cluster: RHEL510 - HA LVM should only remove missing PVs on stop when they belong to mirrors
Ryan McCabe
rmccabe at fedoraproject.org
Tue May 21 03:47:00 UTC 2013
Gitweb: http://git.fedorahosted.org/git/?p=cluster.git;a=commitdiff;h=d45a4fe2fa265a19a4130161caa4ce92367f9072
Commit: d45a4fe2fa265a19a4130161caa4ce92367f9072
Parent: 1d7a0335016cae0246f455a707f78bfc5d1ff880
Author: John Ruemker <jruemker at redhat.com>
AuthorDate: Tue May 14 15:19:19 2013 -0500
Committer: Ryan McCabe <rmccabe at redhat.com>
CommitterDate: Mon May 20 23:46:49 2013 -0400
HA LVM should only remove missing PVs on stop when they belong to mirrors
This adds --mirrorsonly to the 3 'vgreduce --removemissing' calls in the
LVM agents.
You'll also notice that it adds another self_fence check after we fail to
remove tags. In my previous comment, I pointed out that in the case of
single-host by_lv, after we vgreduce we then can't deactivate the logical
volume again because it doesn't exist. This results in us executing
self_fence, which may have just been a happy accident. But when we avoid
making metadata changes by adding --mirrorsonly, the subsequent deactivation
is still successful, and thus we miss the self_fence logic. So, I added
another check so we still catch the failure and fence ourselves in this
situation.
Resolves: rhbz#962376
Signed-off-by: John Ruemker <jruemker at redhat.com>
Signed-off-by: Jonthan Brassow <jbrassow at redhat.com>
Signed-off-by: Ryan McCabe <rmccabe at redhat.com>
---
rgmanager/src/resources/lvm_by_lv.sh | 14 ++++++++++----
rgmanager/src/resources/lvm_by_vg.sh | 4 ++--
2 files changed, 12 insertions(+), 6 deletions(-)
diff --git a/rgmanager/src/resources/lvm_by_lv.sh b/rgmanager/src/resources/lvm_by_lv.sh
index 4971173..7140076 100644
--- a/rgmanager/src/resources/lvm_by_lv.sh
+++ b/rgmanager/src/resources/lvm_by_lv.sh
@@ -243,13 +243,19 @@ lv_activate_and_tag()
# be removed from the VG via a separate call before
# the tag can be removed.
ocf_log err "Attempting volume group clean-up and retry"
- vgreduce --removemissing --force $OCF_RESKEY_vg_name
+ vgreduce --removemissing --mirrorsonly --force $OCF_RESKEY_vg_name
# Retry tag deletion
lvchange --deltag $tag $lv_path
if [ $? -ne 0 ]; then
- ocf_log err "Failed to delete tag from $lv_path"
- return $OCF_ERR_GENERIC
+ if [ "$self_fence" ]; then
+ ocf_log err "Failed to delete tag from $lv_path: REBOOTING"
+ sync
+ reboot -fn
+ else
+ ocf_log err "Failed to delete tag from $lv_path"
+ fi
+ return $OCF_ERR_GENERIC
fi
fi
@@ -322,7 +328,7 @@ lv_activate()
ocf_log notice "Attempting cleanup of $OCF_RESKEY_vg_name"
- if vgreduce --removemissing --force --config \
+ if vgreduce --removemissing --mirrorsonly --force --config \
"activation { volume_list = \"$OCF_RESKEY_vg_name\" }" \
$OCF_RESKEY_vg_name; then
ocf_log notice "$OCF_RESKEY_vg_name now consistent"
diff --git a/rgmanager/src/resources/lvm_by_vg.sh b/rgmanager/src/resources/lvm_by_vg.sh
index 0dd2aaa..819d0b8 100755
--- a/rgmanager/src/resources/lvm_by_vg.sh
+++ b/rgmanager/src/resources/lvm_by_vg.sh
@@ -202,7 +202,7 @@ function vg_start_clustered
ocf_log err "Failed to activate volume group, $OCF_RESKEY_vg_name"
ocf_log notice "Attempting cleanup of $OCF_RESKEY_vg_name"
- if ! vgreduce --removemissing --force $OCF_RESKEY_vg_name; then
+ if ! vgreduce --removemissing --mirrorsonly --force $OCF_RESKEY_vg_name; then
ocf_log err "Failed to make $OCF_RESKEY_vg_name consistent"
return $OCF_ERR_GENERIC
fi
@@ -398,7 +398,7 @@ function vg_stop_single
# Shut down the volume group
# Do we need to make this resilient?
- vgchange -an $OCF_RESKEY_vg_name
+ vgchange -aln $OCF_RESKEY_vg_name
# Make sure all the logical volumes are inactive
results=(`lvs -o name,attr --noheadings $OCF_RESKEY_vg_name 2> /dev/null`)
More information about the cluster-commits
mailing list