Thread
Commits
-
Reduce rate of walwriter wakeups due to async commits.
- 1f395354d874 17.0 landed
-
walwriter interacts quite badly with synchronous_commit=off
Andres Freund <andres@anarazel.de> — 2023-10-24T23:09:29Z
Hi, I recently mentioned to Heikki that I was seeing latch related wakeups being frequent enough to prevent walwriter from doing a whole lot of work. He asked me to write that set of concerns up, which seems quite fair... Here's a profile of walwriter while the following pgbench run was ongoing: c=1;psql -Xq -c 'drop table if exists testtable_logged; CREATE TABLE testtable_logged(v int not null default 0);' && PGOPTIONS='-c synchronous_commit=off' pgbench -n -c$c -j$c -Mprepared -T150 -f <(echo 'INSERT INTO testtable_logged DEFAULT VALUES;') -P1 Looking at top, walwriter is around 15-20% busy with this workload. Unfortunately a profile quickly shows that little of that work is useful: perf record --call-graph dwarf -m16M -p $(pgrep -f 'walwriter') sleep 3 - 94.42% 0.00% postgres postgres [.] AuxiliaryProcessMain AuxiliaryProcessMain - WalWriterMain + 78.26% WaitLatch + 14.01% XLogBackgroundFlush + 0.51% pgstat_report_wal 0.29% ResetLatch 0.13% pgstat_flush_io + 0.02% asm_sysvec_apic_timer_interrupt 0.01% HandleWalWriterInterrupts (inlined) Confirmed by the distribution of what syscalls are made: perf trace -m128M --summary -p $(pgrep -f 'walwriter') sleep 5 syscall calls errors total min avg max stddev (msec) (msec) (msec) (msec) (%) --------------- -------- ------ -------- --------- --------- --------- ------ epoll_wait 216610 0 3744.984 0.000 0.017 0.113 0.03% read 216602 0 333.905 0.001 0.002 0.029 0.03% fdatasync 27 0 94.703 1.939 3.508 11.279 8.83% pwrite64 2998 0 15.646 0.004 0.005 0.027 0.45% openat 2 0 0.019 0.006 0.010 0.013 34.84% close 2 0 0.004 0.002 0.002 0.003 25.76% We're doing far more latch related work than actual work. The walwriter many many times wakes up without having to do anything. And if you increase the number of clients to e.g. c=8, it gets worse in some ways: perf trace: epoll_wait 291512 0 2364.067 0.001 0.008 0.693 0.10% read 290938 0 479.837 0.001 0.002 0.020 0.05% fdatasync 146 0 410.043 2.508 2.809 7.006 1.90% futex 56384 43982 183.896 0.001 0.003 2.791 1.65% pwrite64 17058 0 105.625 0.004 0.006 4.015 4.61% clock_nanosleep 1 0 1.063 1.063 1.063 1.063 0.00% openat 9 0 0.072 0.006 0.008 0.014 14.35% close 9 0 0.018 0.002 0.002 0.003 5.55% Note that we 5x more lock waits (the futex calls) than writes! I think the problem is mainly that XLogSetAsyncXactLSN() wakes up walwriter whenever it is sleeping, regardless of whether the modified asyncXactLSN will lead to a write. We even wake up walwriter when we haven't changed asyncXactLSN, because our LSN is older than some other backends! So often we'll just wake up walwriter, which finds no work, immediately goes to sleep, just to be woken again. Because of the inherent delay between the checks of XLogCtl->WalWriterSleeping and Latch->is_set, we also sometimes end up with multiple processes signalling walwriter, which can be bad, because it increases the likelihood that some of the signals may be received when we are already holding WALWriteLock, delaying its release... Because of the frequent wakeups, we do something else that's not smart: We write out 8k blocks individually, many times a second. Often thousands of 8k pwrites a second. We also acquire WALWriteLock and call WaitXLogInsertionsToFinish(), even if could already know we're not going to flush! Not cheap, when you do it this many times a second. There is an absolutely basic optimization, helping a it at higher client counts: Don't wake if the new asyncXactLSN is <= the old one. But it doesn't help that much. I think the most important optimization we need is to have XLogSetAsyncXactLSN() only wake up if there is a certain amount of unflushed WAL. Unless walsender is hibernating, walsender will wake up on its own after wal_writer_delay. I think we can just reuse WalWriterFlushAfter for this. E.g. a condition like if (WriteRqstPtr <= LogwrtResult.Write + WalWriterFlushAfter * XLOG_BLCKSZ) return; drastically cuts down on the amount of wakeups, without - I think - loosing guarantees around synchronous_commit=off. 1 client: before: tps = 42926.288765 (without initial connection time) syscall calls errors total min avg max stddev (msec) (msec) (msec) (msec) (%) --------------- -------- ------ -------- --------- --------- --------- ------ epoll_wait 209077 0 3746.918 0.000 0.018 0.143 0.03% read 209073 0 310.532 0.001 0.001 0.021 0.02% fdatasync 25 0 82.673 2.623 3.307 3.457 1.13% pwrite64 2892 0 14.600 0.004 0.005 0.018 0.43% after: tps = 46244.394058 (without initial connection time) syscall calls errors total min avg max stddev (msec) (msec) (msec) (msec) (%) --------------- -------- ------ -------- --------- --------- --------- ------ epoll_wait 25 0 4732.625 0.000 189.305 200.281 4.17% fdatasync 25 0 90.264 2.814 3.611 3.835 1.02% pwrite64 48 0 15.825 0.020 0.330 0.707 12.76% read 21 0 0.117 0.003 0.006 0.007 3.69% 8 clients: tps = 279316.646315 (without initial connection time) postgres (2861734), 1215159 events, 100.0% syscall calls errors total min avg max stddev (msec) (msec) (msec) (msec) (%) --------------- -------- ------ -------- --------- --------- --------- ------ epoll_wait 267517 0 2150.206 0.000 0.008 0.973 0.12% read 266683 0 512.348 0.001 0.002 0.036 0.08% fdatasync 149 0 413.658 2.583 2.776 3.395 0.29% futex 56597 49588 183.174 0.001 0.003 1.047 0.69% pwrite64 17516 0 126.208 0.004 0.007 2.927 3.93% after: tps = 290958.322594 (without initial connection time) postgres (2861534), 1626 events, 100.0% syscall calls errors total min avg max stddev (msec) (msec) (msec) (msec) (%) --------------- -------- ------ -------- --------- --------- --------- ------ epoll_wait 153 0 4383.285 0.000 28.649 32.699 0.92% fdatasync 153 0 464.088 2.452 3.033 19.999 4.88% pwrite64 306 0 80.361 0.049 0.263 0.590 4.38% read 153 0 0.459 0.002 0.003 0.004 1.37% futex 49 46 0.211 0.002 0.004 0.038 17.05% More throughput for less CPU, seems neat :) I'm not addressing that here, but I think we also have the opposite behaviour - we're not waking up walwriter often enough. E.g. if you have lots of bulk dataloads, walwriter will just wake up once per wal_writer_delay, leading to most of the work being done by backends. We should probably wake walsender at the end of XLogInsertRecord() if there is sufficient outstanding WAL. Greetings, Andres Freund -
Re: walwriter interacts quite badly with synchronous_commit=off
Heikki Linnakangas <hlinnaka@iki.fi> — 2023-10-25T09:17:03Z
On 25/10/2023 02:09, Andres Freund wrote: > Because of the inherent delay between the checks of XLogCtl->WalWriterSleeping > and Latch->is_set, we also sometimes end up with multiple processes signalling > walwriter, which can be bad, because it increases the likelihood that some of > the signals may be received when we are already holding WALWriteLock, delaying > its release... That can only happen when walwriter has just come out of "hibernation", ie. when the system has been idle for a while. So probably not a big deal in practice. > I think the most important optimization we need is to have > XLogSetAsyncXactLSN() only wake up if there is a certain amount of unflushed > WAL. Unless walsender is hibernating, walsender will wake up on its own after > wal_writer_delay. I think we can just reuse WalWriterFlushAfter for this. > > E.g. a condition like > if (WriteRqstPtr <= LogwrtResult.Write + WalWriterFlushAfter * XLOG_BLCKSZ) > return; > drastically cuts down on the amount of wakeups, without - I think - loosing > guarantees around synchronous_commit=off. In the patch, you actually did: > + if (WriteRqstPtr <= LogwrtResult.Flush + WalWriterFlushAfter * XLOG_BLCKSZ) > + return; It means that you never wake up the walwriter to merely *write* the WAL. You only wake it up if it's going to also fsync() it. I think that's correct and appropriate, but it took me a while to reach that conclusion: It might be beneficial to wake up the walwriter just to perform a write(), to offload that work from the backend. And walwriter will actually also perform an fsync() after finishing the current segment, so it would make sense to also wake it up when 'asyncXactLSN' crosses a segment boundary. However, if those extra wakeups make sense, they would also make sense when there are no asynchronous commits involved. Therefore those extra wakeups should be done elsewhere, perhaps somewhere around AdvanceXLInsertBuffer(). The condition you have in the patch is appropriate for XLogSetAsyncXactLSN(). Another reason to write the WAL aggressively, even if you don't flush it, would be to reduce the number of lost transactions on a segfault. But we don't give any guarantees on that, and even with the current aggressive logic, we only write when a page is full so you're anyway going to lose the last partial page. It also took me a while to convince myself that this calculation matches the calculation that XLogBackgroundFlush() uses to determine whether it needs to flush or not. XLogBackgroundFlush() first divides the request and result with XLOG_BLCKSZ and then compares the number of blocks, whereas here you perform the calculation in bytes. I think the result is the same, but to make it more clear, let's do it the same way in both places. See attached. It's the same logic as in your patch, just formulatd more clearly IMHO. > Because of the frequent wakeups, we do something else that's not smart: We > write out 8k blocks individually, many times a second. Often thousands of > 8k pwrites a second. Even with this patch, when I bumped up wal_writer_delay to 2 so that the wal writer gets woken up by the async commits rather than the timeout, the write pattern is a bit silly: $ strace -p 1099926 # walwriter strace: Process 1099926 attached epoll_wait(10, [{events=EPOLLIN, data={u32=3704011232, u64=94261056289248}}], 1, 1991) = 1 read(3, "\27\0\0\0\0\0\0\0\0\0\0\0<\312\20\0\350\3\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 1024) = 128 pwrite64(5, "\24\321\5\0\1\0\0\0\0\300\0\373\5\0\0\0+\0\0\0\0\0\0\0\0\n\0\0n\276\242\305"..., 1007616, 49152) = 1007616 fdatasync(5) = 0 pwrite64(5, "\24\321\5\0\1\0\0\0\0 \20\373\5\0\0\0003\0\0\0\0\0\0\0\320\37\20\373\5\0\0\0"..., 16384, 1056768) = 16384 epoll_wait(10, [{events=EPOLLIN, data={u32=3704011232, u64=94261056289248}}], 1, 2000) = 1 read(3, "\27\0\0\0\0\0\0\0\0\0\0\0<\312\20\0\350\3\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 1024) = 128 pwrite64(5, "\24\321\5\0\1\0\0\0\0`\20\373\5\0\0\0+\0\0\0\0\0\0\0\0\n\0\0\5~\23\261"..., 1040384, 1073152) = 1040384 fdatasync(5) = 0 pwrite64(5, "\24\321\4\0\1\0\0\0\0@ \373\5\0\0\0\0\0\0\0\0\0\0\0;\0\0\0\264'\246\3"..., 16384, 2113536) = 16384 epoll_wait(10, [{events=EPOLLIN, data={u32=3704011232, u64=94261056289248}}], 1, 2000) = 1 read(3, "\27\0\0\0\0\0\0\0\0\0\0\0<\312\20\0\350\3\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 1024) = 128 pwrite64(5, "\24\321\5\0\1\0\0\0\0\200 \373\5\0\0\0003\0\0\0\0\0\0\0\320\177 \373\5\0\0\0"..., 1040384, 2129920) = 1040384 fdatasync(5) = 0 In each cycle, the wal writer writes a full 1 MB chunk (wal_writer_flush_after = '1MB'), flushes it, and then perform a smaller write before going to sleep. Those smaller writes seem a bit silly. But I think it's fine. > More throughput for less CPU, seems neat :) Indeed, impressive speedup from such a small patch! > I'm not addressing that here, but I think we also have the opposite behaviour > - we're not waking up walwriter often enough. E.g. if you have lots of bulk > dataloads, walwriter will just wake up once per wal_writer_delay, leading to > most of the work being done by backends. We should probably wake walsender at > the end of XLogInsertRecord() if there is sufficient outstanding WAL. Right, that's basically the same issue that I reasoned through above. I did some quick testing with a few different settings of wal_buffers, wal_writer_flush_after and wal_writer_delay, to try to see that effect. But I was not able to find a case where it makes a difference. -- Heikki Linnakangas Neon (https://neon.tech) -
Re: walwriter interacts quite badly with synchronous_commit=off
Andres Freund <andres@anarazel.de> — 2023-10-25T18:59:41Z
Hi, On 2023-10-25 12:17:03 +0300, Heikki Linnakangas wrote: > On 25/10/2023 02:09, Andres Freund wrote: > > Because of the inherent delay between the checks of XLogCtl->WalWriterSleeping > > and Latch->is_set, we also sometimes end up with multiple processes signalling > > walwriter, which can be bad, because it increases the likelihood that some of > > the signals may be received when we are already holding WALWriteLock, delaying > > its release... > > That can only happen when walwriter has just come out of "hibernation", ie. > when the system has been idle for a while. So probably not a big deal in > practice. Maybe I am missing something here - why can this only happen when hibernating? Even outside of that two backends can decide that that they need to wake up walwriter? We could prevent that, by updating state when requesting walwriter to be woken up. But with the changes we're discussing below, that should be rare. > > I think the most important optimization we need is to have > > XLogSetAsyncXactLSN() only wake up if there is a certain amount of unflushed > > WAL. Unless walsender is hibernating, walsender will wake up on its own after > > wal_writer_delay. I think we can just reuse WalWriterFlushAfter for this. > > > > E.g. a condition like > > if (WriteRqstPtr <= LogwrtResult.Write + WalWriterFlushAfter * XLOG_BLCKSZ) > > return; > > drastically cuts down on the amount of wakeups, without - I think - loosing > > guarantees around synchronous_commit=off. > > In the patch, you actually did: > > > + if (WriteRqstPtr <= LogwrtResult.Flush + WalWriterFlushAfter * XLOG_BLCKSZ) > > + return; > > It means that you never wake up the walwriter to merely *write* the WAL. You > only wake it up if it's going to also fsync() it. I think that's correct and > appropriate, but it took me a while to reach that conclusion: Yea, after writing the email I got worried that just looking at Write would perhaps lead to not flushing data soon enough... > It might be beneficial to wake up the walwriter just to perform a write(), > to offload that work from the backend. And walwriter will actually also > perform an fsync() after finishing the current segment, so it would make > sense to also wake it up when 'asyncXactLSN' crosses a segment boundary. > However, if those extra wakeups make sense, they would also make sense when > there are no asynchronous commits involved. Therefore those extra wakeups > should be done elsewhere, perhaps somewhere around AdvanceXLInsertBuffer(). > The condition you have in the patch is appropriate for > XLogSetAsyncXactLSN(). Yea. I agree we should wake up walsender in other situations too... > Another reason to write the WAL aggressively, even if you don't flush it, > would be to reduce the number of lost transactions on a segfault. But we > don't give any guarantees on that, and even with the current aggressive > logic, we only write when a page is full so you're anyway going to lose the > last partial page. Wal writer does end up writing the trailing partially filled page during the next wal_writer_delay cycle. > It also took me a while to convince myself that this calculation matches the > calculation that XLogBackgroundFlush() uses to determine whether it needs to > flush or not. XLogBackgroundFlush() first divides the request and result > with XLOG_BLCKSZ and then compares the number of blocks, whereas here you > perform the calculation in bytes. I think the result is the same, but to > make it more clear, let's do it the same way in both places. > > See attached. It's the same logic as in your patch, just formulatd more > clearly IMHO. Yep, makes sense! > > Because of the frequent wakeups, we do something else that's not smart: We > > write out 8k blocks individually, many times a second. Often thousands of > > 8k pwrites a second. > > Even with this patch, when I bumped up wal_writer_delay to 2 so that the wal > writer gets woken up by the async commits rather than the timeout, the write > pattern is a bit silly: > > $ strace -p 1099926 # walwriter > strace: Process 1099926 attached > epoll_wait(10, [{events=EPOLLIN, data={u32=3704011232, > u64=94261056289248}}], 1, 1991) = 1 > read(3, > "\27\0\0\0\0\0\0\0\0\0\0\0<\312\20\0\350\3\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., > 1024) = 128 > pwrite64(5, "\24\321\5\0\1\0\0\0\0\300\0\373\5\0\0\0+\0\0\0\0\0\0\0\0\n\0\0n\276\242\305"..., > 1007616, 49152) = 1007616 > fdatasync(5) = 0 > pwrite64(5, "\24\321\5\0\1\0\0\0\0 > \20\373\5\0\0\0003\0\0\0\0\0\0\0\320\37\20\373\5\0\0\0"..., 16384, 1056768) > = 16384 > epoll_wait(10, [{events=EPOLLIN, data={u32=3704011232, > u64=94261056289248}}], 1, 2000) = 1 > read(3, > "\27\0\0\0\0\0\0\0\0\0\0\0<\312\20\0\350\3\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., > 1024) = 128 > pwrite64(5, > "\24\321\5\0\1\0\0\0\0`\20\373\5\0\0\0+\0\0\0\0\0\0\0\0\n\0\0\5~\23\261"..., > 1040384, 1073152) = 1040384 > fdatasync(5) = 0 > pwrite64(5, "\24\321\4\0\1\0\0\0\0@ > \373\5\0\0\0\0\0\0\0\0\0\0\0;\0\0\0\264'\246\3"..., 16384, 2113536) = 16384 > epoll_wait(10, [{events=EPOLLIN, data={u32=3704011232, > u64=94261056289248}}], 1, 2000) = 1 > read(3, > "\27\0\0\0\0\0\0\0\0\0\0\0<\312\20\0\350\3\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., > 1024) = 128 > pwrite64(5, "\24\321\5\0\1\0\0\0\0\200 \373\5\0\0\0003\0\0\0\0\0\0\0\320\177 > \373\5\0\0\0"..., 1040384, 2129920) = 1040384 > fdatasync(5) = 0 > > In each cycle, the wal writer writes a full 1 MB chunk > (wal_writer_flush_after = '1MB'), flushes it, and then perform a smaller > write before going to sleep. I think that's actually somewhat sane - we write out the partial page in the subsequent cycle. That won't happen if the page isn't partially filled or doesn't have an async commit on it. I think we end up with somewhat bogus write patterns in other cases still, but that's really more an issue in XLogBackgroundFlush() and thus deserves a separate patch/thread. > > I'm not addressing that here, but I think we also have the opposite behaviour > > - we're not waking up walwriter often enough. E.g. if you have lots of bulk > > dataloads, walwriter will just wake up once per wal_writer_delay, leading to > > most of the work being done by backends. We should probably wake walsender at > > the end of XLogInsertRecord() if there is sufficient outstanding WAL. > > Right, that's basically the same issue that I reasoned through above. I did > some quick testing with a few different settings of wal_buffers, > wal_writer_flush_after and wal_writer_delay, to try to see that effect. But > I was not able to find a case where it makes a difference. I think in the right set of circumstances it can make quite a bit of difference. E.g. I bulk load 3GB of data in a cluster with s_b 1GB. Then I checkpoint and VACUUM FREEZE it. With wal_writer_delay=1ms that's considerably faster (5.4s) than with wal_writer_delay=2s (8.3s) or even the default 200ms (7.9s), because a fast walwriter makes it much more likely that vacuum won't need to wait for an xlog flush before replacing a buffer in the strategy ring. I think improving this logic would be quite worthwhile! Another benefit of triggering wakeups based on the amount of outstanding writes would be that we could increase wal_writer_delay substantially (with perhaps some adjustment for the partial-trailing-page-with-async-commit case), reducing power usage. It's imo pretty silly that we have wal writer wake up regularly, if it just writes once every few seconds. Greetings, Andres Freund -
Re: walwriter interacts quite badly with synchronous_commit=off
Heikki Linnakangas <hlinnaka@iki.fi> — 2023-10-25T20:04:28Z
On 25/10/2023 21:59, Andres Freund wrote: > On 2023-10-25 12:17:03 +0300, Heikki Linnakangas wrote: >> On 25/10/2023 02:09, Andres Freund wrote: >>> Because of the inherent delay between the checks of XLogCtl->WalWriterSleeping >>> and Latch->is_set, we also sometimes end up with multiple processes signalling >>> walwriter, which can be bad, because it increases the likelihood that some of >>> the signals may be received when we are already holding WALWriteLock, delaying >>> its release... >> >> That can only happen when walwriter has just come out of "hibernation", ie. >> when the system has been idle for a while. So probably not a big deal in >> practice. > > Maybe I am missing something here - why can this only happen when hibernating? > Even outside of that two backends can decide that that they need to wake up > walwriter? Ah sure, multiple backends can decide to wake up walwriter at the same time. I thought you meant that the window for that was somehow wider when XLogCtl->WalWriterSleeping. > We could prevent that, by updating state when requesting walwriter to be woken > up. But with the changes we're discussing below, that should be rare. One small easy thing we could do to reduce the redundant wakeups: only wake up walwriter if asyncXactLSN points to different page than prevAsyncXactLSN. -- Heikki Linnakangas Neon (https://neon.tech)
-
Re: walwriter interacts quite badly with synchronous_commit=off
Heikki Linnakangas <hlinnaka@iki.fi> — 2023-11-27T15:55:34Z
On 25/10/2023 21:59, Andres Freund wrote: >> See attached. It's the same logic as in your patch, just formulatd more >> clearly IMHO. > Yep, makes sense! Pushed this. Thanks for the investigation! -- Heikki Linnakangas Neon (https://neon.tech)
-
Re: walwriter interacts quite badly with synchronous_commit=off
Andres Freund <andres@anarazel.de> — 2023-11-27T17:13:59Z
On 2023-11-27 17:55:34 +0200, Heikki Linnakangas wrote: > On 25/10/2023 21:59, Andres Freund wrote: > > > See attached. It's the same logic as in your patch, just formulatd more > > > clearly IMHO. > > Yep, makes sense! > > Pushed this. Thanks for the investigation! Thanks!