On Oct 21, 2013, at 2:40 PM, Keith Wesolowski <[email protected]> wrote:
> On Mon, Oct 14, 2013 at 01:22:45PM -0700, Matthew Ahrens wrote: > >> commit 9dcb97198338ba2d8764dd5604b278118612f74d >> Author: Brian Behlendorf <[email protected]> >> Date: Thu Oct 25 13:02:31 2012 -0700 >> >> Log I/Os longer than zio_delay_max (30s default) >> >> seems reasonable; I'd want to get input from others who are more familiar >> with FMA. > > This doesn't seem relevant: we don't even have zio_delay_max, and in > illumos if an I/O never completes we hit the deadman and panic (which > seems right to me, since I don't see how you recover from this). I'd > also be curious to know what's being done with this telemetry other than > exposure via 'zfs events'. I can see where people want to know when I/Os take a suspiciously long time. I could also argue that 30 seconds is an arbitrary and probably not useful guess for the general case. Of course, the default in Solaris (sd) is 60 seconds, which is equally arbitrary and arguably useless. The question is: does this log I/Os that have not completed in 30 seconds, or does it log I/Os that have completed in more than 30 seconds? The difference is subtle, but important. -- richard -- [email protected] +1-760-896-4422
_______________________________________________ developer mailing list [email protected] http://lists.open-zfs.org/mailman/listinfo/developer
