Tuning kern.maxswzone is minor compared to hangs in "kmem a" state
Eugene Grosbein
eugen at grosbein.net
Mon Feb 3 05:02:53 UTC 2014
On 03.02.2014 02:46, Matthew Rezny wrote:
> Just as I was about to hit send, the box panicked. Finally a panic and
> not just a hang so there's a backtrace! That should be the first hint
> as to where in the kernel I'm hitting trouble. The panic points the
> finger at UFS. The filesystem has seen some thrashing but it did just
> fsck without significant error before that build was started. So, I'm
> thinking it's not on-disk corruption but an in-memory problem. It blew
> up writing a softdep, which starts me wondering if the buffers used for
> softupdates bookkeeping are overflowing in the kernel. That would make
> some sense, rapidly creating (svn/tar) or deleting (rm) many small
> files will create a huge amount of activity for softupdates to track,
> defer and reorder. The rm case is probably the one than can trigger it
> the fastest because deletes are focusing the beating on the metadata
> without wasting time between on file data I/O.
See also http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/176857
More information about the freebsd-stable
mailing list