Read replicas do not log 503 errors to stderr

[laurenceisla]

Environment

• PostgreSQL version: 15.6
• PostgREST version: 12.2.3
• Operating system: ubuntu

Description of issue

PostgREST logged the following (requests queries redacted):

127.0.0.1 - service_role [06/Jan/2025:17:44:02 +0000] "GET /rpc/<an_rpc> HTTP/1.1" 503 - "" "node"
06/Jan/2025:17:44:03 +0000: Successfully connected to PostgreSQL 15.6 ...
06/Jan/2025:17:44:03 +0000: Config reloaded
06/Jan/2025:17:44:03 +0000: Schema cache queried in 35.5 milliseconds
06/Jan/2025:17:44:03 +0000: Schema cache loaded ...
127.0.0.1 - service_role [06/Jan/2025:17:44:02 +0000] "GET /rpc/<another_rpc> HTTP/1.1" 503 - "" "node"
06/Jan/2025:17:44:03 +0000: Successfully connected to PostgreSQL 15.6 ...
06/Jan/2025:17:44:03 +0000: Config reloaded
06/Jan/2025:17:44:04 +0000: Schema cache queried in 37.9 milliseconds
06/Jan/2025:17:44:04 +0000: Schema cache loaded ...

As seen above it does not log the 503 error (when it should) and retries connection immediately. For instance, the 500 errors are logged correctly:

06/Jan/2025:17:44:59 +0000: {"code":"57014","details":null,"hint":null,"message":"canceling statement due to statement timeout"}
127.0.0.1 - service_role [06/Jan/2025:17:44:59 +0000] "GET /rpc/<an_rpc> HTTP/1.1" 500 - "" "node"

---

Now, checking the PostgreSQL logs, it shows that it's an error related to conflicts between the replica and the primary db. In particular, it shows these two errors:

message:"canceling statement due to conflict with recovery"
detail:"User query might have needed to see row versions that must be removed."

message:"terminating connection due to conflict with recovery"
detail:"User query might have needed to see row versions that must be removed."
hint:"In a moment you should be able to reconnect to the database and repeat your command."

Not sure if PostgREST should try to reconnect or not in these cases, since a solution is to set max_standby_archive_delay and max_standby_streaming_delay in PG config to avoid the error.