ZFS and mem management
Pavlo
devgs at ukr.net
Thu Feb 16 08:03:12 UTC 2012
Interesting stuff follows:
>From log: Feb 16 04:28:39 zebra-test kernel: pid 29258 (thread), uid 8,
was killed: out of swap space
Now let's see top log for a night :
Second before:
------------------------------------------------------------------------------------------------------------------------
Thu Feb 16 04:28:38 EET 2012
last pid: 31341; load averages: 0.04, 0.05, 0.00 up 0+15:25:11
04:28:38
61 processes: 1 running, 60 sleeping
Mem: 2514M Active, 2058M Inact, 3109M Wired, 142M Cache, 96K Buf, 106M
Free
Swap:
PID USERNAME THR PRI NICE SIZE RES STATE C TIME WCPU COMMAND
29258 mail 1 45 0 37960K 6064K zio->i 3 0:08 2.10% thread
1750 root 1 44 0 25024K 1772K select 0 4:24 0.10% snmpd
1967 root 1 76 0 5716K 892K piperd 1 1:09 0.00% sh
1913 gs 1 76 0 9704K 1328K pause 0 0:41 0.00% zsh
15920 gs 1 44 0 9400K 2028K select 1 0:33 0.00% top
71345 gs 1 44 0 38132K 936K select 2 0:29 0.00% sshd
1889 gs 1 44 0 9100K 620K nanslp 2 0:13 0.00% iostat
1886 gs 1 44 0 38132K 848K select 3 0:12 0.00% sshd
52290 gs 1 44 0 5844K 856K kqread 0 0:05 0.00% tail
1902 gs 1 44 0 38132K 1044K select 3 0:04 0.00% sshd
1891 satan 1 44 0 9396K 688K select 3 0:04 0.00% screen
1357 nagios 1 44 0 6908K 512K select 0 0:02 0.00% nrpe2
1153 root 1 44 0 6936K 748K select 1 0:01 0.00%
syslogd
1632 root 1 44 0 5428K 432K nanslp 0 0:00 0.00% cron
1912 gs 1 44 0 38132K 792K select 1 0:00 0.00% sshd
71346 gs 1 44 0 9692K 1452K pause 0 0:00 0.00% zsh
1791 root 1 44 0 7992K 372K nanslp 0 0:00 0.00% cron
1903 gs 1 44 0 9704K 1968K pause 2 0:00 0.00% zsh
------------------------------------------------------------------------------------------------------------------------
PID 29258 is still alive (thread is that indexing tool). Also there is a
lot of free and inactive memory. And thread can't possibly ask for a big
piece of memory because of the nature of this tool.
Next second we see:
------------------------------------------------------------------------------------------------------------------------
Thu Feb 16 04:28:39 EET 2012
last pid: 31362; load averages: 0.04, 0.05, 0.00 up 0+15:25:12
04:28:39
60 processes: 1 running, 59 sleeping
Mem: 2512M Active, 2055M Inact, 3092M Wired, 145M Cache, 96K Buf, 125M
Free
Swap:
PID USERNAME THR PRI NICE SIZE RES STATE C TIME WCPU COMMAND
15920 gs 1 44 0 9400K 2028K select 3 0:33 0.10% top
1750 root 1 44 0 25024K 1772K select 0 4:24 0.00% snmpd
1967 root 1 64 0 5716K 892K zio->i 0 1:09 0.00% sh
1913 gs 1 76 0 9704K 1328K pause 0 0:41 0.00% zsh
71345 gs 1 44 0 38132K 936K select 2 0:29 0.00% sshd
1889 gs 1 44 0 9100K 620K nanslp 2 0:13 0.00% iostat
1886 gs 1 44 0 38132K 848K select 3 0:12 0.00% sshd
52290 gs 1 44 0 5844K 856K kqread 0 0:05 0.00% tail
1902 gs 1 44 0 38132K 1044K select 0 0:04 0.00% sshd
1891 satan 1 44 0 9396K 688K select 3 0:04 0.00% screen
1357 nagios 1 44 0 6908K 512K select 2 0:02 0.00% nrpe2
1153 root 1 44 0 6936K 748K select 2 0:01 0.00%
syslogd
1632 root 1 44 0 5428K 432K nanslp 0 0:00 0.00% cron
1912 gs 1 44 0 38132K 792K select 1 0:00 0.00% sshd
71346 gs 1 44 0 9692K 1452K pause 0 0:00 0.00% zsh
1791 root 1 44 0 7992K 372K nanslp 0 0:00 0.00% cron
1903 gs 1 44 0 9704K 1968K pause 2 0:00 0.00% zsh
1497 root 1 44 0 5396K 708K select 0 0:00 0.00%
syslogd
------------------------------------------------------------------------------------------------------------------------
It is dead although inactive and active queues of memory didn't even get
touched. Please tell me, is that appropriate behaviour? And why this
never happens on Linux for example (and maybe on other systems as well)?
More information about the freebsd-fs
mailing list