Revert part of "Log I/Os longer than zio_delay_max (30s default)"
authorMatt Johnston <matt@fugro-fsi.com.au>
Fri, 21 Dec 2012 02:15:34 +0000 (10:15 +0800)
committerBrian Behlendorf <behlendorf1@llnl.gov>
Mon, 7 Jan 2013 18:51:04 +0000 (10:51 -0800)
commit72f53c5694bfd57b56e79fc3b7c4390bce4072d2
treef71ac31075af2f780e11b00276aa272ced5921ff
parent15f9d4e1c2b0ee26546e79d599df1b2adc2e5026
Revert part of "Log I/Os longer than zio_delay_max (30s default)"

This reverts commit 9dcb97198338ba2d8764dd5604b278118612f74
which was originally introduced to debug occasional slow I/Os.
These I/Os would complete eventually but were observed to take
several 100 seconds.

The root cause of this issue was the CFQ scheduler which can,
under certain conditions, excessively delay an I/O from being
issued to the device.  This issue was mitigated somewhat by
commit 84daaddedbfc9cf4bd1490d8a6f4b2967051e308 which ensures
the I/O elevator gets changed even for DM style devices.

This change isn't in any way harmful but it does conflict with
a required change to properly account from I/O wait time.
Because Linux does not export the io_schedule_timeout() function
we must instead rely  on io_schedule() via cv_wait_io().

The additional debugging information which was added to the
delay event has been intentionally left in place.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
module/zfs/zio.c