On Sun, Sep 6, 2015 at 7:34 AM, 周正中(德歌) <dege....@alibaba-inc.com> wrote:
> > >On 2015-09-10 19:39:59 +0800, 张广舟(明虚) wrote: > >> We found there is a fsync call when CLOG buffer > > >> is written out in SlruPhysicalWritePage(). It is often called when a > >> backend > >> needs to check transaction status with SimpleLruReadPage(). > > >That's when there's not enough buffers available some other, and your > >case dirty, needs to be written out. > > >You could try increasing the max (32) in CLOGShmemBuffers() further. > > I think increasing CLOGShmemBuffers have no help for our case, because > PG will call pg_fsync once for every CLOG PAGE(BLOCKSZ) when ExtendCLOG。 > > Ideally that should not happen every time, CLOG page is written only when there is no non-dirty CLOG buffer to accommodate a new page during ExtendCLOG. I think it will happen only if there is no flush of CLOG pages by Checkpoint or by any other process trying to commit the transaction during last 32 pages worth of transactions (which is 4 (TRANS_PER_BYTE) * 8192 (BLKSIZE) * 32 (CLOG_BUFFERS)). What kind of workload you are using, can you once try by commenting the sync call and see if there is any gain (I understand that is not right, but at least we will come to know how much we can save by moving such a sync to checkpoint). With Regards, Amit Kapila. EnterpriseDB: http://www.enterprisedb.com
