Summary: Root cause of swap space full

Hi Managers,

I've got only 4 replies, I'm still not very sure what the root cause
was. It's hard to do forensic analysis when the problem can't be
repeated. (well, thank goodness). Thanks for all of you! Here's what
I've got:

From: Andrew_Cordova 

The only way I could see that this particular file system causing the
crash would be if it had not been mounted. If you had say 4 disks, two
active  w/2 mirrors.   /weblogs is mounted from active disk 2 and for
some reason it is offline. what you have now is  just the /weblogs
mount point, w/ no disk mounted under it. 

The app is stupid, it just writes to where it is told, which is
/weblogs, which in this scenario is just the mount point w/ no
underlying disk.  if /weblogs was full, then effectively / is full. 
just as if /home was not a separate file system and a user was to fill
it up, then again, w/o quotas enabled, they have filled /. 

i have not seen this scenario w/ a fs created on rootdg, but have seen
it when a datadg mount that we had lost the disk, the app just kept
writting to /mountpoint till it ran us out of space. /tmp eventually
runs out of space because it is effectively part of / and you have a
down system, reboot clears /tmp - you login and clean space. 

an unusual scenario for sure, but as you know in this business the
unusual is the norm. one other item to chk is that if / is still
unusually full, then you may have /weblogs mounted over the data still
written under /weblogs the mount point. 

From Darren Dunham                                           

I'm assuming that the filesystem you mention is a VxFS filesystem.  I
can't think of any reason that filling it would cause a swap issue.
Instead your likely reasons are..

1) Running too many processes.
2) Running a process with a memory leak.
3) Writing too much in a tmpfs filesystem.

From Hichael Morton"  

solaris 8 permits logins, etc. to a machine with full partitions.
even with only 1 partition, logins are permitted.
of course, no writes can take place.

solaris eventually takes control of all the swap space and distributes
as needed to processes.  free swap space, over time, will be reduced
normally.  (there are posts in the archives about swap and how the os
manages swap.)

the full file system would not cause the problem you discribed though
you should check for memory leaks, etc.

From "Haywood, Steven" 

Possibly people were dumping stuff in /tmp ? This filesystem normally 
shares space with swap...

--- Meg Wall <meg991@yahoo.com> wrote:
> Hi Managers,
> 
> We had a server hung last night because of swap space
> full. This box had one of non-cratical file system
> full for about one day, (vx_nospace -
> /dev/vx/dsk/rootdg/weblogs_lv file system full) and
> another team didn't take care. This server has 4G swap
> and plenty memory, so people argue that file system
> full is not the root cause of the server crash. I want
> to say it is, but can't convince people with the
> details. Why the swap full then? 
> 
> Thanks! I will summarize.
> 
> Meg
Yahoo! Tax Center - File online, calculators, forms, and more
http://tax.yahoo.com
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers