"too many clients already": direct connections exhausted max_connections

"too many clients already": diagnose and fix

Symptom

FATAL: too many clients already on new connections.

Cause

Direct Postgres connections hit max_connections, which Neon sets by compute size: 104 at 0.25 CU, 209 at 0.5 CU, 419 at 1 CU, up to a 4000 cap at large sizes. Seven of those are reserved for the Neon superuser, so a 0.25 CU compute gives your app 97. Serverless apps opening a direct connection per instance or per request burn through this fast.

Confirm

1. Check you are on the direct string: the hostname lacks -pooler.
2. Run against the direct string: SHOW max_connections; and count actives:

SELECT count(*) FROM pg_stat_activity WHERE datname = '[your-db]';

3. Or look at the Monitoring page in the Console for the connection graph.

Fix

• Move application traffic to the pooled connection string (-pooler hostname). PgBouncer multiplexes up to 10,000 clients onto the real backends.
• Keep the direct string for migrations and admin only.
• If the direct side is legitimately busy (many admin tools), raise the compute size, which raises max_connections.

Verify

Connection count on the direct string drops; the error stops under the same load that produced it.

Neon's pooling docs (verified live 2026-09-26): maxconnections by compute size table (0.25 CU = 104, 1 CU = 419, capped at 4000), 7 reserved for the superuser, SHOW maxconnections and the pgstatactivity query, Console Monitoring page; pooled connections support up to 10,000 concurrent connections.