Coroot UI fails to see one of postgres instances.

[Compunctus]

Hi. Coroot doesn't show one of the replicas in my 3-server postgres setup as a postgres node. Also it fails to see replication roles, and doesn't recognize other deployments (gitlab/nextcloud) as postgres.

The relevant metrics are present in prometheus:

Could you please advise on how to handle this? I get the feeling that coroot doesn't like low-query situations or my labels.

[def]

Is psqlha-postgresql-ha-postgresql-2 displayed on the Instances tab?
Can you also check the Coroot's log for errors?

Replication roles are currently discovered only for clusters managed by the k8s operators for Postgres (Zalando, CrunchyData, Percona).

[Compunctus]

Yep, it's displayed on instances tab. All three are green on the instances tab, now psqlha-postgresql-ha-postgresql-1 is not recognised as a postgres service.

I do see new errors in logs, namely:

couldn't find actual instance for "postgres", initial instance is "10.1.18.172:80" (map[])

and that ip (18.172) does belong to pod psqlha-postgresql-ha-postgresql-1

[def]

Please show the output of the following Prometheus queries:

• container_net_tcp_listen_info{listen_addr="10.1.18.172:80"}
• container_net_tcp_listen_info{container_id=~".*psqlha-postgresql-ha-postgresql-1.*"}

[Compunctus]

Both metrics were empty. I also noticed that they were simply wrong for most IP addresses - showing ports that could never exist in a pod. Restarted daemonset, now those metrics seem valid. Waiting for 10 minutes to let coroot settle.

No noticeable errors were found in node-agent logs prior to restarting the daemonset.

P.S. psqlha-postgresql-ha-postgresql-1 was deployed on a node which had it's network connectivity abruptly cut off recently. Then it was rebooted without restoring networking first. Could help in narrowing it down, i guess...