Hi, RAID developers!
I noticed that write-mostly logic in current kernel is broken.
It seems that read_balance() always chooses write-mostly disk when one exists,
unless other normal disk happens to have zero outstanding requests.
This patch fixes it - tested on 3.13.7 but should apply cleanly to git trunk.
BTW good_sectors logic looks broken too, but I couldn't figure out
what that code is supposed to do, so no fix for that.
I think that the following commit broke it:
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/drivers/md/raid1.c?id=9dedf60313fa4dddfd5b9b226a0ef12a512bf9dc
"best" logic was split to "best_dist" and "best_pending" but no change
was made for write-mostly branch and for good_sectors too.
~
:wq
With best regards,
Vladimir Savkin.
--- linux-3.10.33/drivers/md/raid1.c.orig 2014-03-16 01:11:43.000000000 +0400
+++ linux-3.10.33/drivers/md/raid1.c 2014-03-16 01:23:31.000000000 +0400
@@ -498,6 +498,8 @@
int sectors;
int best_good_sectors;
int best_disk, best_dist_disk, best_pending_disk;
+ int writemostly_disk;
+ int writemostly_good_sectors;
int has_nonrot_disk;
int disk;
sector_t best_dist;
@@ -519,7 +521,9 @@
best_dist = MaxSector;
best_pending_disk = -1;
min_pending = UINT_MAX;
+ writemostly_disk = -1;
best_good_sectors = 0;
+ writemostly_good_sectors = 0;
has_nonrot_disk = 0;
choose_next_idle = 0;
@@ -548,16 +552,16 @@
if (test_bit(WriteMostly, &rdev->flags)) {
/* Don't balance among write-mostly, just
* use the first as a last resort */
- if (best_disk < 0) {
+ if (writemostly_disk < 0) {
if (is_badblock(rdev, this_sector, sectors,
&first_bad, &bad_sectors)) {
if (first_bad < this_sector)
/* Cannot use this */
continue;
- best_good_sectors = first_bad - this_sector;
+ writemostly_good_sectors = first_bad - this_sector;
} else
- best_good_sectors = sectors;
- best_disk = disk;
+ writemostly_good_sectors = sectors;
+ writemostly_disk = disk;
}
continue;
}
@@ -664,6 +668,14 @@
best_disk = best_dist_disk;
}
+ /*
+ * If there is still no good disk, try write-mostly.
+ */
+ if ( (best_disk == -1) && (writemostly_disk >= 0) ) {
+ best_disk = writemostly_disk;
+ best_good_sectors = writemostly_good_sectors;
+ }
+
if (best_disk >= 0) {
rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
if (!rdev)
