156 lines
4.6 KiB
Diff
156 lines
4.6 KiB
Diff
commit 69217b67c0d018f129c7cbf526aebf0b236be701
|
|
Author: Chris Friesen <chris.friesen@windriver.com>
|
|
Date: Thu Sep 17 15:26:16 2015 -0400
|
|
|
|
CGCS-2553/CGTS-2534: tweak LVM success criteria
|
|
|
|
It turns out that activating an LVM LV which has a snapshot (or activating
|
|
the snapshot) will take an amount of time that is proportional to the
|
|
delta between the snapshot and the original volume.
|
|
|
|
Because of this it's possible that running "vgchange" could take a long
|
|
time, since it also activates the LVs.
|
|
|
|
If this happens, rather than timeout the whole script we want to log which
|
|
LVs/snapshots havn't yet been activated, and then just continue on.
|
|
Accordingly, we want to set the internal timeout in the "start" operation
|
|
to something less than the timeout for the "start" action.
|
|
|
|
There will be corresponding changes in cinder to properly handle this case.
|
|
|
|
diff --git a/heartbeat/LVM b/heartbeat/LVM
|
|
index bd1a47a..24b0244 100755
|
|
--- a/heartbeat/LVM
|
|
+++ b/heartbeat/LVM
|
|
@@ -186,6 +186,81 @@ LVM_monitor() {
|
|
}
|
|
|
|
#
|
|
+# Activate one volume explicitly.
|
|
+#
|
|
+activate_volume() {
|
|
+ ocf_run lvchange $1 /dev/${2}/$3
|
|
+ if [ $? -eq 0 ] ; then
|
|
+ ocf_log info "Succesfully activated $LV."
|
|
+ else
|
|
+ ocf_log err "Problem activating $LV."
|
|
+ fi
|
|
+}
|
|
+
|
|
+#
|
|
+# Kick off parallel activation of all volumes
|
|
+#
|
|
+activate_all_volumes() {
|
|
+ VG=$1
|
|
+ shift
|
|
+ lvchange_args="$*"
|
|
+
|
|
+ # Get the list of volumes, without the first line which is column headings.
|
|
+ VOLS=`lvs $VG |tail -n +2`
|
|
+
|
|
+ while read -r LINE; do
|
|
+ # Convert the line into an array.
|
|
+ LINE_ARRAY=($LINE)
|
|
+
|
|
+ # First array element is the volume/snapshot name.
|
|
+ LV=${LINE_ARRAY[0]}
|
|
+
|
|
+ # Third array element is the attributes.
|
|
+ ATTR=${LINE_ARRAY[2]}
|
|
+
|
|
+ # Fifth character in the attributes is "a" if it's active.
|
|
+ ACTIVE=${ATTR:4:1}
|
|
+ if [ "$ACTIVE" == "a" ]; then
|
|
+ ocf_log info "$LV is already active."
|
|
+ continue
|
|
+ fi
|
|
+
|
|
+ SNAPSHOT_ORIGIN=${LINE_ARRAY[4]}
|
|
+ if [ "$SNAPSHOT_ORIGIN" != "" ] ; then
|
|
+ # If this is a snapshot, don't activate it.
|
|
+ continue
|
|
+ fi
|
|
+
|
|
+ ( activate_volume "$*" $VG $LV ) &
|
|
+ done <<< "$VOLS"
|
|
+}
|
|
+
|
|
+#
|
|
+# Scan for inactive volumes and log any that are found.
|
|
+#
|
|
+log_inactive_volumes() {
|
|
+ # Get the list of volumes, without the first line which is column headings.
|
|
+ VOLS=`lvs $1 |tail -n +2`
|
|
+
|
|
+ while read -r LINE; do
|
|
+ # Convert the line into an array.
|
|
+ LINE_ARRAY=($LINE)
|
|
+
|
|
+ # First array element is the volume/snapshot name.
|
|
+ LV=${LINE_ARRAY[0]}
|
|
+
|
|
+ # Third array element is the attributes.
|
|
+ ATTR=${LINE_ARRAY[2]}
|
|
+
|
|
+ # Fifth character in the attributes is "a" if it's active.
|
|
+ ACTIVE=${ATTR:4:1}
|
|
+ if [ "$ACTIVE" != "a" ]; then
|
|
+ ocf_log err "Volume $LV is not active after expiry of timeout."
|
|
+ fi
|
|
+ done <<< "$VOLS"
|
|
+}
|
|
+
|
|
+#
|
|
# Enable LVM volume
|
|
#
|
|
LVM_start() {
|
|
@@ -218,7 +293,47 @@ LVM_start() {
|
|
vgchange_options="$vgchange_options --monitor y"
|
|
fi
|
|
|
|
- ocf_run vgchange $vgchange_options $1 || return $OCF_ERR_GENERIC
|
|
+ # Kick off activation of all volumes. If it doesn't complete within
|
|
+ # the timeout period, then we'll log the not-yet-activated volumes and
|
|
+ # continue on.
|
|
+ (ocf_run vgchange $vgchange_options $1) & PID=$!
|
|
+
|
|
+ # Check every second for up to TIMEOUT seconds whether the vgchange has
|
|
+ # completed.
|
|
+ TIMEOUT=300
|
|
+ TIMED_OUT=true
|
|
+ SECONDS=0;
|
|
+ PARALLEL_ACTIVATE_DELAY=10
|
|
+ PARALLEL_ACTIVATE_DONE=false
|
|
+ while [ $SECONDS -lt $TIMEOUT ] ; do
|
|
+ kill -0 $PID &> /dev/null
|
|
+ if [ $? -eq 1 ] ; then
|
|
+ # process with pid of $PID doesn't exist, vgchange command completed
|
|
+ TIMED_OUT=false
|
|
+ break
|
|
+ fi
|
|
+ if [ $SECONDS -ge $PARALLEL_ACTIVATE_DELAY ] && \
|
|
+ [ "$PARALLEL_ACTIVATE_DONE" != true ] && \
|
|
+ [ "$1" == "cinder-volumes" ] ; then
|
|
+ # This will kick off parallel activation of all LVs in the VG.
|
|
+ # The delay is to ensure the VG is activated first.
|
|
+ PARALLEL_ACTIVATE_DONE=true
|
|
+ ocf_log info Explicitly activating all volumes in $1 with: $vgchange_options
|
|
+ activate_all_volumes $1 $vgchange_options
|
|
+ fi
|
|
+ sleep 1
|
|
+ done
|
|
+
|
|
+ if [ "$TIMED_OUT" = true ] ; then
|
|
+ ocf_log err "Timed out running ocf_run vgchange $vgchange_options $1"
|
|
+ log_inactive_volumes $1
|
|
+ else
|
|
+ # Child process completed, get its status.
|
|
+ wait $PID
|
|
+ if [ $? -ne 0 ] ; then
|
|
+ return $OCF_ERR_GENERIC
|
|
+ fi
|
|
+ fi
|
|
|
|
if LVM_status $1; then
|
|
: OK Volume $1 activated just fine!
|