| 1 | #!/bin/ksh |
|---|
| 2 | # IBM_PROLOG_BEGIN_TAG |
|---|
| 3 | # This is an automatically generated prolog. |
|---|
| 4 | # |
|---|
| 5 | # |
|---|
| 6 | # |
|---|
| 7 | # Licensed Materials - Property of IBM |
|---|
| 8 | # |
|---|
| 9 | # (C) COPYRIGHT International Business Machines Corp. 1999,2006 |
|---|
| 10 | # All Rights Reserved |
|---|
| 11 | # |
|---|
| 12 | # US Government Users Restricted Rights - Use, duplication or |
|---|
| 13 | # disclosure restricted by GSA ADP Schedule Contract with IBM Corp. |
|---|
| 14 | # |
|---|
| 15 | # IBM_PROLOG_END_TAG |
|---|
| 16 | # @(#)13 1.3.3.20 src/avs/fs/mmfs/ts/admin/mmshutdown.sh, mmfs, avs_rgpfs24, rgpfs240610b 1/23/06 02:25:22 |
|---|
| 17 | ############################################################################## |
|---|
| 18 | # |
|---|
| 19 | # GPFS filesystems on the specified nodes will be unmounted and then |
|---|
| 20 | # the GPFS daemons on the specified nodes will be shut down. |
|---|
| 21 | # |
|---|
| 22 | # Usage: |
|---|
| 23 | # |
|---|
| 24 | # mmshutdown [ -t unmountTimeout ] |
|---|
| 25 | # [-a | -N {Node[,Node...] | NodeFile | NodeClass}] |
|---|
| 26 | # |
|---|
| 27 | # where: |
|---|
| 28 | # |
|---|
| 29 | # -t timeout Will timeout the unmount command after this amount of time |
|---|
| 30 | # and shut the GPFS daemons down anyway. |
|---|
| 31 | # |
|---|
| 32 | # -a Stop the daemon on all nodes in the GPFS cluster. |
|---|
| 33 | # |
|---|
| 34 | # -N Node,Node,... Specify the nodes to be shut down. |
|---|
| 35 | # -N NodeFile NodeClass may be one of several possible node classes |
|---|
| 36 | # -N NodeClass (e.g., quorumnodes, managernodes, nsdnodes, etc.) |
|---|
| 37 | # |
|---|
| 38 | # |
|---|
| 39 | # If not explicitly specified otherwise, the daemon is shut down on the local |
|---|
| 40 | # node only. |
|---|
| 41 | # |
|---|
| 42 | # |
|---|
| 43 | # Obsolete options: |
|---|
| 44 | # |
|---|
| 45 | # -C NodesetId Start the daemon on all nodes in the specified nodeset. |
|---|
| 46 | # Assumed to be the same as -a. |
|---|
| 47 | # |
|---|
| 48 | # -W NodeFile Start the daemon on all nodes whose reliable hostnames |
|---|
| 49 | # are listed one per line in NodeFilename. |
|---|
| 50 | # Cannot be specified with -a, -N, -w, or -n. |
|---|
| 51 | # |
|---|
| 52 | # -w nodenames Start the daemon on all nodes whose reliable hostnames |
|---|
| 53 | # are in the comma-separated nodenames list. |
|---|
| 54 | # Cannot be specified with -a, -N or -W. |
|---|
| 55 | # If both -w and -n are specified, the lists are combined. |
|---|
| 56 | # |
|---|
| 57 | # -n nodenums Start the daemon on all nodes whose node numbers are |
|---|
| 58 | # in the comma-separated nodenums list. |
|---|
| 59 | # Cannot be specified with -a, -N, or -W. |
|---|
| 60 | # If both -w and -n are specified, the lists are combined. |
|---|
| 61 | # |
|---|
| 62 | ############################################################################## |
|---|
| 63 | |
|---|
| 64 | # Include global declarations and service routines. |
|---|
| 65 | . /usr/lpp/mmfs/bin/mmglobfuncs |
|---|
| 66 | . /usr/lpp/mmfs/bin/mmsdrfsdef |
|---|
| 67 | |
|---|
| 68 | sourceFile="mmshutdown.sh" |
|---|
| 69 | [[ -n $DEBUG || -n $DEBUGmmshutdown ]] && set -x |
|---|
| 70 | $mmTRACE_ENTER "$*" |
|---|
| 71 | |
|---|
| 72 | |
|---|
| 73 | # Local variables |
|---|
| 74 | usageMsg=246 |
|---|
| 75 | typeset -i maxTimeout=900 |
|---|
| 76 | typeset -i timeout=0 |
|---|
| 77 | typeset -i nodes=0 |
|---|
| 78 | rc=0 |
|---|
| 79 | |
|---|
| 80 | |
|---|
| 81 | ####################### |
|---|
| 82 | # Mainline processing |
|---|
| 83 | ####################### |
|---|
| 84 | |
|---|
| 85 | |
|---|
| 86 | ################################## |
|---|
| 87 | # Process each of the arguments. |
|---|
| 88 | ################################## |
|---|
| 89 | [[ $arg1 = '-?' || $arg1 = '-h' || $arg1 = '--help' || $arg1 = '--' ]] && \ |
|---|
| 90 | syntaxError "help" $usageMsg |
|---|
| 91 | |
|---|
| 92 | while getopts :aC:n:N:t:W:w: OPT |
|---|
| 93 | do |
|---|
| 94 | case $OPT in |
|---|
| 95 | |
|---|
| 96 | a) [[ -n $aflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 97 | aflag="-$OPT" |
|---|
| 98 | [[ -n $Cflag || -n $Wflag || -n $wflag || -n $nflag || -n $Nflag ]] && \ |
|---|
| 99 | syntaxError "invalidCombination" \ |
|---|
| 100 | $usageMsg $aflag $Cflag $Wflag $wflag $nflag $Nflag |
|---|
| 101 | callGpfsInit=yes |
|---|
| 102 | ;; |
|---|
| 103 | |
|---|
| 104 | C) # syntaxError "obsoleteOption" $usageMsg "-$OPT" |
|---|
| 105 | [[ -n $Cflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 106 | Cflag="-$OPT" |
|---|
| 107 | nodesetId="$OPTARG" |
|---|
| 108 | [[ -n $aflag || -n $Wflag || -n $wflag || -n $nflag || -n $Nflag ]] && \ |
|---|
| 109 | syntaxError "invalidCombination" \ |
|---|
| 110 | $usageMsg $Cflag $aflag $Wflag $wflag $nflag $Nflag |
|---|
| 111 | callGpfsInit=yes |
|---|
| 112 | ;; |
|---|
| 113 | |
|---|
| 114 | n) [[ -n $nflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 115 | nflag="-$OPT" |
|---|
| 116 | nodenums="$OPTARG" |
|---|
| 117 | [[ -n $Cflag || -n $Wflag || -n $aflag || -n $Nflag ]] && \ |
|---|
| 118 | syntaxError "invalidCombination" \ |
|---|
| 119 | $usageMsg $nflag $Cflag $Wflag $aflag $Nflag |
|---|
| 120 | callGpfsInit=yes |
|---|
| 121 | ;; |
|---|
| 122 | |
|---|
| 123 | N) [[ -n $Nflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 124 | Nflag="-$OPT" |
|---|
| 125 | nodenames="$OPTARG" |
|---|
| 126 | [[ -n $Cflag || -n $Wflag || -n $aflag || -n $nflag || -n $wflag ]] && \ |
|---|
| 127 | syntaxError "invalidCombination" \ |
|---|
| 128 | $usageMsg $Nflag $Cflag $Wflag $aflag $nflag $wflag |
|---|
| 129 | callGpfsInit=yes |
|---|
| 130 | ;; |
|---|
| 131 | |
|---|
| 132 | t) [[ -n $tflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 133 | tflag="-$OPT" |
|---|
| 134 | timeout=$(checkIntRange $tflag $OPTARG) |
|---|
| 135 | [[ $? -ne 0 ]] && cleanupAndExit |
|---|
| 136 | ;; |
|---|
| 137 | |
|---|
| 138 | W) [[ -n $Wflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 139 | Wflag="-$OPT" |
|---|
| 140 | Wcoll="$OPTARG" |
|---|
| 141 | [[ -n $Cflag || -n $aflag || -n $wflag || -n $nflag || -n $Nflag ]] && \ |
|---|
| 142 | syntaxError "invalidCombination" \ |
|---|
| 143 | $usageMsg $Wflag $Cflag $aflag $wflag $nflag $Nflag |
|---|
| 144 | ;; |
|---|
| 145 | |
|---|
| 146 | w) [[ -n $wflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 147 | wflag="-$OPT" |
|---|
| 148 | nodenames="$OPTARG" |
|---|
| 149 | [[ -n $Cflag || -n $Wflag || -n $aflag || -n $Nflag ]] && \ |
|---|
| 150 | syntaxError "invalidCombination" \ |
|---|
| 151 | $usageMsg $wflag $Cflag $Wflag $aflag $Nflag |
|---|
| 152 | ;; |
|---|
| 153 | |
|---|
| 154 | :) syntaxError "missingValue" $usageMsg $OPTARG |
|---|
| 155 | ;; |
|---|
| 156 | |
|---|
| 157 | +[aCnNtwW]) |
|---|
| 158 | syntaxError "invalidOption" $usageMsg "$OPT" |
|---|
| 159 | ;; |
|---|
| 160 | |
|---|
| 161 | *) syntaxError "invalidOption" $usageMsg $OPTARG |
|---|
| 162 | ;; |
|---|
| 163 | |
|---|
| 164 | esac |
|---|
| 165 | done # end of while getopts :aC:n:N:t:W:w: OPT do |
|---|
| 166 | |
|---|
| 167 | shift OPTIND-1 |
|---|
| 168 | [[ $# != 0 ]] && syntaxError "extraArg" $usageMsg $1 |
|---|
| 169 | |
|---|
| 170 | # Rather than fail, convert the obsolete -C option to -a. |
|---|
| 171 | [[ -n $Cflag ]] && aflag="-a" |
|---|
| 172 | |
|---|
| 173 | |
|---|
| 174 | ########################################################################### |
|---|
| 175 | # Set up trap exception handling and call the gpfsInit function if needed. |
|---|
| 176 | # It will ensure that the local copy of the mmsdrfs and the rest of the |
|---|
| 177 | # GPFS system files are up-to-date. There is no need to lock the sdr. |
|---|
| 178 | # Note: There is no need to call gpfsInit if we are trying to quickly |
|---|
| 179 | # shut down the daemon on the local node only. |
|---|
| 180 | ########################################################################### |
|---|
| 181 | trap pretrap2 HUP INT QUIT KILL |
|---|
| 182 | if [[ -n $callGpfsInit ]] |
|---|
| 183 | then |
|---|
| 184 | gpfsInitOutput=$(gpfsInit nolock) |
|---|
| 185 | setGlobalVar $? $gpfsInitOutput |
|---|
| 186 | fi |
|---|
| 187 | |
|---|
| 188 | |
|---|
| 189 | ####################################################### |
|---|
| 190 | # Create a file containing all of the specified nodes. |
|---|
| 191 | ####################################################### |
|---|
| 192 | if [[ -n $aflag ]] |
|---|
| 193 | then |
|---|
| 194 | # Get a list of the nodes. |
|---|
| 195 | getNodeList $REL_HOSTNAME_Field $GLOBAL_ID $mmsdrfsFile > $nodefile |
|---|
| 196 | |
|---|
| 197 | elif [[ -n $Nflag ]] |
|---|
| 198 | then |
|---|
| 199 | # Convert the passed data into a file containing admin node names. |
|---|
| 200 | createVerifiedNodefile $nodenames $REL_HOSTNAME_Field no $nodefile |
|---|
| 201 | [[ $? -ne 0 ]] && cleanupAndExit |
|---|
| 202 | |
|---|
| 203 | elif [[ -n $Wflag ]] |
|---|
| 204 | then |
|---|
| 205 | # Verify the input file is readable. |
|---|
| 206 | if [[ ! -f $Wcoll || ! -r $Wcoll ]] |
|---|
| 207 | then |
|---|
| 208 | printErrorMsg 43 $mmcmd $Wcoll |
|---|
| 209 | cleanupAndExit |
|---|
| 210 | fi |
|---|
| 211 | |
|---|
| 212 | # Filter out comment lines and localhost entries. |
|---|
| 213 | $grep -v -e "localhost" -e "^#" "$Wcoll" > $tmpfile |
|---|
| 214 | |
|---|
| 215 | # Convert any entries in the node file into admin node names. |
|---|
| 216 | if [[ -s $tmpfile ]] |
|---|
| 217 | then |
|---|
| 218 | createVerifiedNodefile $tmpfile $REL_HOSTNAME_Field no $nodefile |
|---|
| 219 | [[ $? -ne 0 ]] && cleanupAndExit |
|---|
| 220 | else |
|---|
| 221 | # No node names were specified. |
|---|
| 222 | printErrorMsg 328 $mmcmd $Wcoll |
|---|
| 223 | cleanupAndExit |
|---|
| 224 | fi |
|---|
| 225 | |
|---|
| 226 | else |
|---|
| 227 | # Either no option was specified, or we have some combination of -w and -n. |
|---|
| 228 | |
|---|
| 229 | # Convert the node names list (if any) into a file. |
|---|
| 230 | $rm -f $tmpfile |
|---|
| 231 | if [[ -n $nodenames ]] |
|---|
| 232 | then |
|---|
| 233 | for i in $(print $nodenames | $tr "," " ") |
|---|
| 234 | do |
|---|
| 235 | print -- "$i" >> $tmpfile |
|---|
| 236 | done |
|---|
| 237 | fi |
|---|
| 238 | |
|---|
| 239 | # Append the node number list (if any) to the node file. |
|---|
| 240 | if [[ -n $nodenums ]] |
|---|
| 241 | then |
|---|
| 242 | for i in $(print $nodenums | $tr "," " ") |
|---|
| 243 | do |
|---|
| 244 | print -- "$i" >> $tmpfile |
|---|
| 245 | done |
|---|
| 246 | fi |
|---|
| 247 | |
|---|
| 248 | # Convert the entries in the node file into admin node names. |
|---|
| 249 | if [[ -s $tmpfile ]] |
|---|
| 250 | then |
|---|
| 251 | createVerifiedNodefile $tmpfile $REL_HOSTNAME_Field no $nodefile |
|---|
| 252 | [[ $? -ne 0 ]] && cleanupAndExit |
|---|
| 253 | fi |
|---|
| 254 | |
|---|
| 255 | fi # end of if [[ -n $aflag ]] |
|---|
| 256 | |
|---|
| 257 | # Determine the number of nodes that are affected by the command. |
|---|
| 258 | [[ -s $nodefile ]] && nodes=$($cat $nodefile | $wc -l) |
|---|
| 259 | |
|---|
| 260 | # If a timeout value was not set, estimate one. |
|---|
| 261 | if [[ $timeout = 0 ]] |
|---|
| 262 | then |
|---|
| 263 | (( timeout = 60 + 3 * nodes )) |
|---|
| 264 | [[ $nodes -gt 500 ]] && \ |
|---|
| 265 | (( maxTimeout = maxTimeout + nodes / 500 * 300 )) |
|---|
| 266 | [[ $timeout -gt $maxTimeout ]] && timeout=$maxTimeout |
|---|
| 267 | fi |
|---|
| 268 | |
|---|
| 269 | # Ensure we have the proper credentials. |
|---|
| 270 | [[ $getCredCalled = no ]] && getCred |
|---|
| 271 | |
|---|
| 272 | |
|---|
| 273 | ################################ |
|---|
| 274 | # Unmount the GPFS filesystems. |
|---|
| 275 | ################################ |
|---|
| 276 | |
|---|
| 277 | # Starting force unmount ... |
|---|
| 278 | printInfoMsg 241 "$(date)" $mmcmd |
|---|
| 279 | |
|---|
| 280 | if [[ ! -s $nodefile ]] |
|---|
| 281 | then |
|---|
| 282 | # Unmount the gpfs file systems on the local node only. |
|---|
| 283 | $mmremote unmountFileSystems all -f & |
|---|
| 284 | pid=$! |
|---|
| 285 | else |
|---|
| 286 | # Unmount the gpfs file systems on a number of nodes. |
|---|
| 287 | $mmcommon onall $nodefile $unreachedNodes unmountFileSystems all -f & |
|---|
| 288 | pid=$! |
|---|
| 289 | fi |
|---|
| 290 | |
|---|
| 291 | # Wait until the unmount command finishes or the timeout expires. |
|---|
| 292 | waited=0 |
|---|
| 293 | mincount=0 |
|---|
| 294 | retrydelay=5 |
|---|
| 295 | while [[ $waited -lt $timeout ]] |
|---|
| 296 | do |
|---|
| 297 | $sleep $retrydelay |
|---|
| 298 | (( waited = waited + retrydelay )) |
|---|
| 299 | (( mincount = mincount + retrydelay )) |
|---|
| 300 | |
|---|
| 301 | if ! kill -0 $pid 2>/dev/null |
|---|
| 302 | then |
|---|
| 303 | # unmount finished. |
|---|
| 304 | waited=-1 |
|---|
| 305 | break |
|---|
| 306 | fi |
|---|
| 307 | |
|---|
| 308 | if [[ $mincount -ge 60 ]] |
|---|
| 309 | then |
|---|
| 310 | (( togo = timeout - waited )) |
|---|
| 311 | mincount=0 |
|---|
| 312 | printInfoMsg 242 "$(date)" $mmcmd $waited $togo |
|---|
| 313 | fi |
|---|
| 314 | done # while [[ $waited -lt $timeout ]] |
|---|
| 315 | |
|---|
| 316 | # Inform the user if the unmount did not finish. |
|---|
| 317 | [[ $waited -gt 0 ]] && \ |
|---|
| 318 | printInfoMsg 243 "$(date)" $mmcmd $waited |
|---|
| 319 | |
|---|
| 320 | |
|---|
| 321 | ############################## |
|---|
| 322 | # Shut down the GPFS daemons. |
|---|
| 323 | ############################## |
|---|
| 324 | printInfoMsg 244 "$(date)" $mmcmd |
|---|
| 325 | |
|---|
| 326 | if [[ ! -s $nodefile ]] |
|---|
| 327 | then |
|---|
| 328 | # Shut down GPFS on the local node only. |
|---|
| 329 | $mmremote shutdownDaemon |
|---|
| 330 | rc=$? |
|---|
| 331 | else |
|---|
| 332 | # Shut down GPFS on a number of nodes. |
|---|
| 333 | $mmcommon onall $nodefile $unreachedNodes shutdownDaemon |
|---|
| 334 | rc=$? |
|---|
| 335 | fi |
|---|
| 336 | |
|---|
| 337 | # If any nodes could not be reached, tell the user which ones. |
|---|
| 338 | if [[ -s $unreachedNodes ]] |
|---|
| 339 | then |
|---|
| 340 | # The following nodes could not be reached: . . . |
|---|
| 341 | printErrorMsg 270 $mmcmd |
|---|
| 342 | $cat $unreachedNodes 1>&2 |
|---|
| 343 | fi |
|---|
| 344 | |
|---|
| 345 | # All finished. |
|---|
| 346 | printInfoMsg 245 "$(date)" $mmcmd |
|---|
| 347 | |
|---|
| 348 | cleanupAndExit $rc |
|---|
| 349 | |
|---|
