|
 |
cb8e9e |
From 0d2dd0b1395397fced99b62dd826d4ab18fb94f1 Mon Sep 17 00:00:00 2001
|
|
|
cb8e9e |
From: Xavier Hernandez <xhernandez@datalab.es>
|
|
|
cb8e9e |
Date: Fri, 7 Aug 2015 12:37:52 +0200
|
|
|
cb8e9e |
Subject: [PATCH 318/320] cluster/ec: Fix write size in self-heal
|
|
|
cb8e9e |
|
|
|
cb8e9e |
Self-heal was always using a fixed block size to heal a file. This
|
|
|
cb8e9e |
was incorrect for dispersed volumes with a number of data bricks not
|
|
|
cb8e9e |
being a power of 2.
|
|
|
cb8e9e |
|
|
|
cb8e9e |
This patch adjusts the block size to a multiple of the stripe size
|
|
|
cb8e9e |
of the volume. It also propagates errors detected during the data
|
|
|
cb8e9e |
heal to stop healing the file and not mark it as healed.
|
|
|
cb8e9e |
|
|
|
cb8e9e |
> Change-Id: I9ee3fde98a9e5d6116fd096ceef88686fd1d28e2
|
|
|
cb8e9e |
> BUG: 1251446
|
|
|
cb8e9e |
> Signed-off-by: Xavier Hernandez <xhernandez@datalab.es>
|
|
|
cb8e9e |
> Reviewed-on: http://review.gluster.org/11862
|
|
|
cb8e9e |
> Tested-by: NetBSD Build System <jenkins@build.gluster.org>
|
|
|
cb8e9e |
> Tested-by: Gluster Build System <jenkins@build.gluster.com>
|
|
|
cb8e9e |
> Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
cb8e9e |
|
|
|
cb8e9e |
3.7: http://review.gluster.org/11869
|
|
|
cb8e9e |
|
|
|
cb8e9e |
BUG: 1241862
|
|
|
cb8e9e |
Change-Id: I3bbe6ed6ff60b3efcc08d7425678bb9aeb5ddb11
|
|
|
cb8e9e |
Signed-off-by: Xavier Hernandez <xhernandez@datalab.es>
|
|
|
cb8e9e |
Reviewed-on: https://code.engineering.redhat.com/gerrit/56691
|
|
|
cb8e9e |
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
cb8e9e |
Tested-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
cb8e9e |
---
|
|
|
cb8e9e |
tests/bugs/disperse/bug-1251446.t | 50 +++++++++++++++++++++++++++++++++++++
|
|
|
cb8e9e |
xlators/cluster/ec/src/ec-data.h | 1 +
|
|
|
cb8e9e |
xlators/cluster/ec/src/ec-heal.c | 9 ++++++
|
|
|
cb8e9e |
3 files changed, 60 insertions(+), 0 deletions(-)
|
|
|
cb8e9e |
create mode 100644 tests/bugs/disperse/bug-1251446.t
|
|
|
cb8e9e |
|
|
|
cb8e9e |
diff --git a/tests/bugs/disperse/bug-1251446.t b/tests/bugs/disperse/bug-1251446.t
|
|
|
cb8e9e |
new file mode 100644
|
|
|
cb8e9e |
index 0000000..f805539
|
|
|
cb8e9e |
--- /dev/null
|
|
|
cb8e9e |
+++ b/tests/bugs/disperse/bug-1251446.t
|
|
|
cb8e9e |
@@ -0,0 +1,50 @@
|
|
|
cb8e9e |
+#!/bin/bash
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+. $(dirname $0)/../../include.rc
|
|
|
cb8e9e |
+. $(dirname $0)/../../volume.rc
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+cleanup
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+TEST glusterd
|
|
|
cb8e9e |
+TEST pidof glusterd
|
|
|
cb8e9e |
+TEST $CLI volume create $V0 disperse 4 redundancy 1 $H0:$B0/${V0}{0..3}
|
|
|
cb8e9e |
+TEST $CLI volume start $V0
|
|
|
cb8e9e |
+TEST glusterfs --entry-timeout=0 --attribute-timeout=0 -s $H0 --volfile-id $V0 $M0
|
|
|
cb8e9e |
+EXPECT_WITHIN $CHILD_UP_TIMEOUT "4" ec_child_up_count $V0 0
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+TEST dd if=/dev/urandom of=$M0/test1 bs=1024k count=2
|
|
|
cb8e9e |
+cs=$(sha1sum $M0/test1 | awk '{ print $1 }')
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+TEST kill_brick $V0 $H0 $B0/${V0}0
|
|
|
cb8e9e |
+EXPECT '3' online_brick_count
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+TEST cp $M0/test1 $M0/test2
|
|
|
cb8e9e |
+EXPECT "$cs" echo $(sha1sum $M0/test2 | awk '{ print $1 }')
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+TEST $CLI volume start $V0 force
|
|
|
cb8e9e |
+EXPECT '4' online_brick_count
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT "[0-9][0-9]*" get_shd_process_pid
|
|
|
cb8e9e |
+TEST $CLI volume heal $V0 full
|
|
|
cb8e9e |
+EXPECT_WITHIN $HEAL_TIMEOUT "0" get_pending_heal_count $V0
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+EXPECT "699392" stat -c "%s" $B0/${V0}0/test2
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+# force cache clear
|
|
|
cb8e9e |
+EXPECT_WITHIN $UMOUNT_TIMEOUT "Y" force_umount $M0
|
|
|
cb8e9e |
+TEST $CLI volume stop $V0
|
|
|
cb8e9e |
+TEST $CLI volume start $V0
|
|
|
cb8e9e |
+TEST glusterfs --entry-timeout=0 --attribute-timeout=0 -s $H0 --volfile-id $V0 $M0
|
|
|
cb8e9e |
+EXPECT_WITHIN $CHILD_UP_TIMEOUT "4" ec_child_up_count $V0 0
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+TEST kill_brick $V0 $H0 $B0/${V0}3
|
|
|
cb8e9e |
+EXPECT '3' online_brick_count
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+EXPECT "$cs" echo $(sha1sum $M0/test2 | awk '{ print $1 }')
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+## cleanup
|
|
|
cb8e9e |
+EXPECT_WITHIN $UMOUNT_TIMEOUT "Y" force_umount $M0
|
|
|
cb8e9e |
+TEST $CLI volume stop $V0
|
|
|
cb8e9e |
+TEST $CLI volume delete $V0
|
|
|
cb8e9e |
+
|
|
|
cb8e9e |
+cleanup;
|
|
|
cb8e9e |
diff --git a/xlators/cluster/ec/src/ec-data.h b/xlators/cluster/ec/src/ec-data.h
|
|
|
cb8e9e |
index ec470e9..1008706 100644
|
|
|
cb8e9e |
--- a/xlators/cluster/ec/src/ec-data.h
|
|
|
cb8e9e |
+++ b/xlators/cluster/ec/src/ec-data.h
|
|
|
cb8e9e |
@@ -285,6 +285,7 @@ struct _ec_heal
|
|
|
cb8e9e |
fd_t *fd;
|
|
|
cb8e9e |
int32_t partial;
|
|
|
cb8e9e |
int32_t done;
|
|
|
cb8e9e |
+ int32_t error;
|
|
|
cb8e9e |
gf_boolean_t nameheal;
|
|
|
cb8e9e |
uintptr_t available;
|
|
|
cb8e9e |
uintptr_t good;
|
|
|
cb8e9e |
diff --git a/xlators/cluster/ec/src/ec-heal.c b/xlators/cluster/ec/src/ec-heal.c
|
|
|
cb8e9e |
index a7c97a5..fde7e31 100644
|
|
|
cb8e9e |
--- a/xlators/cluster/ec/src/ec-heal.c
|
|
|
cb8e9e |
+++ b/xlators/cluster/ec/src/ec-heal.c
|
|
|
cb8e9e |
@@ -1779,6 +1779,7 @@ ec_heal_block_done (call_frame_t *frame, void *cookie, xlator_t *this,
|
|
|
cb8e9e |
|
|
|
cb8e9e |
fop->heal = NULL;
|
|
|
cb8e9e |
heal->fop = NULL;
|
|
|
cb8e9e |
+ heal->error = op_ret < 0 ? op_errno : 0;
|
|
|
cb8e9e |
syncbarrier_wake (heal->data);
|
|
|
cb8e9e |
return 0;
|
|
|
cb8e9e |
}
|
|
|
cb8e9e |
@@ -1789,6 +1790,9 @@ ec_sync_heal_block (call_frame_t *frame, xlator_t *this, ec_heal_t *heal)
|
|
|
cb8e9e |
ec_heal_block (frame, this, heal->bad|heal->good, EC_MINIMUM_ONE,
|
|
|
cb8e9e |
ec_heal_block_done, heal);
|
|
|
cb8e9e |
syncbarrier_wait (heal->data, 1);
|
|
|
cb8e9e |
+ if (heal->error != 0) {
|
|
|
cb8e9e |
+ return -heal->error;
|
|
|
cb8e9e |
+ }
|
|
|
cb8e9e |
if (heal->bad == 0)
|
|
|
cb8e9e |
return -ENOTCONN;
|
|
|
cb8e9e |
return 0;
|
|
|
cb8e9e |
@@ -1814,6 +1818,11 @@ ec_rebuild_data (call_frame_t *frame, ec_t *ec, fd_t *fd, uint64_t size,
|
|
|
cb8e9e |
pool = ec->xl->ctx->iobuf_pool;
|
|
|
cb8e9e |
heal->total_size = size;
|
|
|
cb8e9e |
heal->size = iobpool_default_pagesize (pool);
|
|
|
cb8e9e |
+ /* We need to adjust the size to a multiple of the stripe size of the
|
|
|
cb8e9e |
+ * volume. Otherwise writes would need to fill gaps (head and/or tail)
|
|
|
cb8e9e |
+ * with existent data from the bad bricks. This could be garbage on a
|
|
|
cb8e9e |
+ * damaged file or it could fail if there aren't enough bricks. */
|
|
|
cb8e9e |
+ heal->size -= heal->size % ec->stripe_size;
|
|
|
cb8e9e |
heal->bad = ec_char_array_to_mask (healed_sinks, ec->nodes);
|
|
|
cb8e9e |
heal->good = ec_char_array_to_mask (sources, ec->nodes);
|
|
|
cb8e9e |
heal->iatt.ia_type = IA_IFREG;
|
|
|
cb8e9e |
--
|
|
|
cb8e9e |
1.7.1
|
|
|
cb8e9e |
|