feat: add support for pg_stat_user_indexes, pg_index_properties, pg_index_size_bytes

[michael-todorovic]

This PR adds support for pg_stat_user_indexes in order to get scan, read, fetch statistics about indexes. This can be used to find out unused/low-used indexes

This relates to issue #1065

[michael-todorovic]

You can identify unused indexes with this promql:

count((pg_stat_user_indexes_idx_scan_total{environment="production"} < 15)
* on (datname, environment, indexrelname, instance, job, relname, schemaname)
group_left(is_primary, is_unique, is_ready, is_valid)
(pg_index_properties{is_primary="0"} * on (datname, environment, indexrelname, instance, job, relname, schemaname)
  pg_index_size_bytes > 104857600)) by (instance, schemaname, relname, indexrelname)

It searches for production indexes scanned less than 15 times, which are not primary keys (you need to keep those indexes for PG itself) and that are bigger than 100MB

The metrics themselves look like:

pg_stat_user_indexes_idx_scan_total{datname="my_db", environment="production", indexrelname="idx_name", instance="my_server", job="pg_exporter_preview", relname="my_table", schemaname="my_schema"} 0
pg_stat_user_indexes_idx_tup_fetch{datname="my_db", environment="production", indexrelname="idx_name", instance="my_server", job="pg_exporter_preview", relname="my_table", schemaname="my_schema"} 0
pg_stat_user_indexes_idx_tup_read{datname="my_db", environment="production", indexrelname="idx_name", instance="my_server", job="pg_exporter_preview", relname="my_table", schemaname="my_schema"} 0
pg_index_properties{datname="my_db", environment="production", indexrelname="idx_name", instance="my_server", is_primary="0", is_ready="1", is_unique="0", is_valid="1", job="pg_exporter_preview", relname="my_table", schemaname="my_schema"} 1
pg_index_size_bytes{datname="my_db", environment="production", indexrelname="idx_name", instance="my_server", job="pg_exporter_preview", relname="my_table", schemaname="my_schema"} 9654042624

[chris-downs]

To give this a bump, we are now using this branch in production with no issues and it's very useful. It would be great to get it into mainline.

[sysadmind]

In addition to the comments on the code, all of the dates for the Copyright need to be updated to the current year for the files that you have created. I think the guys are good here, my feedback is mostly about being able to maintain this code long term.

[sysadmind]

I'm not sure why this import is in it's own section. This should go with the stdlib imports above.

[sysadmind]

We don't typically use a function to build the query unless we need dynamic columns. In this case it should just be a static query const.

[sysadmind]

I don't think that this slice adds anything. We can scan directly into the vars above.

[sysadmind]

Do these metrics make sense when these values are not known? I feel that if too many are NULL/unknown, the metric is useless and should just be skipped.

[sysadmind]

Again, what does the slice improve vs just using the values below?

[sysadmind]

Again, what does the slice improve vs just using the values below?