Re: swap space leak in 5.0_STABLE

[Mark Davies <mark%ecs.vuw.ac.nz@localhost>]

On Monday 10 August 2009 10:18:51 matthew green wrote:
>    Already tried that and no change.  Also this is happening on some
>    machines that don't run X servers (the one I mentioned that went from
>    10% to 27% in 1/2 hour was one such - and its now at 66% after
>    another 8 hours).
>
> what's happening on this machine?

This particular machine is a xen domu that is used by students to remotely 
log in (ssh and vnc) and "run things".  It died an hour or so after 
sending the previous so I took the opportunity to put its kernel back to 
the Jun 26 version but it hasn't helped - after 12 hours its at

Device      1K-blocks     Used    Avail Capacity  Priority
/dev/xbd0b    2097152  2097144        8   100%    0

with these current processes

  UID  PID PPID  CPU PRI NI   VSZ   RSS WCHAN    STAT TTY        TIME 
COMMAND
    0    0    0    0 125  0     0 11044 schedule DKl  ?       0:38.75 
[system]
    0    1    0    0  85  0  2996   204 wait     Is   ?       0:00.02 init
    0  123    1    0  85  0  2992   532 kqueue   Ss   ?       
0:00.20 /usr/sbin/syslogd -s
    0  149    1  471  85  0 40296  4032 sigwait  Isl  ?       
0:00.70 /usr/sbin/named
    0  153    1    0  85  0  3112   788 select   Is   ?       
0:00.12 /usr/sbin/rpcbind -l
    0  196    1    0  85  0  5140  9236 select   Ss   ?       
0:07.91 /usr/sbin/amd -l syslog -x error,nostats -c 900 -r -p -a /am
    0  223    1    0  85  0  3024   200 select   Ss   ?       
0:00.04 /usr/sbin/rpc.lockd
    0  231    1    0  85  0  3024   200 select   Is   ?       
0:00.04 /usr/sbin/rpc.statd
    0  236    1    0  85  0  5684  5880 pause    Ss   ?       
0:00.74 /usr/sbin/ntpd -A
    0  280    1 2511  85  0  3016     4 kqueue   IWs  ?       
0:00.00 /usr/sbin/powerd
    0  311    1    0  85  0  6308   628 select   Is   ?       
0:00.19 /usr/sbin/sshd
    0  336    1    0  85  0  3036   292 kqueue   Is   ?       
0:00.03 /usr/sbin/inetd -l
  110  357    1    0  85  0  7208     4 select   IWs  ?       
0:00.01 /usr/pkg/bin/dbus-daemon --system
    0  362    1    0  85  0  2960   484 nanoslp  Is   ?       
0:00.07 /usr/sbin/cron
    0 4624  362    0  85  0  2960  1052 piperd   I    ?       0:00.00 cron: 
running job
    0 6752  311    0  85  0  6476  4024 select   Ss   ?       0:00.03 sshd: 
root@pts/0 (sshd)
    0 4540 6752    0  85  0  2208  2016 pause    Ss   ttyp0   0:00.03 -tcsh
    0 4646 4540    0  43  0  3024   872 -        O+   ttyp0   0:00.00 
ps -laxw
25787  258    1    0  85  0 15808   364 select   I    ttyp1 - 0:00.92 
Xvnc :1 -desktop X -httpd /usr/pkg/share/vnc/classes -auth /u
    0  358    1 1517  85  0  2972     4 ttyraw   IWs+ xencons 
0:00.00 /usr/libexec/getty Pc console


The "stuff" thats been run today according to lastcomm is mostly java and 
javac and various editors.

Since reverting the kernel hasn't helped trying to think what else is 
different in the last week or so on this and the other machines showing 
the issue and the only thing I can come up with is that we put sun-jdk14 
back on the systems (rather than the sun-jdk15 and native 1.5 or 1.6 
normally used) to use with jmlspecs 
(http://sourceforge.net/projects/jmlspecs/files/) and this appears to be 
the java thats predominantly happening on the above machine today.

So is this a case of some situation where a (linux?) process can terminate 
but the memory is not released for reuse that is being tickled by this 
particular use of jmlspecs/sun-jdk14.

I've tried to trigger the issue personally today but haven't managed to do 
it.

The particular 2nd year assignment that is using jmlspecs is due tomorrow 
so will be intersting to see if the problem continues after that.

cheers
mark