Re: [HACKERS] subscription worker signalling wal writer too much

2017-10-05 Thread Robert Haas
On Tue, Sep 19, 2017 at 8:48 AM, Kyotaro HORIGUCHI wrote: > Though a bit uneasy to have similar code on both side > (XLogBackgroundFlush and XLogSetAsyncXactLSN) but +1 to this from > me. This patch wasn't formatted very nicely; attached is a version that

Re: [HACKERS] subscription worker signalling wal writer too much

2017-09-19 Thread Kyotaro HORIGUCHI
At Sat, 26 Aug 2017 14:45:20 -0700, Jeff Janes wrote in

Re: [HACKERS] subscription worker signalling wal writer too much

2017-08-26 Thread Jeff Janes
On Mon, Jul 3, 2017 at 8:26 PM, Jeff Janes wrote: > On Sat, Jun 24, 2017 at 5:09 PM, Andres Freund wrote: > >> On 2017-06-15 15:06:43 -0700, Jeff Janes wrote: >> >> >> > >> > Wouldn't it >> > better, and much simpler, just to have reverted the first

Re: [HACKERS] subscription worker signalling wal writer too much

2017-08-25 Thread Robert Haas
On Mon, Jul 3, 2017 at 11:26 PM, Jeff Janes wrote: > My understanding is that you can't start on a new file until the old file is > completely synced, because the book keeping can currently only handle one > file at a time. So if you signal the wal writer to do the sync,

Re: [HACKERS] subscription worker signalling wal writer too much

2017-07-03 Thread Jeff Janes
On Sat, Jun 24, 2017 at 5:09 PM, Andres Freund wrote: > Hi, > > On 2017-06-15 15:06:43 -0700, Jeff Janes wrote: > > > It looks like only limited consolidation was happening, the number of > kills > > invoked was more than half of the number of SetLatch. I think the reason >

Re: [HACKERS] subscription worker signalling wal writer too much

2017-06-24 Thread Andres Freund
Hi, On 2017-06-15 15:06:43 -0700, Jeff Janes wrote: > > Well, wal_writer_delay doesn't work if walwriter is in sleep mode, and > > this afaics would allow wal writer to go into sleep mode with half a > > page filled, and it'd not be woken up again until the page is filled. > > No? > > > > If it

Re: [HACKERS] subscription worker signalling wal writer too much

2017-06-20 Thread Jeff Janes
On Thu, Jun 15, 2017 at 3:06 PM, Jeff Janes wrote: > > This new patch is simpler than the previous one, and more effective at > speeding up replication. I assume it would speed up pgbench with > synchronous_commit turned off (or against unlogged tables) as well, but I >

Re: [HACKERS] subscription worker signalling wal writer too much

2017-06-15 Thread Jeff Janes
On Wed, Jun 14, 2017 at 4:29 PM, Andres Freund wrote: > On 2017-06-14 16:24:27 -0700, Jeff Janes wrote: > > On Wed, Jun 14, 2017 at 3:20 PM, Andres Freund > wrote: > > > > > On 2017-06-14 15:08:49 -0700, Jeff Janes wrote: > > > > On Wed, Jun 14, 2017 at

Re: [HACKERS] subscription worker signalling wal writer too much

2017-06-14 Thread Andres Freund
On 2017-06-14 16:24:27 -0700, Jeff Janes wrote: > On Wed, Jun 14, 2017 at 3:20 PM, Andres Freund wrote: > > > On 2017-06-14 15:08:49 -0700, Jeff Janes wrote: > > > On Wed, Jun 14, 2017 at 11:55 AM, Jeff Janes > > wrote: > > > > > > > If I publish a

Re: [HACKERS] subscription worker signalling wal writer too much

2017-06-14 Thread Jeff Janes
On Wed, Jun 14, 2017 at 3:20 PM, Andres Freund wrote: > On 2017-06-14 15:08:49 -0700, Jeff Janes wrote: > > On Wed, Jun 14, 2017 at 11:55 AM, Jeff Janes > wrote: > > > > > If I publish a pgbench workload and subscribe to it, the subscription > > >

Re: [HACKERS] subscription worker signalling wal writer too much

2017-06-14 Thread Andres Freund
On 2017-06-14 15:08:49 -0700, Jeff Janes wrote: > On Wed, Jun 14, 2017 at 11:55 AM, Jeff Janes wrote: > > > If I publish a pgbench workload and subscribe to it, the subscription > > worker is signalling the wal writer thousands of times a second, once for > > every async

Re: [HACKERS] subscription worker signalling wal writer too much

2017-06-14 Thread Jeff Janes
On Wed, Jun 14, 2017 at 11:55 AM, Jeff Janes wrote: > If I publish a pgbench workload and subscribe to it, the subscription > worker is signalling the wal writer thousands of times a second, once for > every async commit. This has a noticeable performance cost. > I've

[HACKERS] subscription worker signalling wal writer too much

2017-06-14 Thread Jeff Janes
If I publish a pgbench workload and subscribe to it, the subscription worker is signalling the wal writer thousands of times a second, once for every async commit. This has a noticeable performance cost. I don't think it is ever necessary to signal the wal writer here, unless wal writer is