Обсуждение: BUG #10734: PostgreSQL 9.3.4 shutdown forever in zfsonlinux 0.6.3-1 filesystem

The following bug has been logged on the website:

Bug reference:      10734
Logged by:          digoal.zhou
Email address:      digoal@126.com
PostgreSQL version: 9.3.4
Operating system:   CentOS 6.5 x64
Description:

I have a machine DELL R720xd, and 1 ssd and 12 4TB SATA disks.
OS: CentOS 6.5 x64
ZFS: 0.6.3 installed by yum.
PostgreSQL 9.3.4.
zpool desc (12 SATA disks strip as vdev), (1 ssd partition as SLOG 2GB, and
another partition as L2ARC 800GB)
zpool ashift use 12. (4KB).
zfs nondefault parameters :
zp1/data01  recordsize            128K                   local
zp1/data01  mountpoint            /data01                local
zp1/data01  checksum              off                    local
zp1/data01  compression           lz4                    local
zp1/data01  primarycache          metadata               local

I use this dataset as $PGDATA.
and insert 500000000 row test data, use update by pkey and pgbench test it.
after all , checkpoint, vacuum freeze the table.
and then shutdown -m fast.
but shutdown very slow, about 10 mins.
and strace checkpoint_pid, I see some lseek and write system calls.
why?
it's a postgresql bug?

On 23 Červen 2014, 9:17, digoal@126.com wrote:
> it's a postgresql bug?

I doubt that. Most likely it's caused by the zfs filesystem, which has
interesting features but it's not exactly for free. Try to use some other
'traditional' filesystem (e.g. ext4 or xfs) and check how long it takes
with it.

There are ways to tune zfs, but pgsql-bugs is not the right list to
discuss that. Please ask in pgsql-performance.

regards
Tomas