Version: 1.0.5-1 Architecture: amd64 Hi,
I'm seeing the same problem here too. I'm performing AMANDA tape backups, using pbzip2 to compress the output from 'tar' before it goes to the AMANDA taper program. When a tape becomes full (after 200GB of compressed data is written, which takes several hours), the AMANDA taper program will stop reading output from pbzip2 until I've changed the tape. But during this time pbzip2 rapidly allocates memory. If I don't change the tape within a few minutes, all virtual memory is exhausted by pbzip2, which is then killed by the kernel OOM killer, causing the backup to fail. Interestingly, if I'm really quick at changing the tape over, pbzip2 seems to resume normally, and its memory usage gradually drops back down to a normal level. I've been looking for a quick way to reproduce this problem, and have so far come up with the attached shell script. Be careful, your system may crash! Do not run it as root! It takes less than 10 minutes to exhaust 4GB memory on my 4-way SMP system. Regards, -- Steven Chamberlain ste...@pyro.eu.org
pbzip2-bug-537048.sh
Description: Bourne shell script