the body reads Fixes #2357.
  • No governed path is touched.
  • The diff stays within claim 6461825148. It changes the test file plus an empty changeset carrying the claim lines.
  • git merge-tree against f660fd8 is clean.
• Mechanism assumption 1 holds: not a hang.
  • Each enrolling run enrols every eligible person the marketer can see, not only its own world's. Members therefore grow from 7 to 51 down the file, and the last Enroll Members: the lead-status and contact-department criteria are both shown and both required whatever the source, so a contact run requires a lead field it ignores — show only the chosen source's criterion (#2194 T1-U05) #2241 leads case is the slowest.
  • Measured: 0.5–2.5 s per case alone, and 0.8–4.9 s in a full local test:coverage run. The Enroll Members: the lead-status and contact-department criteria are both shown and both required whatever the source, so a contact run requires a lead field it ignores — show only the chosen source's criterion (#2194 T1-U05) #2241 submit cases ran 4947 and 4906 ms, at the edge of the 5 s default where main CI failed twice.
  • Refused runs stay under 0.3 s.
• Ruling held:
  • The 17 world()-seeding cases carry WORLD_BUDGET_MS = 30_000 as the per-case trailing argument. That is the repo's existing convention and the same size as ENROLL_BUDGET_MS in test/campaign-enrollment-doors.test.ts, which covers the same growth.
  • There is no retry, no skip, and no global testTimeout change.
  • The reason for the growth is written beside the constant.
  • The seat considered scoping each run to its own world, for example with a marketer per world. It does not require that here: the sibling suite already accepts the same growth with the same budget, and the cases' assertions read only their own world's keys.
• Tests:
  • The head passes 30 of 30 in isolation.
  • pnpm verify exit 0: 239 files, 4592 passed, 1 skipped.
  • CI per-case times are not measured, because the reporter prints per file only. The PR says so.
• Gates: CI on cb2c9c3 has 8 of 9 check runs success. Build and Test (22.x) is still running.
• Deviation noted: a model-named trailer was amended out before the PR, with a force-with-lease push on the thread's own branch. History on main is untouched.
• Landing: non-governed. The seat flips the PR to ready and arms auto-merge in this stroke.

---

Generated by Claude Code