Обсуждение: PGSQL, checkpoints, and file system syncs
Hello PGSQL performance community,
You might remember that I pinged you in July 2012 to introduce the TPC-V benchmark. I am now back with more data, and a question about checkpoints. As far as the plans for the benchmark, we are hoping to release a benchmarking kit for multi-VM servers this year (and of course one can always simply configure it to run on one database)
I am now dealing with a situation of performance dips when checkpoints complete. To simplify the discussion, I have reproduced the problem on a single VM/single database.
Complete config info is in the attached files. Briefly, it is a 6-vCPU VM with 91G of memory, and 70GB in PGSQL shared buffers. The host has 512GB of memory and 4 sockets of Westmere (E7-4870) processors with HT enabled.
The data tablespace is on an ext4 file system on a (virtual) disk which is striped on 16 SSD drives in RAID 0. This is obviously overkill for the load we are putting on this one VM, but in the usual benchmarking config, the 16 SSDs are shared by 24 VMs. Log is on an ext3 file system on 4 spinning drives in RAID 1.
We are running PGSQL version 9.2 on RHEL 6.4; here are some parameters of interest (postgresql.conf is in the attachment):
checkpoint_segments = 1200
checkpoint_timeout = 360s
checkpoint_completion_target = 0.8
wal_sync_method = open_datasync
wal_buffers = 16MB
wal_writer_delay = 10ms
effective_io_concurrency = 10
effective_cache_size = 1024MB
When running tests, I noticed that when a checkpoint completes, we have a big burst of writes to the data disk. The log disk has a very steady write rate that is not affected by checkpoints except for the known phenomenon of more bytes in each log write when a new checkpoint period starts. In a multi-VM config with all VMs sharing the same data disks, when these write bursts happen, all VMs take a hit.
So I set out to see what causes this write burst. After playing around with PGSQL parameters and observing its behavior, it appears that the bursts aren’t produced by the database engine; they are produced by the file system. I suspect PGSQL has to issue a sync(2)/fsync(2)/sync_file_range(2) system call at the completion of the checkpoint to ensure that all blocks are flushed to disk before creating a checkpoint marker. To test this, I ran a loop to call sync(8) once a second.
The graphs in file “run280.mht” have the throughput, data disk activity, and checkpoint start/completion timestamps for the baseline case. You can see that the checkpoint completion, the write burst, and the throughput dip all occur at the same time, so much so that it is hard to see the checkpoint completion line under the graph of writes. It looks like the file system does a mini flush every 30 seconds. The file “run274.mht” is the case with sync commands running in the background. You can see that everything is more smooth.
Is there something I can set in the PGSQL parameters or in the file system parameters to force a steady flow of writes to disk rather than waiting for a sync system call? Mounting with “commit=1” did not make a difference.
Thanks,
Reza
Вложения
On 04/03/2014 08:39 PM, Reza Taheri wrote: > Hello PGSQL performance community, > You might remember that I pinged you in July 2012 to introduce the TPC-V benchmark. I am now back with more data, and aquestion about checkpoints. As far as the plans for the benchmark, we are hoping to release a benchmarking kit for multi-VMservers this year (and of course one can always simply configure it to run on one database) > > I am now dealing with a situation of performance dips when checkpoints complete. To simplify the discussion, I have reproducedthe problem on a single VM/single database. > > Complete config info is in the attached files. Briefly, it is a 6-vCPU VM with 91G of memory, and 70GB in PGSQL sharedbuffers. The host has 512GB of memory and 4 sockets of Westmere (E7-4870) processors with HT enabled. > > The data tablespace is on an ext4 file system on a (virtual) disk which is striped on 16 SSD drives in RAID 0. This isobviously overkill for the load we are putting on this one VM, but in the usual benchmarking config, the 16 SSDs are sharedby 24 VMs. Log is on an ext3 file system on 4 spinning drives in RAID 1. > > We are running PGSQL version 9.2 on RHEL 6.4; here are some parameters of interest (postgresql.conf is in the attachment): > checkpoint_segments = 1200 > checkpoint_timeout = 360s > checkpoint_completion_target = 0.8 > wal_sync_method = open_datasync > wal_buffers = 16MB > wal_writer_delay = 10ms > effective_io_concurrency = 10 > effective_cache_size = 1024MB > > When running tests, I noticed that when a checkpoint completes, we have a big burst of writes to the data disk. The logdisk has a very steady write rate that is not affected by checkpoints except for the known phenomenon of more bytes ineach log write when a new checkpoint period starts. In a multi-VM config with all VMs sharing the same data disks, whenthese write bursts happen, all VMs take a hit. > > So I set out to see what causes this write burst. After playing around with PGSQL parameters and observing its behavior,it appears that the bursts aren't produced by the database engine; they are produced by the file system. I suspectPGSQL has to issue a sync(2)/fsync(2)/sync_file_range(2) system call at the completion of the checkpoint to ensurethat all blocks are flushed to disk before creating a checkpoint marker. To test this, I ran a loop to call sync(8)once a second. > > The graphs in file "run280.mht" have the throughput, data disk activity, and checkpoint start/completion timestamps forthe baseline case. You can see that the checkpoint completion, the write burst, and the throughput dip all occur at thesame time, so much so that it is hard to see the checkpoint completion line under the graph of writes. It looks like thefile system does a mini flush every 30 seconds. The file "run274.mht" is the case with sync commands running in the background.You can see that everything is more smooth. > > Is there something I can set in the PGSQL parameters or in the file system parameters to force a steady flow of writesto disk rather than waiting for a sync system call? Mounting with "commit=1" did not make a difference. Try setting the vm.dirty_bytes sysctl. Something like 256MB might be a good starting point. This comes up fairly often, see e.g.: http://www.postgresql.org/message-id/flat/27C32FD4-0142-44FE-8488-9F366DC75966@mr-paradox.net - Heikki
> Try setting the vm.dirty_bytes sysctl. Something like 256MB might be a good > starting point. > > This comes up fairly often, see e.g.: > http://www.postgresql.org/message-id/flat/27C32FD4-0142-44FE-8488- > 9F366DC75966@mr-paradox.net > > - Heikki Thanks, Heikki. That sounds like my problem alright. I will play with these parameters right away, and will report back. Cheers, Reza
Hi Reza, vm.dirty_bytes indeed makes sense, but just in case: how exactly is your ext4 mount? Particularly, have you disabled barrier? Ilya On Thu, Apr 3, 2014 at 8:11 PM, Reza Taheri <rtaheri@vmware.com> wrote: >> Try setting the vm.dirty_bytes sysctl. Something like 256MB might be a good >> starting point. -- Ilya Kosmodemiansky, PostgreSQL-Consulting.com tel. +14084142500 cell. +4915144336040 ik@postgresql-consulting.com
Hi Ilya, I mount the esx4 file system with: nofail,noatime,nodiratime,nobarrier,commit=1 The esxt3 file system for the log is mounted with: nofail,noatime,nodiratime,data=writeback Attached is a tar file with charts from last night's runs. File 0_dirty_background_bytes.htm has the default with dirty_background_bytes=0and dirty_background_ratio=10 (it's a warm-up run, so it starts with low throughput and a high readrate). I still see a little bit of burstiness with dirty_background_bytes=50MB, but no burstiness with dirty_background_ratiofrom 10MB down to even 1MB. Thanks everyone for the advice, Reza > -----Original Message----- > From: Ilya Kosmodemiansky [mailto:ilya.kosmodemiansky@postgresql- > consulting.com] > Sent: Thursday, April 03, 2014 11:12 PM > To: Reza Taheri > Cc: Heikki Linnakangas; pgsql-performance@postgresql.org > Subject: Re: [PERFORM] PGSQL, checkpoints, and file system syncs > > Hi Reza, > > vm.dirty_bytes indeed makes sense, but just in case: how exactly is your > ext4 mount? Particularly, have you disabled barrier? > > Ilya > > On Thu, Apr 3, 2014 at 8:11 PM, Reza Taheri <rtaheri@vmware.com> wrote: > >> Try setting the vm.dirty_bytes sysctl. Something like 256MB might be > >> a good starting point. > > > > -- > Ilya Kosmodemiansky, > > PostgreSQL-Consulting.com > tel. +14084142500 > cell. +4915144336040 > ik@postgresql-consulting.com
Вложения
On Thu, Apr 3, 2014 at 09:01:08PM +0300, Heikki Linnakangas wrote: > >Is there something I can set in the PGSQL parameters or in the file > >system parameters to force a steady flow of writes to disk rather > >than waiting for a sync system call? Mounting with "commit=1" did not > >make a difference. > > Try setting the vm.dirty_bytes sysctl. Something like 256MB might be a > good starting point. > > This comes up fairly often, see e.g.: > http://www.postgresql.org/message-id/flat/27C32FD4-0142-44FE-8488-9F36 > 6DC75966@mr-paradox.net Uh, should he set vm.dirty_bytes or vm.dirty_background_bytes? It is my understanding that vm.dirty_background_bytes starts the I/O while still accepting writes, while vm.dirty_bytes stops accepting writes during the I/O, which isn't optimal. See: https://www.kernel.org/doc/Documentation/sysctl/vm.txt dirty_bytes Contains the amount of dirty memory at which a process generating disk writes will itself start writeback. dirty_background_bytes Contains the amount of dirty memory at which the background kernel flusher threads will start writeback. I think we want the flusher to be active, not necessarly the writing process. -- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB http://enterprisedb.com + Everyone has their own god. +