RE: [cobalt-developers] RaQ3 memory leak problems?

["Matthew Nuzum" <cobalt@xxxxxxxxxxxxx>]

> At 10 am a cronjob of user root runs the PERL programm FCHECK with
> nicelevel
> -15 which parses the filesystem for changes. That one always has the
> effect
> to free up a lot of memory for unknown reasons. So between 6am (end of
> cron.daily) and 10 am (start of FCHECK) the server is always hovering
> close
> to a crash with next to no free memory going around.
> 
> > What are you running in cron daily, that might cause the problem?
> 
> That's very hard to tell. Aside from the usual stuff there is
Webalizer,
> plus
> a shellscript which fetches the latest anti-virus definitions from
> Kaspersky
> Labs and a shellscript which dumps the MySQL database to a backup file
on
> the
> hardisk.
> 
> Disabling Webalizer and the MySQL-backup didn't do a difference (tried
> that a
> while ago), so I'm a little clueless here.
> 

It appears that heavy I/O processes cause memory to be freed up.  That
sounds funny, because usually if you have any free memory it's sucked
into the disk cache.

Either way, the common thread appears to be disk I/O which leads me to
the question, can you trust MRTG in this case as accurately judging free
mem usage?

Is it possible that it's seeing memory as free if it's in use by the
disk cache?  For example, the output of free says:
             total       used       free     shared    buffers
cached
Mem:       1673360    1589856      83504       1916     218112
1142900
-/+ buffers/cache:     228844    1444516
Swap:      1046224      12436    1033788

If you read this you might think that I have only 84 MB of Ram
available, but if PostgreSQL get's hit on this server, the -/+
buffers/cache: free will get used, effectively giving me plenty of RAM.
(BTW, this is my db server, not a cobalt)

My point is, maybe the problem is still there, even when the MRTG graph
is reporting lots of free memory.  P.S. I've not used compiled
programming languages in quite some time, so I'm not much help with real
debugging.

Matthew Nuzum
www.bearfruit.org
cobalt@xxxxxxxxxxxxx