On 10/09/11 14:57, Daniel Pittman wrote:

On Fri, Sep 9, 2011 at 21:41, Craig Sanders<cas@taz.net.au> wrote:

...

On Sat, Sep 10, 2011 at 12:30:07PM +1000, Tim Connors wrote:

...

...

I woke up yesterday morning to find my system load avg at over 100 - due [...] worse, it crashed later even without mlocate running. backuppc scheduled

It's true, the linux VMM is a bit shit. A load of 100 is just a slight bit crap.

all waiting for IO. it's hardly linux' fault that some process is going apeshit creating a hard-link farm.

What puzzles me in this discussion is that linking and deleting, in BackupPC, are a single-threaded operation. One Perl process running, chewing through one file system operation after another. So, while I have no doubt it caused this, it feels like there is some underlying problem lurking in there, waiting for the next time something has a similar meta-data heavy set of operations happening...

For what it's worth, we've seen issues with BackupPC causing excessive laod and nearly killing machines in a manner much like this too. I wasn't investigating that one so can't provide much useful info, other than another data point of "it does happen". Toby