Hi! Thank you for your interest and experiment results. > 13 сент. 2017 г., в 15:43, Ants Aasma <ants.aa...@eesti.ee> написал(а): > > On Thu, Aug 31, 2017 at 9:02 AM, Andrey Borodin <x4...@yandex-team.ru> wrote: >> When we have accumulated diff blocknumbers for most of segments we can >> significantly speed up method of WAL scanning. If we have blocknumbers for >> all segments we can skip WAL scanning at all. > > Have you measured that the WAL scanning is actually a significant > issue? As a quick experiment I hacked up pg_waldump to just dump block > references to stdout in binary format. It scanned 2.8GB of WAL in 3.17 > seconds, outputting 9.3M block refs per second. WAL was generated with > pgbench, synchronous commit off, using 4 cores for 10 minutes - making > the ratio of work from generating WAL to parsing it be about 750:1. >
No, I had not done this measurement myself. Sure, parsing WAL, when it is in RAM, is not very expensive. Though, it can be even cheaper before formatting WAL. I just want to figure out what is the best place for this, if backuping exec is sharing CPUs with postmaster. Best regards, Andrey Borodin. -- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers
