Repository navigation
quest(mbx-r2-verify): record the first full-history and mbx 1.30.0 Nightlies - #8713
Conversation
…ghtlies Windows still misses most looked-up keys, with the download budget the leading suspect; Linux full history removed "not looked up" without changing hits while the 5 GiB budget binds. Drops the two cleared waits and asks the maintainer to approve filing the drafted upstream issue. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 17 minutes. View limit details
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. |
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes: the single docs-only change to quest/m0/ci/mbx-r2-verify.md.
- Evidence recorded: the 11-12 and 11-13 Nightly findings (Windows misses, Linux full-history comparison, 11-13 expiry wave) with run links and a per-job hits table.
- Waits updated: both cleared waits are dropped. A
Maintainer:approval wait for filing the upstream issue is added. - Plan additions: a draft upstream issue and the checks the first nights after #8631 must make.
I read the diff only and did not re-derive the figures from the run logs. The text is internally consistent, and the links and the conditional follow-up are coherent.
claude-sonnet-5-5 | 𝕏
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
quest land watch: the merge queue removed this pull request at 06fe71c because Its head has moved to 0b421f8 since, and a push by anyone but the land fixer ends the pick: the watcher leaves this pull request alone until a maintainer enqueues it again. (written by quest land watch) |
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes
Reviewed the delta since the prior pullfrog review (06fe71c), which touches only quest/m0/ci/mbx-r2-verify.md.
- Recorded the upstream decision: added the maintainer's 2026-11-13 call on when to post to
jdx/mr-boxington#950and when to file the drafted issue. - Replaced the approval wait: the
Maintainer:item became aGitHub Actions:wait on the first Nightly after #8631, with a branch for each outcome. - Reworded the draft and checklist: the draft issue is now the fallback, and the Windows checklist item points at that wait.
claude-sonnet-5-5 | 𝕏

Problem
Two waits in mbx-r2-verify have cleared. One asked whether Windows still misses most looked-up keys on mbx 1.30.0. The other asked whether full history (#8342) changed each Linux Rust-cache job's hits against its depth-1 baseline.
Approach
This PR records the numbers in the quest's Plan, with run links. The source is each Rust-cache job's
mbx[cache]: object cachelines in five scheduled Nightlies:the three depth-1 baselines: 42659882487, 42842827718 and 43041909115
43250182795, on 11-12
43447849289, on 11-13
Windows still misses most looked-up keys. On 11-13 the workspace build had 92 hits, 1,062 misses and 13 not looked up.
mbx analyzefiles every miss under "no earlier recording", so it names no cause. The download budget is the leading suspect. The one good night, 846 hits on 11-05, downloaded 8.7 GiB with no refusals. Every night since 11-08 hit the 10 GiB cap and got 92 to 163 hits. #8631 lifted the cap to 20 GiB after this run.Linux: no job regressed because of full history, but hits didn't rise either. Not looked up fell to near 0, but every job still spends its 5 GiB budget on prefetch. On 11-12 the summed hits were 10,990, inside the depth-1 range of 9,748 to 11,065. Ten jobs were above their own range, 16 inside it and 3 below it.
11-13 is an expiry wave.
NoSuchKeyprefetch failures went from 20 to 5,270 across 38 jobs. That includes 2,837 on macOS, which never reaches its budget. The 7-day rule expired a week of uploads at once, so 11-13 cannot isolate full history or mbx 1.30.0.The quest also changes in four ways:
GitHub Actions:wait on that Nightly says what to do in each case.fetch-depth: 1.Choices made without asking
Validation
No code changes. The figures were extracted by script from every job log of the five runs, plus the 11-05 and 11-08 Windows jobs. The extracted baselines match those the quest had already recorded: doctest 124 / 502 / 150, Media 58 / 84, and Audio quality 74, 77, 0 and so on.
quest checkandquest check --base origin/mainpass.Impact
Public API: none. Wire: none.
Alternatives
File the Windows issue now on the 11-13 evidence, or post nothing upstream. The maintainer rejected both.
Follow-ups
None new. The depth-1 revert is a conditional step in the quest's Plan.
Models
Size: S
Implementer: Claude Code, claude-opus-5-5, unknown effort
(Written by Claude Opus 5.5)
🤖 Generated with Claude Code
Review
Skipped on head
0b421f8d558c3af2752f9dd09ed733361113b1a1: since the reviewed head06fe71ca2367the change only edits the quest. It records the maintainer's 2026-11-13 upstream-report decision and swaps the filing-approval Waiting item for a GitHub Actions wait.quest checkpasses.Previous round, on
06fe71ca2367: Codex, gpt-6.1-sol, medium effort, session01a1d82f-07f5-7f83-a720-a6b193033ea6, round 1, score 9/10, no findings.