supabase / supavisor / 37323859913

81%

Build Type

push

github

Committed by web-flow

Commit Message

fix: one pool disconnecting no longer wipes metrics for a tenant's other pools (#1221)

## Summary

- When a tenant has more than one active pool (for example, a busy
transaction-mode pool and an occasional session-mode connection), and
just one of those pools disconnects, Supavisor deletes the metrics for
the **whole tenant** — not just the pool that disconnected. This means
the still-running pool's metrics disappear too, even though it never
stopped working.
- This matches a live customer issue: `supavisor_*` Prometheus metrics
vanish for hours at a time on a project with steady transaction-mode
traffic plus an intermittent session-mode connection.
- The bug is in `Supavisor.TenantsMetrics.handle_info/2`
(`lib/supavisor/tenants_metrics.ex`): it clears the cache using only the
tenant name as the key, not the specific pool that went away.
- POOLER-830
- 
## Test plan

- `mix test test/supavisor/tenants_metrics_test.exs` — 1 test, 0
failures.
- The test: registers a transaction-mode pool and a session-mode pool
for the same tenant, confirms both show up in the cached metrics, then
drops only the session-mode pool. After that, the transaction-mode
pool's metrics are gone too, even though it's still connected —
reproducing the bug.

---------

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>

Coverage Stats

5 of 5 new or added lines in 1 file covered. (100.0%)

3 existing lines in 3 files now uncovered.

3085 of 3806 relevant lines covered (81.06%)

49548.23 hits per line