Thread: Per query FDW network stat collection
Hello, I have implemented per query network stat collection for FDW. It is done in a similar way to how buffer and WAL stats are collected and it can be seen with a new NETWORK option for explain command: explain (analyze, network) insert into itrtest values (2, 'blah'); QUERY PLAN ----------------------------------------------------------------------------------------------- Insert on itrtest (cost=0.00..0.01 rows=0 width=0) (actual time=0.544..0.544 rows=0 loops=1) Network: FDW bytes sent=197 received=72, wait_time=0.689 -> Result (cost=0.00..0.01 rows=1 width=36) (actual time=0.003..0.003 rows=1 loops=1) Planning Time: 0.025 ms Execution Time: 0.701 ms (5 rows) I am yet to add corresponding columns to pg_stat_statements, write tests and documentation, but before I go ahead with that, I would like to know what the community thinks about the patch. Regards, Ilya Gladyshev
Attachment
On Tue, Aug 24, 2021 at 5:12 PM Ilya Gladyshev <i.gladyshev@postgrespro.ru> wrote: > > I have implemented per query network stat collection for FDW. It is done > in a similar way to how buffer and WAL stats are collected and it can be > seen with a new NETWORK option for explain command: > > explain (analyze, network) insert into itrtest values (2, 'blah'); > > QUERY PLAN > ----------------------------------------------------------------------------------------------- > Insert on itrtest (cost=0.00..0.01 rows=0 width=0) (actual > time=0.544..0.544 rows=0 loops=1) > Network: FDW bytes sent=197 received=72, wait_time=0.689 > [...] It sound like a really useful metric to have. However I'm not sure that having a new "network" option is the best way for that. It seems confusing as IIUC it won't be catching all network activity (like fe/be activity, or network disk...) but only FDW activity. I think it would be better to have those information retrieved when using the verbose option rather than a new one. Similarly, I'm afraid that INSTRUMENT_NETWORK could be misleading, although I don't have any better proposal right now.
On 24.08.2021 12:19, Julien Rouhaud wrote: > However I'm not sure that having a new "network" option is the best > way for that. It seems confusing as IIUC it won't be catching all > network activity (like fe/be activity, or network disk...) but only > FDW activity. I think it would be better to have those information > retrieved when using the verbose option rather than a new one. > Similarly, I'm afraid that INSTRUMENT_NETWORK could be misleading, > although I don't have any better proposal right now. I am also doubtful about this naming. Initially, I wanted to add fe/be activity as one of the metrics, but then decided to restrict myself to FDW for now. However, I decided to leave "network" as it is, because to me it makes sense to have all the network-related metrics under a single explain option (and a single instrumentation flag perhaps), in case more are added later. The struct fields used for collection internally tell explicitly that they are meant to be used only for FDW stats and the explain output also mentions that the displayed stats are for FDW network activity.