Dashboards and database metrics ?

[vadius45]

Hello,

I am evaluating different monitoring tools for Postgres and I came across pgwatch very recently. I truly like the comprehensive documentation and the vision : it just works !
My plan is to run the daemon on each database host using the unix socket to connect to the cluster(s), with a user that is granted pg_monitor role. Therefore, I don't have to manage & rotate the credentials at scale. And I'd use the prometheus sink, metrics would be scraped from there.

Am torn between two options:

• postgres-continuous-discovery mode to fetch the table stats in every application database. That seems reasonable to capture and expose these metrics for application teams . And it saves me the hassle of refreshing the sources every time there's a new database created (or dropped)
• but then I have lots of dbname entries in the grafana dashboards : for most of them, it does not make sense to have the database-level view like number of connections / sessions or replication because the scope of these metrics is the cluster , not the individual databases.

My ask to the pgwatch community: how do you deal with cluster-level and database-level metrics and the various dashboards ?

[pashagolub]

hey there!

Yeah, been there! :) What I would do is

• have a postges database under full preset monitoring
• add postgres-continuous-discovery for everything else except postgres under basic (or other) preset

This way you will have comprehensive insight into the cluster itself through postgres monitoring source,
and you will have additional per database info from discovery. You might want to create your own preset that will suit your needs.

And of course you may update your dashboards! Just change the definition of the dashboard variable and you're done!

[vadius45]

oh, that's clever !
Let me give it a try

[vadius45]

Hi,
I took inspiration from the presets and defined two different sections for each cluster:_

• one with kind postgres for the instance level metrics like replication / stat_activity / wal / replication ..., with -instance- in its name . And I include only postgres db (maybe redundant because kind is postgres)
• one with kind postgres-continuous-discovery for the db level metrics like tables stats , with -userdb- in its name . And I exclude postgres db.

In Grafana, the prometheus query for dbname is amended to filter on -instance- or -userdb- depending on the dashboards.