Filter Instance by Database (or the other way around?)

[rkrisztian]

In the dashboard filters, we can select only one Instance (host machine) on the dashboard, but multiple Database (PG database server instance). This make me conclude that the dashboard was designed also for cases when multiple PostgreSQL databases may run on the same host. However, the current solution also has a flaw: I can select an incorrect Database for a given Instance. As a result, the graphs will become empty or show "no data" (N/A). This is especially confusing if someone (accidentally) saves the dashboard with a non-All default for Database. I would like to prevent this confusion from happening.

For this reason, I think the dashboard should be adapted so that there is a clear "has" relationship between Instance and Database, or in other words: either selecting an Instance should refresh the selectable Database values, or selecting a Database should refresh the available Instances. This would then require the use of Chained Variables, and the order of variables would start to matter.

I tried to implement this improvement for myself, but for some reason nothing I tried work. After hours of trying I gave up. Given I'm no Grafana expert, I thought maybe you can help, but not for me personally, of course. It could be a great improvement for the template itself.

Thanks for considering!

[rkrisztian]

Now that I have more experience with this topic, I’ve discovered the following:

• If Instance depends on Database, then using the pg_up metric for filtering no longer works, because it doesn’t include the datname label. If we switch to a metric like pg_database_size_bytes, we run into another issue: for hosts that have scraping problems (and therefore only expose metrics like pg_exporter_last_scrape_error), no instances will appear in the filter. This makes it a poor solution overall.

• To make Database depend on Instance, we can modify the query from label_values(datname) to something like label_values(pg_database_size_bytes{instance="$instance"}, datname). In the case of a scraping error, the Database filter will likely become empty (though I haven’t tested this). Still, I believe this behavior is preferable. In the happy path, if you select an instance that doesn’t contain the selected database, the Database filter will reset to All.