Fix mountpoint after full fs transfer if needed

[zfs-ubuntu/.git] / zfs-replicate-all

diff --git a/zfs-replicate-all b/zfs-replicate-all
index 160bc7c314242c06e8c71f66ecf7f7075a7c9ee2..54f96eb97b042072217d00dea8573b45f4171581 100755 (executable)
--- a/zfs-replicate-all
+++ b/zfs-replicate-all
@@ -14,11 +14,12 @@ logfile="$logdir/zfs-replicate.log"
 mylogfile="$logdir/zfs-replicate-all.log"
 date=`date`
 starttime=`date +%s`
+tmpfile=`mktemp`
 
 # Make sure we aren't already running
 SCRIPT_NAME=${0##*/}
-PROCESS_LIST=`tempfile`
-ps -ef | grep -e "$SCRIPT_NAME" | grep -v grep | grep -v $$ | grep -v $PPID >> $PROCESS_LIST
+PROCESS_LIST=`mktemp`
+ps -ef | grep -e "$SCRIPT_NAME" | grep -v grep | grep -v tail | grep -v less | grep -v $$ | grep -v $PPID >> $PROCESS_LIST
 if [[ $? == 0 ]]; then
    echo "$date Another $SCRIPT_NAME process is already running" >> $mylogfile
    cat $PROCESS_LIST >> $mylogfile
@@ -54,8 +55,11 @@ fatal_and_exit() {
   # echo message to terminal
   echo -e 2>&1 "$1"
   # send email notification
-  echo -e "$1" | $mailx -s "zfs scrub on $hostname failed" "$mailto"
-  exit 1
+  echo -e "$1" | $mailx -s "zfs-replicate-all on $hostname failed" "$2"
+  # exit with bad return code unless 3rd argument was defined that says not to
+  if [[ -z "$3" ]]; then
+    exit 1
+  fi
 }
 
 # This function executes the replicate command and checks the stoptime
@@ -85,39 +89,54 @@ snapshot_age() {
   fi
 }
 
+# Replicate every zfs filesystem specified in the config file
+echo "$date ZFS replicate started" >> $logfile
+echo "$date ZFS replicate started" | tee -a $mylogfile
+
+# Loop over each backup pool
+backup_pools=$backup_pool
+for backup_pool in $backup_pools; do
+
 # Import the local backup pool if needed and the option is given to do so, else error out
-zpool list -H "$backup_pool" >/dev/null 2>&1
+zpool list -H $backup_pool > $tmpfile 2>&1
 if [ $? != 0 ]; then
   if [[ $import_export_backup_pool == 1 ]]; then
-    zpool import $backup_pool
+    zpool import -f $backup_pool >> $tmpfile 2>&1
     if [ $? != 0 ]; then
-      fatal_and_exit "-E- unable to import the backup pool $backup_pool" "$mailto"
+      msgs=`cat $tmpfile`
+      fatal_and_exit "-E- unable to import the backup pool '$backup_pool' on $hostname: $msgs" "$mailto"
     fi
   else 
-    fatal_and_exit "-E- The local backup pool, '$backup_pool' doesn't seem to exist." "$mailto"
+    msgs=`cat $tmpfile`
+    fatal_and_exit "-E- The local backup pool on $hostname, '$backup_pool' doesn't seem to exist: $msgs" "$mailto"
   fi
 fi
+rm -f $tmpfile
 
 # Obtain the zpool guid for the local backup pool
 backup_pool_guid=`zpool get guid $backup_pool 2>&1 | grep $backup_pool | awk '{ print $3 }'`
 zpool get guid $backup_pool > /dev/null 2>&1
 if [ $? != 0 ]; then
-  fatal_and_exit "-E- Unable to extract the guid for the local backup pool: $backup_pool" "$mailto"
+  fatal_and_exit "-E- Unable to extract the guid for the local backup pool on $hostname: $backup_pool" "$mailto"
 fi
 
 # Setup our backup marker names
 current_backup_marker=@current-backup-${backup_pool_guid}
 previous_backup_marker=@previous-backup-${backup_pool_guid}
 
-# Auto snapshot every zfs filesystem on the system specified below
-echo "$date ZFS replicate started" >> $logfile
-echo "$date ZFS replicate started" | tee -a $mylogfile
-
 # Sort the filesystems to replicate by the oldest backup first
-tmpfile=`tempfile`
 for filesystem in $filesystems_to_replicate; do
-  age=`snapshot_age $filesystem`
-  echo $filesystem $age >> $tmpfile
+  if [[ $filesystem =~ ':' ]]; then
+    dst_pool=${filesystem%%:*}
+    filesystem=${filesystem#*:} # remove src_pool from string
+  else
+    dst_pool=$backup_pool
+  fi
+  # Only backup filesystems that are specified to go this backup_pool
+  if [[ $backup_pool == $dst_pool ]]; then
+    age=`snapshot_age $filesystem`
+    echo $filesystem $age >> $tmpfile
+  fi
 done
 sorted_filesystems=`cat $tmpfile | sort -n -k 2 | awk '{ print $1 }'`
 rm -f $tmpfile
@@ -125,16 +144,29 @@ rm -f $tmpfile
 # Replicate the sorted filesystems
 for filesystem in $sorted_filesystems; do
   echo "-> Replicating $remote:$filesystem to ${backup_pool}/${filesystem}" | tee -a $mylogfile
-  replicate $remote $filesystem
+  replicate $remote $filesystem $backup_pool
 done
 
 # Export the local pool if told to do so
 if [[ $import_export_backup_pool == 1 ]]; then
-  zpool export $backup_pool
+  # Don't export the pool if there is a currently running zfs-scrub operation
+  ps -ef | grep "zfs-scrub" | grep -q "${backup_pool}" | grep -v grep
   if [ $? != 0 ]; then
-    fatal_and_exit "-E- unable to export the local backup pool $backup_pool" "$mailto"
+    zpool export $backup_pool >> $tmpfile 2>&1
+    if [ $? != 0 ]; then
+      sleep 300
+      zpool export $backup_pool >> $tmpfile 2>&1
+      if [ $? != 0 ]; then
+        lsof /$backup_pool/* >> $tmpfile 2>&1
+       msgs=`cat $tmpfile`
+        fatal_and_exit "-E- unable to export the local backup pool $backup_pool on $hostname: $msgs" "$mailto" 0
+      fi
+    fi
   fi
 fi
+rm -f $tmpfile
+
+done
 
 # All done
 echo `date` ZFS replicate complete >> $logfile