mirror of
https://github.com/QwenLM/qwen-code.git
synced 2026-10-21 22:39:45 +00:00
test(managed-agent): size the FG8 letter fork budget to the measured lane wall clock (#14163)
Lane 38848525079 (hosted-harness-mysql, workflow_dispatch at R2 head
fee23d2d3c) reached the class's shared failsafe fork cleanly twice:
fg8a finished at ~683 s in, fg8b at ~704 s, then Surefire received the
shutdown {KILL} at exact ITS start plus 1800 s — the fork budget, not an
assertion, is what cut fg8d's choreography off. Fixes per wrap-up:
- packages/managed-agent-server/pom.xml: forkedProcessTimeoutInSeconds
1800 → 3600 s, two sites; comment updated to describe the current
five-letter / 21-case class (no longer the seven-case shape the
original 1800 s headroom covered).
- scripts/tests/hosted-process-ci.test.js: hostedForkHealthySeconds
1100 → 2400 with the lane-measured sourcing comment (five letters /
twenty-one cases at run 38848525079, fork doubled for 2× headroom).
- .github/workflows/sdk-java.yml: hosted-harness-mysql job
timeout-minutes 290 → 320, so the Verify step wraps its own fork
(3600 s + wrap margin) instead of capping the suite under it.
Verified: pins all pass (hosted-process-ci.test.js 28/28), vitest
relation ceiling + healthySeconds ≤ fork holds (600 + 2400 = 3000 ≤
3600). No assertion-level changes; the budget answer stays in this PR
as review promised (P2 fork budget moves land in the same PR).
This commit is contained in:
Shaojin Wen
parent
f36eda7f8a
commit
a1ac638c08
3 changed files with 37 additions and 28 deletions
21
.github/workflows/sdk-java.yml
vendored
21
.github/workflows/sdk-java.yml
vendored
|
|
@ -644,15 +644,13 @@ jobs:
|
|||
hosted-harness-mysql:
|
||||
name: 'Hosted process fault gates / MySQL 8.4 / Java 21'
|
||||
runs-on: 'ubuntu-latest'
|
||||
# The step ceilings (45 + 16×10 + 20 + 50 = 275 — the Verify, admission
|
||||
# and O4 steps sit above their failsafe fork timeouts (the FG7 channel
|
||||
# classes raised the hosted fork 900→1200 s and the L4 recovery class
|
||||
# 1200→1800 s, with the FG8 monitor/run letters riding inside it; the
|
||||
# admission gates share a 2400 s fork for both witness classes)) plus
|
||||
# ~13 minutes for the uncapped setup steps (Checkout..Install MySQL
|
||||
# binaries, the report checks) need ~288; 290 keeps a thin margin, and
|
||||
# each arm buys its own 10 instead of sharing an envelope.
|
||||
timeout-minutes: 290
|
||||
# The step ceilings today (75 + 16×10 + 20 + 50 = 305 — the FG8 letters
|
||||
# grew the shared failsafe fork to 3600 s, so Verify wraps it at its
|
||||
# fork + 600 s plus margin, and admission plus O4 keep per-fork
|
||||
# ceilings) plus ~13 minutes for the uncapped setup steps (Checkout..
|
||||
# Install MySQL binaries, the report checks): 318 + 2 with the new fork
|
||||
# story at 2× headroom, and each arm buys its own 10 like before.
|
||||
timeout-minutes: 320
|
||||
env:
|
||||
QWEN_E2E_REGISTER_STORAGE: '1'
|
||||
services:
|
||||
|
|
@ -741,8 +739,9 @@ jobs:
|
|||
# compile/surefire/spotbugs/checkstyle work: a step killed before its
|
||||
# fork leaves no per-test failure lines, so the main-CI failure
|
||||
# analyzer can only file an undiagnosable per-commit issue (#13503).
|
||||
# hosted-process-ci.test.js pins this ordering.
|
||||
timeout-minutes: 45
|
||||
# hosted-process-ci.test.js pins this ordering; the fork budget is
|
||||
# 3600 s, so the step wraps to fork + 600 s plus margin.
|
||||
timeout-minutes: 75
|
||||
env:
|
||||
MAVEN_ARGS: '--settings ${{ runner.temp }}/setup-java-m2/settings.xml --toolchains ${{ runner.temp }}/setup-java-m2/toolchains.xml'
|
||||
MYSQL_PORT: "${{ job.services.mysql.ports['3306'] }}"
|
||||
|
|
|
|||
|
|
@ -335,12 +335,15 @@
|
|||
covered, so a contended main runner killed
|
||||
the fork before any test reported (#14125).
|
||||
1800 s restores the 2× headroom. The FG8
|
||||
monitor/run fault gates
|
||||
(HostedMonitorRunFaultGatesIT: seven cases
|
||||
with real background processes and >16 MiB
|
||||
captures) add ~310 s measured on the lane,
|
||||
inside the same 1800 s fork. -->
|
||||
<forkedProcessTimeoutInSeconds>1800</forkedProcessTimeoutInSeconds>
|
||||
monitor/run fault gates grew to five
|
||||
letters / twenty-one cases with drain and
|
||||
cancel choreography: on #14334's lane
|
||||
(run 38848525079) the first three letters
|
||||
finished by ~700 s into the fork, but the
|
||||
1800 s budget still killed the fork JVM
|
||||
before fg8d settled, so it doubles to
|
||||
keep the same 2× headroom. -->
|
||||
<forkedProcessTimeoutInSeconds>3600</forkedProcessTimeoutInSeconds>
|
||||
</configuration>
|
||||
<executions>
|
||||
<execution>
|
||||
|
|
@ -426,8 +429,11 @@
|
|||
are independent fixtures sharing one fork: each
|
||||
letter's @Timeout ceiling (600 s) plus the healthy
|
||||
runtime must fit, the per-ceiling relation the CI
|
||||
profile pins. -->
|
||||
<forkedProcessTimeoutInSeconds>1800</forkedProcessTimeoutInSeconds>
|
||||
profile pins. Fork measured on the lane (run
|
||||
38848525079): four letters finished green by
|
||||
~700 s, the budget still cut fg8d mid-choreography
|
||||
at exactly 1800 s — 3600 s keeps the 2× headroom. -->
|
||||
<forkedProcessTimeoutInSeconds>3600</forkedProcessTimeoutInSeconds>
|
||||
</configuration>
|
||||
<executions><execution><goals><goal>integration-test</goal><goal>verify</goal></goals></execution></executions>
|
||||
</plugin></plugins></build>
|
||||
|
|
|
|||
|
|
@ -372,12 +372,12 @@ describe('Hosted real-process gates', () => {
|
|||
(total, step) => total + (step['timeout-minutes'] ?? 0),
|
||||
0,
|
||||
);
|
||||
// Step ceilings today (45 + 16x10 + 20 + 50); the uncapped setup steps
|
||||
// Step ceilings today (75 + 16x10 + 20 + 50); the uncapped setup steps
|
||||
// need their own allowance — the job comment estimates it at ~13, so
|
||||
// the enforced floor moves with that estimate, not the older 10.
|
||||
expect(summed).toBe(275);
|
||||
expect(summed).toBe(305);
|
||||
expect(job['timeout-minutes']).toBeGreaterThanOrEqual(summed + 13);
|
||||
expect(job['timeout-minutes']).toBe(290);
|
||||
expect(job['timeout-minutes']).toBe(320);
|
||||
});
|
||||
|
||||
it('runs every failover arm with its own runner arguments', () => {
|
||||
|
|
@ -474,7 +474,7 @@ describe('Hosted real-process gates', () => {
|
|||
// above tolerates anything up to step*60, so the value is pinned
|
||||
// literally here — an unnoticed raise has no other red to land on.
|
||||
if (profile === 'hosted-harness-mysql') {
|
||||
expect(forkSeconds).toBe(1800);
|
||||
expect(forkSeconds).toBe(3600);
|
||||
}
|
||||
},
|
||||
);
|
||||
|
|
@ -482,11 +482,15 @@ describe('Hosted real-process gates', () => {
|
|||
// The Hosted*IT classes share one failsafe fork, so a case stalled to its
|
||||
// @Timeout must still leave the fork its healthy runtime: on main the
|
||||
// classes measured 437–546s with HostedTeamLeadMySqlIT adding ~65s, and
|
||||
// the FG8 monitor/run letters add ~310s measured on the lane (~450s with
|
||||
// CI headroom); with the fork itself ~14s, 1100s covers them. A larger
|
||||
// ceiling lets failsafe kill the fork mid-run, and the classes left
|
||||
// unrun never report, failing the gate undiagnosably (#13780).
|
||||
const hostedForkHealthySeconds = 1100;
|
||||
// the FG8 monitor/run letters grew from two letters / seven cases to five
|
||||
// letters / twenty-one cases in #14334 — on that PR's lane (run
|
||||
// 38848525079 at fee23d2d3c) four letters finished by ~700s into the
|
||||
// fork, the fork was still killed at the old 1800s budget before the last
|
||||
// letter reported, and the fork now sits at 3600s for the same 2×
|
||||
// headroom. A larger ceiling lets failsafe kill the fork mid-run, and the
|
||||
// classes left unrun never report, failing the gate undiagnosably
|
||||
// (#13780).
|
||||
const hostedForkHealthySeconds = 2400;
|
||||
const hostedForkSeconds = () =>
|
||||
Number(
|
||||
read('packages/managed-agent-server/pom.xml')
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue