[DRBD-user] [patch] Wait for all DRBD devices to be registered with udev

[Christopher Boumenot]

I am using DRBD in combination with Heartbeat, and I've noticed that on 
occasion not all of the DRBD devices are properly configured.  I've 
tracked the issue down to DRBD and udev.  When DRBD waits for udev to 
register its devices it only waits for the very first device to be 
registered.  This is fine if you have one device, but if you have more 
devices the scripts can incorrectly continue execution.  As a result 
Heartbeat's init script are executed, and when it attempts to mount a 
DRBD backed partition it fails.

After Hearbeat's init script has finished executing the cluster is in 
the state Primary/Unknown (active), and Unknown/Secondary (standby), and 
the DRBD's connection state is WFConnection.  I've attached a patch that 
addresses this issue by ensuring every device is configured before 
continuing execution.


Index: scripts/drbd
===================================================================
--- scripts/drbd        (revision 2144)
+++ scripts/drbd        (working copy)
@@ -21,7 +21,7 @@
  PROC_DRBD="/proc/drbd"
  MODPROBE="modprobe"
  RMMOD="rmmod"
-UDEV_TIMEOUT=10
+UDEV_TIMEOUT_ORIG=10
  ADD_MOD_PARAM=""

  if [ -f $DEFAULTFILE ]; then
@@ -45,9 +45,14 @@
      RESOURCE=${RESOURCE%%\ *}
      DEVICE=`$DRBDADM sh-dev $RESOURCE` || exit 20

-    while [ ! -e $DEVICE ] && [ $UDEV_TIMEOUT -gt 0 ] ; do
-        sleep 1
-       UDEV_TIMEOUT=$(( $UDEV_TIMEOUT-1 ))
+    for resource in `$DRBDADM sh-resources`; do
+       for dev in `$DRBDADM sh-dev $resource`; do
+           UDEV_TIMEOUT=$UDEV_TIMEOUT_ORIG
+           while [ ! -e $dev ] && [ $UDEV_TIMEOUT -gt 0 ] ; do
+                       sleep 1
+                       UDEV_TIMEOUT=$(( $UDEV_TIMEOUT-1 ))
+           done
+       done
      done
  }