Thread: Vacuum timing in pg_stat_all_tables
In light of bb8dff9995f (add cost delay time to progress views), looking at the output of 30a6ed0ce4b (track per-relation time spent on vacuum and analyze), it struck me as a bit unclear of what the time is really showing.
Do we want to do something similar for the table views? Or if not, we should probably at least document the effect of cost based vacuum delay on those timings - as in if they are including it or not (which I do believe they are).
While more stats are always nice :), I think just being clear about it in the docs would perhaps be enough for now? Maybe just appending something along the line of "(including cost based delaying)"?
On Tue, Mar 04, 2025 at 03:12:18PM +0100, Magnus Hagander wrote: > In light of bb8dff9995f (add cost delay time to progress views), looking at > the output of 30a6ed0ce4b (track per-relation time spent on vacuum and > analyze), it struck me as a bit unclear of what the time is really showing. > > Do we want to do something similar for the table views? Or if not, we > should probably at least document the effect of cost based vacuum delay on > those timings - as in if they are including it or not (which I do believe > they are). I could see it being useful to have the total cost delay time in those views. The information in the progress views goes away when vacuuming is done, while the table views would retain it indefinitely. That being said, I haven't judged the feasibility of adding it. I'm sure it can be done, but I don't know whether it requires reworking commit bb8dff9995f. > While more stats are always nice :), I think just being clear about it in > the docs would perhaps be enough for now? Maybe just appending something > along the line of "(including cost based delaying)"? I think this is also a reasonable thing to do regardless of whether we add the cost delay time to the table views. -- nathan
Hi, On Tue, Mar 04, 2025 at 08:54:13AM -0600, Nathan Bossart wrote: > On Tue, Mar 04, 2025 at 03:12:18PM +0100, Magnus Hagander wrote: > > In light of bb8dff9995f (add cost delay time to progress views), looking at > > the output of 30a6ed0ce4b (track per-relation time spent on vacuum and > > analyze), it struck me as a bit unclear of what the time is really showing. > > > > Do we want to do something similar for the table views? Or if not, we > > should probably at least document the effect of cost based vacuum delay on > > those timings - as in if they are including it or not (which I do believe > > they are). > > I could see it being useful to have the total cost delay time in those > views. The information in the progress views goes away when vacuuming is > done, while the table views would retain it indefinitely. +1, I think that could be useful to "retain" this information on a per-table basis. > > While more stats are always nice :), I think just being clear about it in > > the docs would perhaps be enough for now? Maybe just appending something > > along the line of "(including cost based delaying)"? Like "more stats are always nice" I think that "more explanations in the doc" are always nice, so I don't see any reason why not to add this extra explanation. Regards, -- Bertrand Drouvot PostgreSQL Contributors Team RDS Open Source Databases Amazon Web Services: https://aws.amazon.com