On Tue, 2008-09-09 at 22:45 +0200, Ralf Hildebrandt wrote: > * Timo Sirainen <[EMAIL PROTECTED]>: > > > Wonder if there are some race condition issues that don't show up with > > Core 2 duo, but do show up with 4 core / 2 CPUs. I know there are some > > things I should probably not be doing (or should maybe add some memory > > barrier code).
Could you try running these on the server for a while: http://dovecot.org/tmp/memory-concurrent-rw.c http://dovecot.org/tmp/memory-concurrent-rw-pread.c I wrote them a while ago to test this memory behavior. They shouldn't print anything if they're running successfully. > > How often do you see those errors? > > Just once so far. It's been more than once since I've seen you post the mail_index_sync_ext_reset backtrace at least 3 times. :) But I actually meant any errors (related to index files), not just this. > > Does mmap_disable=yes change it? > > What's the performance impact of that? Shouldn't be much, although it could use somewhat more memory since it's trying to emulate mmap().
signature.asc
Description: This is a digitally signed message part