From: Jonathan Brassow <jbrassow@redhat.com>
Date: Mon, 17 Dec 2007 16:13:31 -0600
Subject: [md] dm-mirror: shedule_timeout call causes slowdown
Message-id: 1197929611.28090.13.camel@hydrogen
O-Subject: [RHEL 5.2 PATCH] dm-mirror: shedule_timeout() call causing big process slowdowns
Bugzilla: 358881

Bug #358881:

The small sleep in drivers/md/dm-raid1.c:do_work() was designed to:
1) Allow the userspace process to more easily proceed to reconfigure a
mirror in the event of a failure.
2) Allow remote machines (cluster mirror) to finish recovery of a region
before resubmitting I/O.

However, the sleep is having adverse effects on nominal I?o - sometimes
slowing processes down by greater than a factor of 10.  This can make
applications appear hung.

This problem occurs most frequently in cluster mirrors under heavy I/O
(or during failure scenarios).  It is possible in single machine
mirroring, but hasn't been seen yet.

This bug does not corrupt data, but it certainly affects "availability"
- which is especially bad considering mirrors are used when you want to
increase availability.

 brassow

Acked-by: Alasdair G Kergon <agk@redhat.com>

diff --git a/drivers/md/dm-raid1.c b/drivers/md/dm-raid1.c
index d4ea9cb..b22ab8f 100644
--- a/drivers/md/dm-raid1.c
+++ b/drivers/md/dm-raid1.c
@@ -1293,10 +1293,8 @@ static int do_mirror(struct mirror_set *ms)
 
 static void do_work(void *data)
 {
-	while (do_mirror(data)) {
-		set_current_state(TASK_INTERRUPTIBLE);
-		schedule_timeout(HZ/5);
-	}
+	while (do_mirror(data))
+		schedule();
 }
 
 /*-----------------------------------------------------------------