Date: Thu, 21 Jul 2011 22:04:06 +0200 From: Ivan Voras <ivoras@freebsd.org> To: Luiz Otavio O Souza <lists.br@gmail.com> Cc: freebsd-fs@freebsd.org Subject: Re: ZFS and large directories - caveat report Message-ID: <CAF-QHFUf_PV0RQ5sUJhnZGD04-BSbE0_mNW%2BshBGB-7YaPz-1w@mail.gmail.com> In-Reply-To: <CAF-QHFVGWO9-sxZAw-sFyoVZVTs=HQN0AX=4OpeaGv4_qLhWLw@mail.gmail.com> References: <j09hk8$svj$1@dough.gmane.org> <13577F3E-DE59-44F4-98F7-9587E26499B8@gmail.com> <CAF-QHFUs7OAAwcFvSymc3YYX3Cdq%2BQUUSP_4OFXCegCbzLkOUA@mail.gmail.com> <E3F8D6EC-2CCA-49A2-9BFD-982BF01A3531@bitgravity.com> <5542D910-0C5C-4B2B-885F-CC92901367F0@gmail.com> <CAF-QHFVGWO9-sxZAw-sFyoVZVTs=HQN0AX=4OpeaGv4_qLhWLw@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 21 July 2011 21:45, Ivan Voras <ivoras@freebsd.org> wrote: > On 21 July 2011 21:36, Luiz Otavio O Souza <lists.br@gmail.com> wrote: > >> But i guess i need to increase the arc_meta_limit as well: >> >> vfs.zfs.arc_meta_limit: 536870912 >> vfs.zfs.arc_meta_used: 579461312 > > You also have arc_meta_used larger than arc_meta_limit ... but not > nearly as big a difference as on my system. > > Can anyone speculate if raising vfs.zfs.arc_meta_limit would help? Well, it didn't help me - I raised it above what used to be arc_meta_used and after the reboot arc_meta_used simply rose again over arc_meta_limit. Here's another "symptom": while "find" is running, I do a ls of my home directory on the same zpool and get delays, always at the same place (after the postgresql source file): www:~> ll total 233 drwxrwxr-x 2 ivoras ivoras 8 Jun 1 2009 backup/ -rw-rw-r-- 1 ivoras ivoras 593 Nov 7 2007 c1.php -rw-rw-r-- 1 ivoras ivoras 37682863 Apr 30 2009 cms.tgz drwxrwxr-x 4 ivoras ivoras 4 Feb 12 2008 devel/ -rw-r--r-- 1 ivoras ivoras 44372 May 24 2007 etcdirs.tgz -rw-r--r-- 1 root ivoras 215397 Nov 22 2007 lock_profile.txt -rw-r--r-- 1 ivoras ivoras 18336 Nov 21 2007 lockmgr.diff -rw-rw-r-- 1 ivoras ivoras 32590585 Oct 31 2007 melc.sql -rw-r----- 1 ivoras ivoras 1712 Oct 15 2008 newreq.pem -rw-rw-r-- 1 root ivoras 3330572 Apr 30 2009 postgresql-server-8.3.1.tbz load: 0.38 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 3.25r 0.00u 0.01s 0% 2140k load: 0.67 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 5.00r 0.00u 0.01s 0% 2140k load: 0.67 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 5.47r 0.00u 0.01s 0% 2140k load: 0.67 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 6.34r 0.00u 0.01s 0% 2140k load: 0.67 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 8.30r 0.00u 0.01s 0% 2140k load: 0.67 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 8.70r 0.00u 0.01s 0% 2140k load: 0.67 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 9.17r 0.00u 0.01s 0% 2140k load: 0.70 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 11.08r 0.00u 0.01s 0% 2140k load: 0.70 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 11.54r 0.00u 0.01s 0% 2140k load: 0.70 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 13.00r 0.00u 0.01s 0% 2140k load: 0.70 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 13.49r 0.00u 0.01s 0% 2140k load: 0.70 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 14.10r 0.00u 0.01s 0% 2140k load: 0.64 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 14.62r 0.00u 0.01s 0% 2140k load: 0.64 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 17.41r 0.00u 0.01s 0% 2140k load: 0.64 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 19.74r 0.00u 0.01s 0% 2140k load: 0.75 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 20.53r 0.00u 0.01s 0% 2140k load: 0.75 cmd: ls 1786 [tx->tx_quiesce_done_cv)] 21.30r 0.00u 0.01s 0% 2140k lrwxrwxr-x 1 ivoras ivoras 9 Apr 23 2009 services@ -> /services -rw-rw-r-- 1 ivoras ivoras 160060550 Oct 31 2007 services.tgz -rw-rw-r-- 1 ivoras ivoras 0 Nov 1 2007 stress.txt drwxrwxr-x 5 ivoras ivoras 8 Mar 10 2010 temp/ -rw-rw-r-- 1 ivoras ivoras 257002 Oct 31 2007 ule.tgz -rw-r--r-- 1 ivoras ivoras 8965978 May 24 2007 wwwdirs.tgz The "load.." lines in between are me hitting Ctrl-T for process info. Observe almost 20 seconds delay in the middle of "ls"!
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAF-QHFUf_PV0RQ5sUJhnZGD04-BSbE0_mNW%2BshBGB-7YaPz-1w>