* Add a collector for pg_buffercache_summary()
* Requires PostgreSQL >= 16.
---------
Signed-off-by: Peter Nuttall <peter.nuttall@snowflake.com>
Co-authored-by: Ben Kochie <superq@gmail.com>
* Export query itself together with queryId in stat_statement metrics
* The feature must be enabled via flag.
* Limit length of selected query.
The query is not added to every metrics, but instead of new metric stat_statement_query_id is introduced that contains mapping between queryId and query.
Fixes: #813
---------
Signed-off-by: Jakub Štiller <stiller@2n.cz>
Signed-off-by: Jakub Štiller <jakstiller@atlas.cz>
This no longer returns an error when creating a collector.instance when the database cannot be reached for the version query. This will resolve the entire postgresCollector not being registered for metrics collection when a database is not available. If the version query fails, the scrape will fail.
Resolves#880
Signed-off-by: Joe Adams <github@joeadams.io>
The intent is to use the instance struct to hold the connection to the database as well as metadata about the instance. Currently this metadata only includes the version of postgres for the instance which can be used in the collectors to decide what query to run. In the future this could hold more metadata but for now it keeps the Collector interface arguments to a reasonable number.
Signed-off-by: Joe Adams <github@joeadams.io>
Updates the exporter-toolkit to the latest version
* Adds new landing page feature.
* Allow metrics path to be on `/`.
Signed-off-by: SuperQ <superq@gmail.com>
The pg_database collector was not respecting the --exclude-databases flag and causing problems where databases were not accessible. This now respects the list of databases to exclude.
- Adjusts the Collector create func to take a config struct instead of a logger. This allows more changes like this in the future. I figured we would need to do this at some point but I wasn't sure if we could hold off.
- Split the database size collection to a separate query when database is not excluded.
- Comment some probe code that was not useful/accurate
Signed-off-by: Joe Adams <github@joeadams.io>
This moves the metrics that are queried from pg_stat_bgwriter into a dedicated collector instead of dynamically generating queries and metrics from a map. It renames some metrics including adding the `_total` suffix on all of the counters to match prometheus standards. This implementation uses QueryRowContext to allow for later addition of context for cancellation. From the Postgres documentation, it states that there is one line per WAL sender process, but it is unclear how to differentiate between them in any meaningful way. When querying the table, there is no column to identify the row, only metrics about bgwriter.
Signed-off-by: Joe Adams <github@joeadams.io>
