fix(graphile-cache): prevent double-disposal of PostGraphile instances during shutdown

[pyramation]

Summary

Fixes the PostGraphile instance has been released error during test teardown (and any closeAllCaches call) by replacing the disposedKeys string-keyed Set with a WeakSet<GraphileCacheEntry> keyed on entry object references.

Root cause: disposeEntry() added the cache key to disposedKeys as a double-disposal guard, but then deleted it in the finally block after disposal completed. When closeAllCaches() subsequently called graphileCache.clear(), the LRU dispose callback re-invoked disposeEntry() for the same entry — but the guard key was already gone, so it attempted pgl.release() on an already-released instance.

Fix: Track disposal by entry object reference (WeakSet) instead of key string. This eliminates both the race (no finally cleanup needed) and a theoretical key-reuse edge case where different entry objects could share the same cache key string.

Updates since last revision

Added a regression test suite in graphile/graphile-cache/__tests__/graphile-cache.test.ts that:

• Mocks pgl.release() to throw on double-call (matching real PostGraphile behavior)
• Verifies closeAllCaches() calls release() exactly once per entry, not twice
• Verifies single LRU eviction also disposes exactly once
• Covers the empty-cache edge case

CI server-test logs were also checked — the Error disposing PostGraphile message no longer appears in teardown output.

Review & Testing Checklist for Human

• Trace the closeAllCaches flow to confirm the race is resolved: disposeEntry marks the entry in the WeakSet → Promise.allSettled completes → graphileCache.clear() triggers LRU dispose callback → disposeEntry sees entry in WeakSet and no-ops
• Verify removal of the finally block is safe — the WeakSet needs no manual cleanup (entries are GC'd when unreferenced), unlike the old Set<string> which leaked stale keys without the finally. Note: if disposeEntry errors partway through, the entry stays in the WeakSet, preventing a retry — confirm this is acceptable (a failed release would likely fail again anyway)
• Run the server-test suite (cd graphql/server-test && pnpm test) and confirm the Error disposing PostGraphile log no longer appears in output

Notes

• The error was caught and logged (non-fatal), so CI was green even before the fix — the real signal is absence of the error log during teardown
• The regression test mocks pg-cache and @pgpmjs/logger to run in isolation; the full integration path is covered by the existing server-test CI job
• Two files changed: graphile-cache.ts (fix, net −1 line) and new graphile-cache.test.ts (regression test)

Link to Devin session: https://app.devin.ai/sessions/c36c53a01f67404eb07113d24a4b7c67
Requested by: @pyramation

[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 and CI monitoring