[syslog-ng] disk full memory leak

[Alexander Clouter]

Hi,

I hope I did not miss the memo but I think I have stumbled on a memory leak; 
due to my own poor administration :)

When a partition fills up and syslog-ng cannot write out log files to, it 
seems to start soaking up memory.  Looking in the output from 'top' you can 
see the resident memory size ('RES') creeping up and not being released.  It 
seems to be at the rate of logging messages themselves.

I'm guessing this bug has not been seen mainly as whilst syslog-ng soaks up 
memory unattended Linux will kill it once it soaks up all the available RAM; 
probably people would just assume "ahhhh syslog-ng gracefully died when it 
could not write out to the disk", I know I would :)

It is easy to reproduce, I did it by creating a small partition of about 10MB 
and letting it fill up (you can hurry it along by dd'ing 8MB's worth of 
/dev/zero to the partition as a file).  Then whilst running top you can see 
the memory creeping up the moment the partition fills.

I found that after a day of "max'ed out partition time" syslog-ng got as high 
as 512MB of resident memory!  I am pretty sure it is not configuration file 
related but if you would like a copy do ask and I will send it 'offline'.

I'm running the regular 2.0.5 tarball from the website with the Debian 
related package patches applied from 2.0.0-1[1].

If this bug is fixed at least I will be able to lvextend/xfs_growfs a 
partition to 'gracefully' recover....maybe syslog-ng could flush this data 
polling regularly to see if it can write the data at a later time?

Cheers

Alex

[1] http://ftp.debian.org/debian/pool/main/s/syslog-ng/syslog-ng_2.0.0-1.diff.gz