No way to enable logs collection when using docker image

[sebasmagri]

Seems like by default logs collection is not enabled when using the docker image without changes, and there's no environment variable or command switch to enable them.

[JordanP]

I have the same question: how does log collections work under Kubernetes ?

[lfittl]

@JordanP We don't support log collection in Kubernetes today - though looking to add this in the future.

For @sebasmagri's initial question - Sebastian, were you looking to monitor Kubernetes pods, or were you targeting something external, like a cloud provider database?

(generally log collection is enabled by default in recent collector releases - if supported for the current platform, that is)

[sebasmagri]

@lfittl it's for running the collector on ECS and monitoring RDS. We had to go with a custom docker image, but it's still pretty annoying that now the server names are forced to be the rds instance ID, since we create a lot of RDS instances dynamically and discard the old ones, being really a single server instance all the time.

[lfittl]

@sebasmagri Ah - makes sense. You can use the following settings to override the system identifiers:

PGA_API_SYSTEM_ID
PGA_API_SYSTEM_TYPE
PGA_API_SYSTEM_SCOPE

You can see the defaults that get assigned here: https://github.com/pganalyze/collector/blob/master/config/identify_system.go#L15 (note you could e.g. choose to override PGA_API_SYSTEM_ID in your setup to have a stable server record)

[sebasmagri]

Thanks for the pointer @lfittl !

We'd still like to be able to use the docker image as provided though, any chance that logging collection can be enabled by default for this use case?

[lfittl]

@sebasmagri In the case where you are not using the instance ID in the hostname, utilizing the AWS_INSTANCE_ID variable should work (that, or the hostname, is what the collector checks against to determine which code path to take)

[JordanP]

@lfittl beside Kubernetes, does the Collector supports monitoring a PG instance running in a container ?

[lfittl]

@JordanP There are two ways to do that, assuming you are referring to Postgres running inside a Docker container on a VM that you manage:

1. Run the pganalyze-collector on the Docker host, and use the db_log_docker_tail setting - this will directly fetch the logs from the Docker engine. Note that this setting is mostly intended for our own development purposes, and is considered experimental.

2. Run the pganalyze-collector in a Docker container. Have the Postgres container write the log output into a file that sits on a shared volume that the pganalyze-collector container has access to. Then use the LOG_LOCATION setting to specify that log file.

[JordanP]

Excellent ! Option 2 seems doable for us, nice !

[JordanP]

For the record, I managed to have log insights working in a Kubernetes cluster, with what you suggested: Run the pganalyze-collector in a Docker container. Have the Postgres container write the log output into a file that sits on a shared volume that the pganalyze-collector container has access to. Then use the LOG_LOCATION setting to specify that log file.

All good, thanks !

[rauanmayemir]

Is there any follow-up on this?

I still think that it would be very convenient to have a way for kubernetes-managed log collectors like fluentd or vector push logs into pganalyze (whether to the collector or directly to the centralized ingest endpoint).

[ianstanton]

@lfittl How realistic would it be to allow for pganalyze-collector to stream logs from a web socket? We're in a similar situation here, and our logs end up in Loki. We read these logs in other areas of our application via web socket.

[lfittl]

See #503 for an implementation that should work in the typical Kubernetes environment using fluentbit for log routing (tested with CloudNativePG, need to do a bit more testing with other operators).