Matthew Dillon wrote:
:Hi, : :I wrote a script that generates 1 million files in one directory on the :hammer filesystem. The first 100k are created very quickly, then it :starts to get less predictive. It stops completely after creating 836k :files. I can still ping the machine, but I can't ssh into it any more. :It's a head-less system so I can tell what is going on exactly. : :I'm using the attached C file like this: : : cr 1000000 test : :Regards, : : MichaelOooh, nice test. Mine got to 389000 before it deadlocked in the buffer cache. I'll be able to get this fixed today. It looks like a simple issue.
Commit 61/A fixes the problem. But now, after creating around 3 million files and doing a "cpdup /usr/pkgsrc /hammer/pkgsrc", running "make head-src-cvsup" turned out to be extremely slow (actually it was the "cvs update" part). Then I did a "rm -rf /hammer/dir-with-a-million-files" and hammer finally died :) It probably core dumped :( I can try to reproduce it tomorrow with a display connected to it. Another thing I noticed is that when there is a lot of file system activity, other user processes are slowed down a lot (maybe they are just blocked on a condition). At least that's how it feels. Regards, Michael
