Use geometric repack instead of gc --auto for mirror maintenance

[Trent-TSE]

Summary

Post-job mirror maintenance (runMirrorGC) now runs git repack -d --geometric=2 (plus git pack-refs --all --prune) instead of git gc --auto.

Why: every mirror sync fetch runs with gc.auto=0, so each committed sync adds one pack. Once the mirror crosses gc.autoPackLimit (50 packs), gc --auto does a full repack -A -d of every pack, including the multi-GB base pack from hydration (no gc.bigPackThreshold is set). On large monorepos that blows through GC_TIMEOUT_SECS (120s), and because a timed-out GC blocks the sticky-disk commit, the mirror stays over the threshold and every subsequent job with a changed mirror re-attempts the same full repack — visible as bursts of GC timed out after 120s / will not commit sticky disk warnings with ~2-minute post steps, until one job happens to finish inside the window. (The other --auto trigger, >6700 loose objects, only runs an incremental repack -d -l and is cheap; the pack-count trigger is the one that hurts.)

A geometric repack only rolls up the packs (and loose objects) needed to keep pack object counts in a factor-2 progression, so the base pack is never rewritten and the per-run cost is proportional to what recent syncs added; when the progression already holds it's a near no-op (Nothing new to pack.).

- timeout 120 git -c gc.autoDetach=false -C <mirror> gc --auto
+ timeout 120 git -C <mirror> repack -d --geometric=2
+ timeout 120 git -C <mirror> pack-refs --all --prune   # non-fatal

• --geometric needs git ≥ 2.32 (supportsGeometricRepack() parses git version); older gits keep the previous gc --auto path unchanged.
• Timeout / exit-code handling and the fail-safe "no commit on GC failure" semantics in cleanup() are unchanged. Everything still runs in the foreground so nothing is left holding the mount at umount.
• pack-refs replaces the ref packing gc used to do so loose refs from sync fetches don't accumulate; failure only warns (loose refs shadow packed-refs with identical values, so a partial run is harmless).
• A git count-objects -v summary is logged before maintenance ([git-mirror] Object store: packs=N, size-pack=…KiB, loose=N, …) so pack/loose counts can be correlated with maintenance time in real job logs.
• Trade-off: repack --geometric does not prune unreachable objects (no -A/prune), so objects from force-pushed/deleted branches stay in the mirror until it's re-hydrated. That growth is slow relative to the repo itself and is preferable to a maintenance step that can never complete.

dist/index.js rebuilt via npm run build.

Link to Devin session: https://app.devin.ai/sessions/42f58c11558242bdb1bfc805782f9ae9
Open in Devin Desktop: https://app.devin.ai/desktop/session/42f58c11558242bdb1bfc805782f9ae9?variant=devin
Requested by: @Trent-TSE

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled. (Staging)}

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[linear-code]

BLA-4660

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want higher recall? High effort reviews run extra passes and find more bugs. A team admin can switch effort levels in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit 74fe243. Configure here.}