Prisma "P2024: Timed out fetching a connection from the pool"
Prisma waited for a free connection from its pool and gave up after pool_timeout. Either every connection was busy, or the database was briefly slow to respond. The pool wait timing out is often a transient, retryable condition.
What this error means
prisma migrate deploy or a seed/connection step fails with P2024, reporting it timed out fetching a connection and showing the pool size. It frequently clears on retry once load or a slow database recovers.
Error: P2024
Timed out fetching a new connection from the connection pool. (More info:
http://pris.ly/d/connection-pool) (Current connection pool timeout: 10,
connection limit: 5)Common causes
Pool exhausted under concurrency
More concurrent queries than connection_limit means callers wait for a connection; if none frees within pool_timeout, P2024 fires.
Database briefly slow or unresponsive
A transient stall - a busy server, a failover, or a slow query holding connections - can make the pool wait time out even when nothing is misconfigured.
A pooler in front capping connections
PgBouncer/RDS Proxy limits can leave Prisma waiting for an upstream slot, surfacing as a pool timeout.
How to fix it
Retry the transient timeout
When the cause is a brief stall, a bounded retry usually succeeds with no change.
for i in 1 2 3; do
npx prisma migrate deploy && break
echo "pool timeout retry $i"; sleep 5
doneTune pool size and timeout
Raise the connection limit and pool timeout for a contended pipeline via the connection URL.
DATABASE_URL="postgresql://user:pass@db:5432/app?connection_limit=10&pool_timeout=30"Reduce contention during migrations
- Run migrations without competing application load on the same database.
- Use a direct connection for DDL rather than a saturated pooled URL.
- Avoid spawning many parallel Prisma processes against one small pool.
How to prevent it
- Size
connection_limit/pool_timeoutto the workload and server capacity. - Wrap migration steps in a bounded retry for transient stalls.
- Run migrations away from peak concurrent load.