Short version: that stack trace is asyncpg timing out while Prefect (via SQLAlchemy) tries to open a TCP connection to your Postgres database. It’s almost always a networking/connectivity issue (wrong host/port, security group/firewall/VPC routing, DB not listening) or, less commonly, an aggressive connection timeout. What to check first - Can you reach the DB from the same EC2 host? - Try:

psql "host=<host> port=<port> dbname=<db> user=<user> sslmode=require" -c 'select 1'

- Or TCP check:

nc -vz <host> <port>

(should say “succeeded”) - Security groups/NACLs/routes: - Inbound on the DB (e.g., RDS) allows 5432 from your EC2 instance’s SG. - If the DB is on another host, Postgres is listening on 0.0.0.0 (not just localhost). - Connection URL is correct and uses the async driver: - It should look like:

<postgresql+asyncpg://USER:PASS@HOST:5432/DBNAME>

- If your provider requires TLS, include it, e.g.:

...?sslmode=require

- The DB is up and healthy (logs show it accepting connections). Optional: increase Prefect’s DB timeouts/pool limits In Prefect 3.x you can tune these with environment variables (server-side): - Increase connection establishment timeout (default ~5s): -

PREFECT_SERVER_DATABASE__CONNECTION_TIMEOUT=15

- Increase general DB operation timeout (default ~10s): -

PREFECT_SERVER_DATABASE__TIMEOUT=30

- Adjust pool if you’re under load (to avoid pool checkout timeouts): -

PREFECT_SERVER_DATABASE__SQLALCHEMY__POOL_SIZE=10

-

PREFECT_SERVER_DATABASE__SQLALCHEMY__POOL_TIMEOUT=60

-

PREFECT_SERVER_DATABASE__SQLALCHEMY__MAX_OVERFLOW=20

Example env setup

export PREFECT_SERVER_DATABASE__CONNECTION_URL='<postgresql+asyncpg://user:pass@your-host:5432/prefect?sslmode=require>'
export PREFECT_SERVER_DATABASE__CONNECTION_TIMEOUT=15
export PREFECT_SERVER_DATABASE__TIMEOUT=30
export PREFECT_SERVER_DATABASE__SQLALCHEMY__POOL_SIZE=10
export PREFECT_SERVER_DATABASE__SQLALCHEMY__POOL_TIMEOUT=60
export PREFECT_SERVER_DATABASE__SQLALCHEMY__MAX_OVERFLOW=20

References - Prefect 3 database configuration (AsyncPostgresConfiguration) - Prefect 3 settings reference A couple of clarifying questions to zero in: - Where exactly are you seeing this error: Prefect Server (API) logs on the EC2 host, or inside a flow/worker that connects to your own application database? - What’s the DB host (RDS? another EC2?) and can you

psql

to it from the same EC2 machine? - Could you share your connection URL format (redact user/pass), and whether TLS is required by your DB?