Hello, I noticed that a small optimization is possible in the flow of wait stat reporting for the LWLocks, when the pgstat_track_activities is disabled. If the check for pgstat_track_activities is done before invoking LWLockReportWaitStart() instead of inside the pgstat_report_wait_start(), it can save some of the statements execution where the end result of LWLockReportWaitStart() is a NO-OP because pgstat_track_activities = false.
I also see, that there are other callers of pgstat_report_wait_start() as well, which would also have to change to add a check for the pg_stat_activities, if the check is removed from the pgstat_report_wait_start(). Is the pg_stat_activities check inside pgstat_report_wait_start() because of some protocol being followed? Would it be worth making that change. Regards, Neha Cheers, Neha On Thu, Mar 10, 2016 at 12:31 AM, Amit Kapila <amit.kapil...@gmail.com> wrote: > On Fri, Mar 4, 2016 at 7:11 PM, Alexander Korotkov <aekorot...@gmail.com> > wrote: > > > >> > >> If yes, then the only slight worry is that there will lot of repetition > in wait_event_type column, otherwise it is okay. > > > > > > There is morerows attribute of entry tag in Docbook SGML, it behaves > like rowspan in HTML. > > > > Thanks for the suggestion. I have updated the patch to include > wait_event_type information in the wait_event table. > > As asked above by Robert, below is performance data with the patch. > > M/C Details > ------------------ > IBM POWER-8 24 cores, 192 hardware threads > RAM = 492GB > > Performance Data > ---------------------------- > min_wal_size=15GB > max_wal_size=20GB > checkpoint_timeout =15min > maintenance_work_mem = 1GB > checkpoint_completion_target = 0.9 > > > pgbench read-only (median of 3, 5-min runs) > > clients BASE PATCH % > 1 19703.549206 19992.141542 1.4646718364 > 8 120105.542849 127717.835367 6.3380026745 > 64 487334.338764 495861.7211254 1.7498012521 > > > The read-only data shows some improvement with patch, but I think this is > mostly attributed to run-to-run variation. > > pgbench read-write (median of 3, 30-min runs) > > clients BASE PATCH % > 1 1703.275728 1696.568881 -0.3937616729 > 8 8884.406185 9442.387472 6.2804567394 > 64 32648.82798 32113.002416 -1.6411785572 > > > In the above data, the read-write data shows small regression (1.6%) at > higher client-count, but when I ran individually that test, the difference > was 0.5%. I think it is mostly attributed to run-to-run variation as we see > with read-only tests. > > > Thanks to Mithun C Y for doing performance testing of this patch. > > As this patch is adding 4-byte variable to shared memory structure PGPROC, > so this is susceptible to memory alignment issues for shared buffers as > discussed in thread [1], but in general the performance data doesn't > indicate any regression. > > [1] - http://www.postgresql.org/message-id/CAA4eK1+ZeB8PMwwktf+ > 3brs0pt4ux6rs6aom0uip8c6shjw...@mail.gmail.com > > With Regards, > Amit Kapila. > EnterpriseDB: http://www.enterprisedb.com > > > -- > Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) > To make changes to your subscription: > http://www.postgresql.org/mailpref/pgsql-hackers > >
