Under load (or randomly), my app throws errors about failing to get a database connection, requests pile up and time out, and restarting the app temporarily fixes it. Diagnose and fix my connection exhaustion: 1. Establish the facts: my database's max connection limit, my app's pool size per process, and how many processes/instances run. The math often fails here: 10 app instances each opening a pool of 10 connections against a database that allows 60 total. 2. Check for connection leaks: code paths that check out a connection and never release it (errors thrown mid-query, missing release in a finally path, transactions left open). A leak explains the 'restart fixes it for a while' pattern. 3. Find connection hogs: long-running queries, reports or exports run through the app pool, or per-request code that opens multiple connections instead of reusing one. 4. Check for serverless-specific issues if applicable: each function instance creating its own pool, cold instances multiplying connections, no external pooler (like PgBouncer or the managed equivalent) in front of the database. 5. Fix what's found: right-size the pool relative to the database limit and instance count, fix leak paths with proper finally/release handling, move heavy queries off the request path or onto a separate pool, and add monitoring on pool saturation so I see it coming next time. Don't just crank the pool size up; that moves the failure into the database. The deliverable is the diagnosis (which of these it actually was), the fixes, and how to verify: a load test or reproduction that previously exhausted connections now passing, plus the monitoring I should watch.