Matthias van de Meent

There is no user which will use TPC-DS as their main database. The only use case is performance testing and validation, which makes sense for a database product, but only...

> in a window of time Do we need arbitrary window of time, or is an open-ended window from some point in time good enough?

I'm looking into the origin of the visibility map issues in the pg_regress suite indicated [here](https://neon-github-public-dev.s3.amazonaws.com/reports/pr-8573/10725583140/index.html#suites/158be07438eb5188d40b466b6acfaeb3/8951fa9a9820a397/) - they seem to be relatively rare (or at least, rare enough for all...

@thesuhas do you have an issue that explains why we need this, and did you pass that by the SRE team? I'm asking, because this sounds like a metrics explosion...

> Do you think this will suffice? No, because in practice autovacuum will cause the values returned by "10 most recently vacuumed tables" to churn over time, through most tables....

No, if you want to measure how often autovacuum runs, you're probably better off summing up the values of `pg_stat_all_tables.autovacuum_count` for all tables of all databases.

> Won't this happen in this case as well? You will still go through the entirety of pg_stat_all_tables and sum up the autovacuum_count column. Yes, but at least (when summed)...

> > No, if you want to measure how often autovacuum runs, you're probably better off summing up the values of pg_stat_all_tables.autovacuum_count for all tables of all databases. > >...

> So our file updates it's LwLSN and returns false (redo should be applied to the page). Then page is swapped from LFC. How is this possible? We only update...

> doubling number of callbacks? What do you mean by that?