ctdb/config/events.d/60.ganesha

   1 #!/bin/sh
   2 # script to manage nfs in a clustered environment
   3
   4 [ -n "$CTDB_BASE" ] || \
   5     export CTDB_BASE=$(cd -P $(dirname "$0") ; dirname "$PWD")
   6
   7 . $CTDB_BASE/functions
   8
   9 GANRECDIR="/var/lib/nfs/ganesha"
  10 GANRECDIR2="/var/lib/nfs/ganesha/recevents"
  11 GPFS_STATE="/usr/lpp/mmfs/bin/mmgetstate"
  12 GANRECDIR3="/var/lib/nfs/ganesha_local"
  13
  14
  15 service_start ()
  16 {
  17     startstop_ganesha stop
  18     startstop_ganesha start
  19     set_proc "sys/net/ipv4/tcp_tw_recycle" 1
  20 }
  21
  22 service_stop ()
  23 {
  24     startstop_ganesha stop
  25 }
  26
  27 service_reconfigure ()
  28 {
  29     # if the ips have been reallocated, we must restart ganesha
  30     # across all nodes and ping all statd listeners
  31     [ -x $CTDB_BASE/statd-callout ] && {
  32         $CTDB_BASE/statd-callout notify &
  33     } >/dev/null 2>&1
  34 }
  35
  36 loadconfig "nfs"
  37
  38
  39 [ -n "$CTDB_CLUSTER_FILESYSTEM_TYPE" ] || CTDB_CLUSTER_FILESYSTEM_TYPE="gpfs"
  40
  41 service_name="nfs-ganesha-$CTDB_CLUSTER_FILESYSTEM_TYPE"
  42
  43 [ "${CTDB_NFS_SERVER_MODE:-${NFS_SERVER_MODE}}" = "ganesha" ] || exit 0
  44
  45 ctdb_setup_service_state_dir
  46
  47 ctdb_start_stop_service
  48
  49 is_ctdb_managed_service || exit 0
  50
  51 ctdb_service_check_reconfigure
  52
  53 nodenum_file="${service_state_dir}/gpfs_nodenum"
  54
  55 get_cluster_fs_state  ()
  56 {
  57     case $CTDB_CLUSTER_FILESYSTEM_TYPE in
  58         gpfs)
  59             STATE=`$GPFS_STATE | awk 'NR <= 3 {next} {printf "%-6s", $3}'`
  60             echo $STATE
  61             ;;
  62         *)
  63             die "File system $CTDB_CLUSTER_FILESYSTEM_TYPE not supported"
  64             ;;
  65    esac
  66 }
  67
  68 create_nodenum_file()
  69 {
  70     NNUM=$(/usr/lpp/mmfs/bin/mmlsconfig myNodeConfigNumber | awk '{print $2}')
  71     echo $NNUM > $nodenum_file
  72 }
  73
  74 get_nodenum()
  75 {
  76     if [ ! -f $nodenum_file ]; then
  77         create_nodenum_file
  78     fi
  79     cat $nodenum_file
  80 }
  81
  82
  83 create_ganesha_recdirs ()
  84 {
  85     [ -n "$CTDB_GANESHA_REC_SUBDIR" ] || CTDB_GANESHA_REC_SUBDIR=".ganesha"
  86
  87     _mounts=$(mount -t $CTDB_CLUSTER_FILESYSTEM_TYPE)
  88     if [ -z "$_mounts" ]; then
  89       echo "startup $CTDB_CLUSTER_FILESYSTEM_TYPE not ready"
  90       exit 0
  91     fi
  92     _mntpt=$(echo "$_mounts" | sort | awk 'NR == 1 {print $3}')
  93     _link_dst="${_mntpt}/${CTDB_GANESHA_REC_SUBDIR}"
  94     mkdir -vp "$_link_dst"
  95     if [ -e "$GANRECDIR" ]; then
  96         if [ ! -L "$GANRECDIR" ] ; then
  97             rm -vrf "$GANRECDIR"
  98         else
  99             _t=$(readlink "$GANRECDIR")
 100             if [ "$_t" != "$_link_dst" ] ; then
 101                 rm -v "$GANRECDIR"
 102             fi
 103         fi
 104     fi
 105     # This is not an "else".  It also re-creates the link if it was
 106     # removed above!
 107     if [ ! -e "$GANRECDIR" ]; then
 108         ln -sv "$_link_dst" "$GANRECDIR"
 109     fi
 110
 111     mkdir -p $GANRECDIR2
 112     mkdir -p $GANRECDIR3
 113 }
 114
 115 monitor_ganesha_nfsd ()
 116 {
 117         create_ganesha_recdirs
 118         service_name=${service_name}_process
 119
 120         PIDFILE="/var/run/ganesha.pid"
 121         CUR_STATE=`get_cluster_fs_state`
 122         GANESHA="/usr/bin/$CTDB_CLUSTER_FILESYSTEM_TYPE.ganesha.nfsd"
 123         if { read PID < $PIDFILE && \
 124             grep "$GANESHA" "/proc/$PID/cmdline" ; } >/dev/null 2>&1 ; then
 125                 ctdb_counter_init "$service_name"
 126         else
 127             if [ $CUR_STATE = "active" ]; then
 128                 echo "Trying fast restart of NFS service"
 129                 startstop_ganesha restart
 130                 ctdb_counter_incr "$service_name"
 131                 ctdb_check_counter "error" "-ge" "6" "$service_name"
 132             fi
 133         fi
 134
 135         service_name="nfs-ganesha-$CTDB_CLUSTER_FILESYSTEM_TYPE"_service
 136         # check that NFS is posting forward progress
 137         if [ $CUR_STATE = "active" -a "$CTDB_NFS_SKIP_KNFSD_ALIVE_CHECK" != "yes" ] ; then
 138             MAXREDS=2
 139             MAXSTALL=120
 140             RESTART=0
 141
 142             NUMREDS=`ls $GANRECDIR3 | grep "red" | wc -l`
 143             LASTONE=`ls -t $GANRECDIR3 | sed 's/_/ /' | awk 'NR > 1 {next} {printf $1} '`
 144             # Beware of startup
 145             if [ -z $LASTONE ] ; then
 146                 LASTONE=`date +"%s"`
 147             fi
 148             TNOW=$(date +"%s")
 149             TSTALL=$(($TNOW - $LASTONE))
 150             if [ $NUMREDS -ge $MAXREDS ] ; then
 151                 echo restarting because of $NUMREDS red conditions
 152                 RESTART=1
 153                 ctdb_counter_incr "$service_name"
 154                 ctdb_check_counter "error" "-ge" "6" "$service_name"
 155             fi
 156             if [ $TSTALL -ge $MAXSTALL ] ; then
 157                 echo restarting because of $TSTALL second stall
 158                 RESTART=1
 159                 ctdb_counter_incr "$service_name"
 160                 ctdb_check_counter "error" "-ge" "6" "$service_name"
 161             fi
 162             if [ $RESTART -gt 0 ] ; then
 163                 startstop_ganesha restart
 164             else
 165                 ctdb_counter_init "$service_name"
 166             fi
 167         fi
 168 }
 169
 170 ############################################################
 171
 172 case "$1" in
 173      init)
 174         # read statd from persistent database
 175         ;;
 176      startup)
 177         create_ganesha_recdirs
 178         ctdb_service_start
 179         create_nodenum_file
 180         ;;
 181
 182      shutdown)
 183         ctdb_service_stop
 184         ;;
 185
 186      takeip)
 187         if [ -n "$2" ] ; then
 188             case  $CTDB_CLUSTER_FILESYSTEM_TYPE in
 189                 gpfs)
 190                     NNUM=$(get_nodenum)
 191                     TDATE=`date +"%s"`
 192                     TOUCHTGT=$1"_"$TDATE"_"$NNUM"_"$3"_"$4"_"$2
 193                     touch $GANRECDIR2/$TOUCHTGT
 194                     ;;
 195             esac
 196         fi
 197         ctdb_service_set_reconfigure
 198         ;;
 199
 200      releaseip)
 201         if [ -n "$2" ] ; then
 202             case  $CTDB_CLUSTER_FILESYSTEM_TYPE in
 203                 gpfs)
 204                     NNUM=$(get_nodenum)
 205                     TDATE=`date +"%s"`
 206                     TOUCHTGT=$1"_"$TDATE"_"$NNUM"_"$3"_"$4"_"$2
 207                     touch $GANRECDIR2/$TOUCHTGT
 208                     TOUCHTGT="my"$TOUCHTGT
 209                     touch $GANRECDIR2/$TOUCHTGT
 210                 ;;
 211             esac
 212         fi
 213         ctdb_service_set_reconfigure
 214         ;;
 215
 216      monitor)
 217         # Check that directories for shares actually exist.
 218         [ "$CTDB_NFS_SKIP_SHARE_CHECK" = "yes" ] || {
 219             grep Path /etc/ganesha/$CTDB_CLUSTER_FILESYSTEM_TYPE.ganesha.exports.conf |
 220             cut -f2 -d\" | sort -u | ctdb_check_directories
 221         } || exit $?
 222
 223         update_tickles 2049
 224
 225         # check that statd responds to rpc requests
 226         # if statd is not running we try to restart it
 227         # we only do this IF we have a rpc.statd command.
 228         # For platforms where rpc.statd does not exist, we skip
 229         # the check completely
 230         p="rpc.statd"
 231         which $p >/dev/null 2>/dev/null && \
 232             nfs_check_rpc_service "statd" \
 233                 %  10 "verbose restart:b unhealthy" \
 234                 -ge 6 "verbose unhealthy" \
 235                 -eq 4 "verbose restart" \
 236                 -eq 2 "restart:b"
 237
 238         if [ "$CTDB_SKIP_GANESHA_NFSD_CHECK" != "yes" ] ; then
 239             monitor_ganesha_nfsd
 240         fi
 241
 242         # rquotad is sometimes not started correctly on RHEL5
 243         # not a critical service so we dont flag the node as unhealthy
 244         nfs_check_rpc_service "rquotad" \
 245             -gt 0 "verbose restart:b"
 246         ;;
 247
 248      *)
 249         ctdb_standard_event_handler "$@"
 250         ;;
 251 esac
 252
 253 exit 0