fix manifest cache

[zreigz]

Test Plan

Test environment: https://console.your-env.onplural.sh/

Checklist

• I have added a meaningful title and summary to convey the impact of this PR to a user.
• If required, I have updated the Plural documentation accordingly.
• I have added tests to cover my changes.
• I have deployed the agent to a test environment and verified that it works as expected (required only when changing agent code).

Plural Flow: console

[soffi-ai]

Soffi AI Summary

This PR fixes a subtle but impactful bug in the deployment operator's manifest caching layer. The cache previously stored a directory path for a fetched service manifest tarball and would return it as a cache hit as long as the entry was live and the SHA matched — but it never verified that the directory actually still existed on disk. If the directory was deleted or evicted (e.g., by an ephemeral container restart, node pressure, or external cleanup), the operator would return a stale/invalid path, leading to silent sync failures.

The fix adds a filesystem existence check (dirExists) before treating a cache entry as valid. If the cached directory is missing, the entry is explicitly removed from the cache and the manifest is re-fetched from the tarball source. Additionally, on any service sync error, the component SHA cache entry is now expired to force a clean re-evaluation on the next reconcile, preventing a bad cached state from persisting through retries.

Commits

Commit	Summary
927c9e0	Adds a filesystem existence check to the manifest cache hit path. If a cached directory entry is live and the SHA matches but the directory no longer exists on disk, the stale cache entry is removed and the manifest is re-fetched rather than returning an invalid path.
0efc136	Expires the component SHA cache entry whenever a service sync error occurs, ensuring that error state does not persist across reconcile attempts and that the next reconcile starts from a clean slate.

---

Updated: 2026-08-25 14:35 UTC

Deploy in Soffi

[greptile-apps]

Greptile Summary

This PR makes manifest-cache reuse conditional on the cached directory still existing, evicting and refetching entries whose directories disappeared.

• Verifies cached paths with os.Stat before returning them.
• Removes invalid cache entries and logs missing-directory refetches.
• Adds a small directory-existence helper.

Confidence Score: 5/5

The PR appears safe to merge because invalid cached directories are now evicted and refetched without changing valid-cache behavior.

The changed path verifies directory availability before reuse, removes unusable entries, and falls through to the existing manifest-fetch flow; investigated concurrency and error paths did not reveal a new reachable failure.

Important Files Changed

Filename	Overview
go/deployment-operator/pkg/manifests/cache.go	Adds missing-directory detection and eviction to prevent stale manifest-cache paths from being reused; no actionable changed-code defect was identified.

_{Reviews (1): Last reviewed commit: "fix manifest cache" | Re-trigger Greptile}

[michaeljguarino]

what's the bug here?

[zreigz]

what's the bug here?

The agent reused a cached manifest path after rollout (or other changes) and directory was deleted. The agent couldn't fetch a new tarball because relayed on the cache. A later successful sync never cleared that error in Console

[michaeljguarino]

there's probably an associated bug fix to make sure the errors properly clear then too