Connections stuck in idle/ClientRead - Django application
Neon
PythonHey everyone 
We've been battling a weird issue since Dec 16 and I'm honestly stuck. Hoping someone can point us in the right direction.
What's happening:
Our Django app randomly freezes - heartbeat stops etc. When I check
PID | STATE | WAIT_EVENT | DURATION | QUERY
10342 | idle | ClientRead | 32:32 | SELECT pid, state...
7452 | idle | ClientRead | 02:36 | select 1
12732 | idle | ClientRead | 02:36 | select 1
8199 | idle in transaction| ClientRead | 00:42 | SAVEPOINT "s139..."
So Postgres shows
The fact that even
What we've tried:
-
-
-
- TCP keepalives
- Tried both pooler AND direct endpoint - same issue on both
Our setup:
- Django 4.x + psycopg3 (
- AWS ECS Fargate
- Following the Neon Django docs config
Timeline:
This started Dec 16. I noticed there were some "Apply config" events in our Neon project around 9:18-9:43 PM that day. Could something have changed on the infrastructure side?
Impact:
When it happens, we have to restart the container. It's happening multiple times a day now.
Anyone seen anything like this? Or any ideas what else we could check?
Thanks
We've been battling a weird issue since Dec 16 and I'm honestly stuck. Hoping someone can point us in the right direction.
What's happening:
Our Django app randomly freezes - heartbeat stops etc. When I check
pg_stat_activityPID | STATE | WAIT_EVENT | DURATION | QUERY
10342 | idle | ClientRead | 32:32 | SELECT pid, state...
7452 | idle | ClientRead | 02:36 | select 1
12732 | idle | ClientRead | 02:36 | select 1
8199 | idle in transaction| ClientRead | 00:42 | SAVEPOINT "s139..."
So Postgres shows
idleClientReadThe fact that even
select 1What we've tried:
-
CONN_MAX_AGE=0-
DISABLE_SERVER_SIDE_CURSORS=True-
CONN_HEALTH_CHECKS=True- TCP keepalives
- Tried both pooler AND direct endpoint - same issue on both
Our setup:
- Django 4.x + psycopg3 (
psycopg[binary,pool]- AWS ECS Fargate
- Following the Neon Django docs config
Timeline:
This started Dec 16. I noticed there were some "Apply config" events in our Neon project around 9:18-9:43 PM that day. Could something have changed on the infrastructure side?
Impact:
When it happens, we have to restart the container. It's happening multiple times a day now.
Anyone seen anything like this? Or any ideas what else we could check?
Thanks
