ceph/src/init-ceph.in

   1 #!/bin/sh
   2 # Start/stop ceph daemons
   3 # chkconfig: 2345 60 80
   4
   5 ### BEGIN INIT INFO
   6 # Provides:          ceph
   7 # Default-Start:     2 3 4 5
   8 # Default-Stop:      0 1 6
   9 # Required-Start:    $remote_fs $named $network $time
  10 # Required-Stop:     $remote_fs $named $network $time
  11 # Short-Description: Start Ceph distributed file system daemons at boot time
  12 # Description:       Enable Ceph distributed file system services.
  13 ### END INIT INFO
  14
  15 # TODO: on FreeBSD/OSX, use equivalent script file
  16 if [ -e /lib/lsb/init-functions ]; then
  17     . /lib/lsb/init-functions
  18 fi
  19
  20 if [ `dirname $0` = "." ] && [ $PWD != "/etc/init.d" ]; then
  21     # looks like an autotools src dir build
  22     BINDIR=.
  23     SBINDIR=.
  24     LIBEXECDIR=.
  25     ETCDIR=.
  26     ASSUME_DEV=1
  27 else
  28     if [ -e CMakeCache.txt ] && [ -e bin/init-ceph ]; then
  29         # looks like a cmake build directory
  30         CEPH_ROOT=`grep ceph_SOURCE_DIR CMakeCache.txt | cut -d "=" -f 2`
  31         BINDIR=bin
  32         SBINDIR=bin
  33         LIBEXECDIR=$CEPH_ROOT/src
  34         ETCDIR=.
  35         ASSUME_DEV=1
  36         CEPH_LIB=$CEPH_ROOT/build/lib
  37         echo "$PYTHONPATH" | grep -q $CEPH_LIB || export PYTHONPATH=$CEPH_LIB/cython_modules/lib.2:$PYTHONPATH
  38         echo "$LD_LIBRARY_PATH" | grep -q $CEPH_LIB || export LD_LIBRARY_PATH=$CEPH_LIB:$LD_LIBRARY_PATH
  39         echo "$DYLD_LIBRARY_PATH" | grep -q $CEPH_LIB || export DYLD_LIBRARY_PATH=$CEPH_LIB:$DYLD_LIBRARY_PATH
  40     else
  41         BINDIR=@bindir@
  42         SBINDIR=@sbindir@
  43         LIBEXECDIR=@libexecdir@/ceph
  44         ETCDIR=@sysconfdir@/ceph
  45         ASSUME_DEV=0
  46     fi
  47 fi
  48
  49 if [ -n "$CEPH_BIN" ] && [ -n "$CEPH_ROOT" ] && [ -n "$CEPH_BUILD_DIR" ]; then
  50   BINDIR=$CEPH_BIN
  51   SBINDIR=$CEPH_ROOT/src
  52   ETCDIR=$CEPH_BIN
  53   LIBEXECDIR=$CEPH_ROOT/src
  54   ASSUME_DEV=1
  55 fi
  56
  57 if [ `uname` = FreeBSD ]; then
  58   GETOPT=/usr/local/bin/getopt
  59 else
  60   GETOPT=getopt
  61 fi
  62
  63 if id ceph > /dev/null 2>&1; then
  64   SET_CEPHUSER_ARGS=" --setuser ceph --setgroup ceph"
  65 fi
  66
  67 usage_exit() {
  68     echo "usage: $0 [options] {start|stop|restart|condrestart} [mon|osd|mds]..."
  69     printf "Core options:\n"
  70     printf "\t--allhosts / -a           execute (via ssh) on all hosts in conf file\n"
  71     printf "\t--cluster [cluster name]  define the cluster name\n"
  72     printf "\t--conf / -c [conf file]   use [conf file] instead of default\n"
  73     printf "\t--help / -h               show this usage message\n"
  74     printf "\t--hostname [hostname]     override hostname lookup\n"
  75     printf "\t-m [mon addr]             mon address\n"
  76     printf "\n"
  77     printf "Other options:\n"
  78     printf "\t--btrfs                   btrfs\n"
  79     printf "\t--nobtrfs                 no btrfs\n"
  80     printf "\t--btrfsumount             btrfs umount\n"
  81     printf "\t--fsmount                 fsmount\n"
  82     printf "\t--nofsmount               no fsmount\n"
  83     printf "\t--fsumount                fsumount\n"
  84     printf "\t--restart                 restart on core dump\n"
  85     printf "\t--norestart               do not restart on core dump\n"
  86     printf "\t--valgrind                run via valgrind\n"
  87     printf "\t--novalgrind              do not run via valgrind\n"
  88     printf "\t--verbose / -v            be verbose\n"
  89     exit
  90 }
  91
  92 # behave if we are not completely installed (e.g., Debian "removed,
  93 # config remains" state)
  94 test -f $LIBEXECDIR/ceph_common.sh || exit 0
  95
  96 . $LIBEXECDIR/ceph_common.sh
  97
  98 EXIT_STATUS=0
  99
 100 signal_daemon() {
 101     name=$1
 102     daemon=$2
 103     pidfile=$3
 104     signal=$4
 105     action=$5
 106     [ -z "$action" ] && action="Stopping"
 107     printf "$action Ceph $name on $host..."
 108     do_cmd "if [ -e $pidfile ]; then
 109         pid=\`cat $pidfile\`
 110         if ps -p \$pid -o args= | grep -q $daemon; then
 111             cmd=\"kill $signal \$pid\"
 112             printf \"\$cmd...\"
 113             \$cmd
 114         fi
 115     fi"
 116     echo done
 117 }
 118
 119 daemon_is_running() {
 120     name=$1
 121     daemon=$2
 122     daemon_id=$3
 123     pidfile=$4
 124     do_cmd "[ -e $pidfile ] || exit 1   # no pid, presumably not running
 125         pid=\`cat $pidfile\`
 126         ps -p \$pid -o args= | grep $daemon | grep -qwe -i.$daemon_id && exit 0 # running
 127         exit 1  # pid is something else" "" "okfail"
 128 }
 129
 130 stop_daemon() {
 131     name=$1
 132     daemon=$2
 133     pidfile=$3
 134     signal=$4
 135     action=$5
 136     [ -z "$action" ] && action="Stopping"
 137     printf "$action Ceph $name on $host..."
 138     do_cmd "if [ -e $pidfile ] ; then
 139         pid=\`cat $pidfile\`
 140         while ps -p \$pid -o args= | grep -q $daemon; do
 141             cmd=\"kill $signal \$pid\"
 142             printf \"\$cmd...\"
 143             \$cmd
 144             sleep 1
 145             continue
 146         done
 147     fi"
 148     echo done
 149 }
 150
 151 ## command line options
 152 options=
 153
 154 OPTS=$(${GETOPT} -n 'init-ceph' -o 'hvam:c:' -l 'help,verbose,valgrind,novalgrind,allhosts,restart,norestart,btrfs,nobtrfs,fsmount,nofsmount,btrfsumount,fsumount,conf:,cluster:,hostname:' -- "$@")
 155 if [ $? != 0 ]
 156 then
 157     exit 1
 158 fi
 159
 160 eval set -- "$OPTS"
 161
 162 dovalgrind=
 163 docrun=
 164 allhosts=0
 165 monaddr=
 166 dofsmount=1
 167 dofsumount=0
 168 verbose=0
 169 use_default_conf=1
 170
 171 ## set variables like cluster or conf
 172 [ -e /etc/sysconfig/ceph ] && . /etc/sysconfig/ceph
 173 [ -e /etc/default/ceph ] && . /etc/default/ceph
 174
 175
 176 while echo $1 | grep -q '^-'; do     # FIXME: why not '^-'?
 177 case $1 in
 178     -v | --verbose)
 179             verbose=1
 180             ;;
 181     --valgrind)
 182             dovalgrind=1
 183             ;;
 184     --novalgrind)
 185             dovalgrind=0
 186             ;;
 187     --allhosts | -a)
 188             allhosts=1;
 189             ;;
 190     --restart)
 191             docrun=1
 192             ;;
 193     --norestart)
 194             docrun=0
 195             ;;
 196     -h | --help)
 197             usage_exit
 198             ;;
 199     -m )
 200             [ -z "$2" ] && usage_exit
 201             options="$options $1"
 202             shift
 203             MON_ADDR=$1
 204             ;;
 205     --btrfs | --fsmount)
 206             dofsmount=1
 207             ;;
 208     --nobtrfs | --nofsmount)
 209             dofsmount=0
 210             ;;
 211     --btrfsumount | --fsumount)
 212             dofsumount=1
 213             ;;
 214     --conf | -c)
 215             [ -z "$2" ] && usage_exit
 216             options="$options $1"
 217             shift
 218         use_default_conf=0
 219             conf=$1
 220             ;;
 221     --cluster )
 222             [ -z "$2" ] && usage_exit
 223             options="$options $1"
 224             shift
 225             cluster=$1
 226             ;;
 227     --hostname )
 228             [ -z "$2" ] && usage_exit
 229             options="$options $1"
 230             shift
 231             hostname=$1
 232             ;;
 233     -- )
 234             shift
 235             break
 236             ;;
 237     *)
 238             echo unrecognized option \'$1\'
 239             usage_exit
 240             ;;
 241 esac
 242 options="$options $1"
 243 shift
 244 done
 245
 246 # if `--cluster` was not passed in, fallback to looking at the config name
 247 if [ -z "$cluster" ]; then
 248     cluster=`echo $conf | awk -F'/' '{print $(NF)}' | cut -d'.' -f 1`
 249 else
 250     # if we were told to use a given cluster name then $conf needs to be updated
 251     # but just define it if `--conf` was not specified, otherwise we would be silently
 252     # overriding $conf even if it was defined with `--conf`
 253     if [ $use_default_conf -eq 1 ]; then
 254         conf="/etc/ceph/$cluster.conf"
 255     fi
 256 fi
 257
 258
 259 verify_conf
 260
 261 command=$1
 262 [ -n "$*" ] && shift
 263
 264 get_local_name_list
 265 get_name_list "$@"
 266
 267 # Reverse the order if we are stopping
 268
 269 if [ "$command" = "stop" -o "$command" = "onestop" ]; then
 270     for f in $what; do
 271        new_order="$f $new_order"
 272     done
 273     what="$new_order"
 274 fi
 275
 276 for name in $what; do
 277     type=`echo $name | cut -c 1-3`   # e.g. 'mon', if $item is 'mon1'
 278     id=`echo $name | cut -c 4- | sed 's/^\\.//'`
 279     num=$id
 280     name="$type.$id"
 281
 282     check_host $cluster || continue
 283
 284     binary="$BINDIR/ceph-$type"
 285     cmd="$binary -i $id"
 286     if [ $ASSUME_DEV -eq 1 ]; then
 287       cmd="PATH=$PWD:$PATH $cmd"
 288     fi
 289
 290     get_conf run_dir "/var/run/ceph" "run dir"
 291
 292     get_conf pid_file "$run_dir/$type.$id.pid" "pid file"
 293
 294     if [ "$command" = "start" -o "$command" = "onestart" ]; then
 295         if [ -n "$pid_file" ]; then
 296             do_cmd "mkdir -p "`dirname $pid_file`
 297             cmd="$cmd --pid-file $pid_file"
 298         fi
 299
 300         get_conf log_dir "" "log dir"
 301         [ -n "$log_dir" ] && do_cmd "mkdir -p $log_dir"
 302
 303         get_conf auto_start "" "auto start"
 304         if [ "$auto_start" = "no" ] || [ "$auto_start" = "false" ] || [ "$auto_start" = "0" ]; then
 305             if [ -z "$@" ]; then
 306                 echo "Skipping Ceph $name on $host... auto start is disabled"
 307                 continue
 308             fi
 309         fi
 310
 311         if daemon_is_running $name ceph-$type $id $pid_file; then
 312             echo "Starting Ceph $name on $host...already running"
 313             continue
 314         fi
 315
 316         get_conf copy_executable_to "" "copy executable to"
 317         if [ -n "$copy_executable_to" ]; then
 318             scp $binary "$host:$copy_executable_to"
 319             binary="$copy_executable_to"
 320         fi
 321     fi
 322
 323     # conf file
 324     cmd="$cmd -c $conf"
 325
 326     if echo $name | grep -q ^osd; then
 327         get_conf osd_data "/var/lib/ceph/osd/$cluster-$id" "osd data"
 328         get_conf fs_path "$osd_data" "fs path"  # mount point defaults so osd data
 329         get_conf fs_devs "" "devs"
 330         if [ -z "$fs_devs" ]; then
 331             # try to fallback to old keys
 332             get_conf tmp_btrfs_devs "" "btrfs devs"
 333             if [ -n "$tmp_btrfs_devs" ]; then
 334                 fs_devs="$tmp_btrfs_devs"
 335             fi
 336         fi
 337         first_dev=`echo $fs_devs | cut '-d ' -f 1`
 338     fi
 339
 340     # do lockfile, if RH
 341     get_conf lockfile "/var/lock/subsys/ceph" "lock file"
 342     lockdir=`dirname $lockfile`
 343     if [ ! -d "$lockdir" ]; then
 344         lockfile=""
 345     fi
 346
 347     get_conf asok "$run_dir/$cluster-$type.$id.asok" "admin socket"
 348
 349     case "$command" in
 350         start|onestart)
 351             # Increase max_open_files, if the configuration calls for it.
 352             get_conf max_open_files "32768" "max open files"
 353
 354             # build final command
 355             wrap=""
 356             runmode=""
 357             runarg=""
 358
 359             [ -z "$docrun" ] && get_conf_bool docrun "0" "restart on core dump"
 360             [ "$docrun" -eq 1 ] && wrap="$BINDIR/ceph-run"
 361
 362             [ -z "$dovalgrind" ] && get_conf_bool valgrind "" "valgrind"
 363             [ -n "$valgrind" ] && wrap="$wrap valgrind $valgrind"
 364
 365             [ -n "$wrap" ] && runmode="-f &" && runarg="-f"
 366             [ -n "$max_open_files" ] && files="ulimit -n $max_open_files;"
 367
 368             [ -n "$TCMALLOC_MAX_TOTAL_THREAD_CACHE_BYTES" ] && tcmalloc="TCMALLOC_MAX_TOTAL_THREAD_CACHE_BYTES=$TCMALLOC_MAX_TOTAL_THREAD_CACHE_BYTES"
 369
 370             cmd="$files $tcmalloc $wrap $cmd --cluster $cluster ${SET_CEPHUSER_ARGS} $runmode"
 371
 372             if [ $dofsmount -eq 1 ] && [ -n "$fs_devs" ]; then
 373                 get_conf pre_mount "true" "pre mount command"
 374                 get_conf fs_type "" "osd mkfs type"
 375
 376                 if [ -z "$fs_type" ]; then
 377                     # try to fallback to old keys
 378                     get_conf tmp_devs "" "btrfs devs"
 379                     if [ -n "$tmp_devs" ]; then
 380                         fs_type="btrfs"
 381                     else
 382                         echo No filesystem type defined!
 383                         exit 0
 384                     fi
 385                 fi
 386
 387                 get_conf fs_opt "" "osd mount options $fs_type"
 388                 if [ -z "$fs_opt" ]; then
 389                     if [ "$fs_type" = "btrfs" ]; then
 390                         #try to fallback to old keys
 391                         get_conf fs_opt "" "btrfs options"
 392                     fi
 393
 394                     if [ -z "$fs_opt" ]; then
 395                         if [ "$fs_type" = "xfs" ]; then
 396                             fs_opt="rw,noatime,inode64"
 397                         else
 398                             #fallback to use at least noatime
 399                             fs_opt="rw,noatime"
 400                         fi
 401                     fi
 402                 fi
 403
 404                 [ -n "$fs_opt" ] && fs_opt="-o $fs_opt"
 405                 [ -n "$pre_mount" ] && do_cmd "$pre_mount"
 406
 407                 do_root_cmd_okfail "mkdir -p $fs_path"
 408                 if [ "$fs_type" = "btrfs" ]; then
 409                     echo Mounting Btrfs on $host:$fs_path
 410                     do_root_cmd_okfail "modprobe btrfs ; btrfs device scan || btrfsctl -a ; egrep -q '^[^ ]+ $fs_path ' /proc/mounts && umount $fs_path ; mount -t btrfs $fs_opt $first_dev $fs_path"
 411                 else
 412                     echo Mounting $fs_type on $host:$fs_path
 413                     do_root_cmd_okfail "modprobe $fs_type ; egrep -q '^[^ ]+ $fs_path ' /proc/mounts && umount $fs_path ; mount -t $fs_type $fs_opt $first_dev $fs_path"
 414                 fi
 415                 if [ "$ERR" != "0" ]; then
 416                     EXIT_STATUS=$ERR
 417                     continue
 418                 fi
 419             fi
 420
 421             if [ "$type" = "osd" ]; then
 422                 get_conf update_crush "" "osd crush update on start"
 423                 case "${update_crush:-1}" in 1|[Tt][Rr][Uu][Ee])
 424                     # update location in crush
 425                     get_conf osd_location_hook "$BINDIR/ceph-crush-location" "osd crush location hook"
 426                    if [ ! -e "$osd_location_hook" ]; then
 427                        EXIT_STATUS=2
 428                        continue
 429                    fi
 430                     osd_location=`$osd_location_hook --cluster $cluster --id $id --type osd`
 431                     get_conf osd_weight "" "osd crush initial weight"
 432                     defaultweight="$(df -P -k $osd_data/. | tail -1 | awk '{ print sprintf("%.4f",$2/1073741824) }')"
 433                     get_conf osd_keyring "$osd_data/keyring" "keyring"
 434                     do_cmd_okfail "timeout 30 $BINDIR/ceph -c $conf --name=osd.$id --keyring=$osd_keyring osd crush create-or-move -- $id ${osd_weight:-${defaultweight:-1}} $osd_location"
 435                     if [ "$ERR" != "0" ]; then
 436                         EXIT_STATUS=$ERR
 437                         continue
 438                     fi
 439                 esac
 440             fi
 441
 442             echo Starting Ceph $name on $host...
 443             if [ ! -d $run_dir ]; then
 444                 # assume /var/run exists
 445                 install -d -m0770 -o ceph -g ceph /var/run/ceph
 446             fi
 447             get_conf pre_start_eval "" "pre start eval"
 448             [ -n "$pre_start_eval" ] && $pre_start_eval
 449             get_conf pre_start "" "pre start command"
 450             get_conf post_start "" "post start command"
 451             [ -n "$pre_start" ] && do_cmd "$pre_start"
 452             do_cmd_okfail "$cmd" $runarg
 453             if [ "$ERR" != "0" ]; then
 454                 EXIT_STATUS=$ERR
 455                 continue
 456             fi
 457
 458             [ -n "$post_start" ] && do_cmd "$post_start"
 459             [ -n "$lockfile" ] && [ "$?" -eq 0 ] && touch $lockfile
 460             ;;
 461
 462         stop|onestop)
 463             get_conf pre_stop "" "pre stop command"
 464             get_conf post_stop "" "post stop command"
 465             [ -n "$pre_stop" ] && do_cmd "$pre_stop"
 466             stop_daemon $name ceph-$type $pid_file
 467             [ -n "$pidfile" ] && rm -f $pidfile
 468             [ -n "$asok" ] && rm -f $asok
 469             [ -n "$post_stop" ] && do_cmd "$post_stop"
 470             [ -n "$lockfile" ] && [ "$?" -eq 0 ] && rm -f $lockfile
 471             if [ $dofsumount -eq 1 ] && [ -n "$fs_devs" ]; then
 472                 echo Unmounting OSD volume on $host:$fs_path
 473                 do_root_cmd "umount $fs_path || true"
 474             fi
 475             ;;
 476
 477         status)
 478             if daemon_is_running $name ceph-$type $id $pid_file; then
 479                 printf "$name: running "
 480                 do_cmd "$BINDIR/ceph daemon $name version 2>/dev/null" || printf unknown
 481                 printf "\n"
 482             elif [ -e "$pid_file" ]; then
 483                 # daemon is dead, but pid file still exists
 484                 echo "$name: dead."
 485                 EXIT_STATUS=1
 486             else
 487                 # daemon is dead, and pid file is gone
 488                 echo "$name: not running."
 489                 EXIT_STATUS=3
 490             fi
 491             ;;
 492
 493         ssh)
 494             $ssh
 495             ;;
 496
 497         forcestop)
 498             get_conf pre_forcestop "" "pre forcestop command"
 499             get_conf post_forcestop "" "post forcestop command"
 500             [ -n "$pre_forcestop" ] && do_cmd "$pre_forcestop"
 501             stop_daemon $name ceph-$type $pid_file -9
 502             [ -n "$post_forcestop" ] && do_cmd "$post_forcestop"
 503             [ -n "$lockfile" ] && [ "$?" -eq 0 ] && rm -f $lockfile
 504             ;;
 505
 506         killall)
 507             echo "killall ceph-$type on $host"
 508             do_cmd "pkill ^ceph-$type || true"
 509             [ -n "$lockfile" ] && [ "$?" -eq 0 ] && rm -f $lockfile
 510             ;;
 511
 512         force-reload | reload)
 513             signal_daemon $name ceph-$type $pid_file -1 "Reloading"
 514             ;;
 515
 516         restart|onerestart)
 517             $0 $options stop $name
 518             $0 $options start $name
 519             ;;
 520
 521         condrestart)
 522             if daemon_is_running $name ceph-$type $id $pid_file; then
 523                 $0 $options stop $name
 524                 $0 $options start $name
 525             else
 526                 echo "$name: not running."
 527             fi
 528             ;;
 529
 530         cleanlogs)
 531             echo removing logs
 532             [ -n "$log_dir" ] && do_cmd "rm -f $log_dir/$type.$id.*"
 533             ;;
 534
 535         cleanalllogs)
 536             echo removing all logs
 537             [ -n "$log_dir" ] && do_cmd "rm -f $log_dir/* || true"
 538             ;;
 539
 540         *)
 541             usage_exit
 542             ;;
 543     esac
 544 done
 545
 546 # activate latent osds?
 547 if [ "$command" = "start" -a "$BINDIR" != "." ]; then
 548     if [ "$*" = "" ] || echo $* | grep -q ^osd\$ ; then
 549        if [ -x $SBINDIR/ceph-disk ]; then
 550            ceph-disk activate-all
 551        fi
 552     fi
 553 fi
 554
 555 exit $EXIT_STATUS