diff --git a/quest/m0/ci/README.md b/quest/m0/ci/README.md index 061a768aab2..1cd74241d54 100644 --- a/quest/m0/ci/README.md +++ b/quest/m0/ci/README.md @@ -166,5 +166,4 @@ The historical samples locate candidates; they are not controlled benchmarks. - [moq-audio's default-feature test build is clippy-clean and gated](/quest/m0/ci/default-feature-test-lint.md) - the dead `retains`/`fail` under default-feature `--tests` clippy is fixed, and the gate catches it next time - [A release merge opens no stale release PR](/quest/m0/ci/release-pr-after-release.md) - `just rs release` publishes before it proposes, so a merged release isn't re-proposed - [Dependabot leaves the workspace-hack to hakari](/quest/m0/ci/dependabot-workspace-hack.md) - cargo group PRs stop hand-editing `rs/workspace-hack/Cargo.toml` -- [mbx's free-space GC never removes a live target](/quest/m0/ci/mbx-gc.md) - the cargo shim stops deleting targets other sessions use when the host disk runs low - [Bun runs only the test files we name](/quest/m0/ci/bun-test-paths.md) - scripted `bun test` calls pass exact paths, so generated copies never run diff --git a/quest/m0/ci/mbx-gc.md b/quest/m0/ci/mbx-gc.md deleted file mode 100644 index 098f75cd4bb..00000000000 --- a/quest/m0/ci/mbx-gc.md +++ /dev/null @@ -1,47 +0,0 @@ -# [S] mbx's free-space GC never removes a live target - -## Goal - -On agent hosts, a cargo call through the mbx shim never deletes a Cargo -target another session is still using. A run that loses disk headroom gets -a clear failure instead of silently evicting the shared cache. The host -settings that keep it so live in `kixelated/dotfiles`. - -## Plan - -Decided 2026-11-03 (maintainer, /quest-plan of the follow-ups from -[moq-dev/moq#7742](https://github.com/moq-dev/moq/pull/7742)): milestone m0, -in the CI epic. -[moq-dev/moq#7715](https://github.com/moq-dev/moq/pull/7715) only worked -around the same GC for long interop runs. - -- Incident - ([comment](https://github.com/moq-dev/moq/pull/7742#issuecomment-6605544835)): - `mbx[gc]: 11.2 GiB free … under the 50.0 GiB minimum`, then "removed 7 - target directories (… 6 abandoned and 1 live)". It also evicted about - 30 GiB of shared cache objects below `gc.max_size`, because the disk was - under `gc.min_free_size`. -- This is documented mbx behaviour (`mbx gc --help`): below - `gc.min_free_size`, it collects past its budgets, managed targets first. - `~/work/dotfiles/mbx/config.toml` sets no `[gc]` options, so the - defaults apply. On 2026-11-03, `/` on `desktop` was 98% full with 49 GB - free, still under the threshold, so any cargo call can trigger it again. -- Work out why the disk is that full first: other worktrees' targets, `.tmp/` - scratch, Nix store. `quest worktree` reclaims finished checkouts. -- Then protect live targets in dotfiles. Lowering `min_free_size` only - changes when collection runs: below it, mbx still removes unprotected - managed targets. mbx's docs - ([managed targets](https://mr-boxington.jdx.dev/managed-targets)) describe - a running command's lease and a `target.keep` setting. Check which of them - failed for the target logged as live: a binary run straight from - `target/` holds no lease. Then set `target.keep` or the fitting `[gc]` - option. If mbx can't spare live targets at all, report that upstream and - record the issue link here. -- Edit dotfiles in place (`~/work/dotfiles` is live config), and link its - commit here when done. - -Public API: none. Wire: none. - -## Related - -- The queue host keeps the queue's disk - disk policy for the queue host, configured in kixelated/nixos