feat(gate): self-heal preflight for a poisoned module cache - resolution canary, one loud purge-and-retry #132

New issue

Closed

opened 2026-07-15 23:06:52 +00:00 by hexajon · 0 comments

hexajon commented

2026-07-15 23:06:52 +00:00

(Migrated from codeberg.org)

Ask

gate.sh gains a cheap preflight before the check matrix:

Canary: resolve dependencies for one known import (e.g. go list -deps ./internal/cmdresult or go build -n on a tiny package). Cost when healthy: sub-second.
On canary failure: print a loud, single-line diagnosis naming the module cache as the suspect, run go clean -modcache (and -cache if the second attempt still fails), re-run the canary once, and continue if healed. One retry, never a loop.
If still failing after the purge: fail with the canary error AND the purge already ruled out - so I start at the real cause, not the cache.

Acceptance

A deliberately corrupted GOMODCACHE (test can simulate by deleting files inside an extracted module dir in a throwaway GOMODCACHE) yields a gate run that heals itself and passes.
Healthy-path overhead is negligible; the purge only ever triggers on canary failure.
The heal event is unmistakable in gate output (not silent).

A machine-side cache incident (2026-07-15: an external cache GC file-pruned the shared GOMODCACHE and hollowed out extracted module dirs) made ./scripts/gate.sh fail with cryptic per-package errors ("no required module provides package github.com/spf13/cobra" with a perfectly correct go.mod). Go auto-fetches MISSING modules, but a mutated-yet-present cache is trusted by contract - so the gate red-fails with no hint the cache is the culprit, and recovery took human archaeology. ## Ask gate.sh gains a cheap preflight before the check matrix: 1. Canary: resolve dependencies for one known import (e.g. `go list -deps ./internal/cmdresult` or `go build -n` on a tiny package). Cost when healthy: sub-second. 2. On canary failure: print a loud, single-line diagnosis naming the module cache as the suspect, run `go clean -modcache` (and `-cache` if the second attempt still fails), re-run the canary once, and continue if healed. One retry, never a loop. 3. If still failing after the purge: fail with the canary error AND the purge already ruled out - so I start at the real cause, not the cache. ## Acceptance - A deliberately corrupted GOMODCACHE (test can simulate by deleting files inside an extracted module dir in a throwaway GOMODCACHE) yields a gate run that heals itself and passes. - Healthy-path overhead is negligible; the purge only ever triggers on canary failure. - The heal event is unmistakable in gate output (not silent).

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

sixfold-space/madtea#132

No description provided.

Rows
Columns