source: gpfs_3.1_ker2.6.20/lpp/mmfs/bin/mmstartup @ 77

Last change on this file since 77 was 16, checked in by rock, 17 years ago
  • Property svn:executable set to *
File size: 12.8 KB
Line 
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. 2000,2006
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# @(#)78 1.32.1.2 src/avs/fs/mmfs/ts/admin/mmstartup.sh, mmfs, avs_rgpfs24, rgpfs24s003a 5/26/06 09:48:07
17##############################################################################
18#
19#  The GPFS daemons on the specified nodes will be started.
20#
21#  Usage:
22#
23#    mmstartup [-a | -N {Node[,Node...] | NodeFile | NodeClass}]
24#              [-E EnvVar=Value...] [ -T ]
25#
26#  where:
27#
28#    -a            Start the daemon on all nodes in the GPFS cluster.
29#
30#    -N Node,Node,...  Specify the nodes on which the daemon is to be started.
31#    -N NodeFile       NodeClass may be one of several possible node classes
32#    -N NodeClass      (e.g., quorumnodes, managernodes, nsdnodes, etc.)
33#
34#    -E EnvVar=Value  Specify an environment variable to be passed to the GPFS
35#                     daemon.  More than one -E option can be specified.
36#
37#    -T            start tracing at daemon startup and cut a trace report
38#                  at daemon shutdown
39#
40#
41#  If not explicitly specified otherwise, the daemon is started on the local
42#  node only.
43#
44#
45#  Undocumented options:
46#
47#    -e seconds    Estimated startup time (in seconds).  During this time
48#                  period, the mmsdrfs file is guaranteed not to change.
49#                  Specifying a value of 0 disables the locking that would
50#                  otherwise take place for large non-sp clusters.
51#
52#    -f            Force the loading of the kernel extensions (Linux only).
53#
54#    -G            Force trace records to be cut on all nodes when there is
55#                  a daemon failure on some node.
56#
57#    -t {yes|traceFile}  Start the mm commands tracing facility.  If traceFile
58#                   is specified, it must be a fully qualified pathname.
59#                   Otherwise, the trace results are appended to file
60#                   /tmp/mmfs/mmScriptTrace.
61#
62#
63#  Obsolete options:
64#
65#    -C NodesetId  Start the daemon on all nodes in the specified nodeset.
66#                  Assumed to be the same as -a.
67#
68#    -W NodeFile   Start the daemon on all nodes whose reliable hostnames
69#                  are listed one per line in NodeFilename.
70#                  Cannot be specified with -a, -N, -w, or -n.
71#
72#    -w nodenames  Start the daemon on all nodes whose reliable hostnames
73#                  are in the comma-separated nodenames list.
74#                  Cannot be specified with -a, -N or -W.
75#                  If both -w and -n are specified, the lists are combined.
76#
77#    -n nodenums   Start the daemon on all nodes whose node numbers are
78#                  in the comma-separated nodenums list.
79#                  Cannot be specified with -a, -N, or -W.
80#                  If both -w and -n are specified, the lists are combined.
81#
82##############################################################################
83
84# Include global declarations and service routines.
85. /usr/lpp/mmfs/bin/mmglobfuncs
86. /usr/lpp/mmfs/bin/mmsdrfsdef
87
88sourceFile="mmstartup.sh"
89[[ -n $DEBUG || -n $DEBUGmmstartup ]] && set -x
90$mmTRACE_ENTER "$*"
91
92
93# Local variables
94
95usageMsg=391
96typeset -i timeout=0
97typeset -i nodes=0
98rc=0
99nodenames=""
100nodenums=""
101wcoll=""
102aflag=""
103Cflag=""
104eflag=""
105Eflag=""
106Estring=""
107fflag=""
108Gflag=""
109nflag=""
110Nflag=""
111Tflag=""
112tflag=""
113Wflag=""
114wflag=""
115nodeList=""
116
117
118
119#######################
120# Mainline processing.
121#######################
122
123
124##################################
125# Process each of the arguments.
126##################################
127[[ $arg1 = '-?' || $arg1 = '-h' || $arg1 = '--help' || $arg1 = '--' ]] &&  \
128  syntaxError "help" $usageMsg
129
130while getopts :aC:e:E:fGn:N:t:Tw:W: OPT
131do
132  case $OPT in
133
134    a) [[ -n $aflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
135       aflag="-$OPT"
136       [[ -n $Cflag || -n $Wflag || -n $wflag || -n $nflag || -n $Nflag ]] &&  \
137         syntaxError "invalidCombination"  \
138           $usageMsg $aflag $Cflag $Wflag $wflag $nflag $Nflag
139       ;;
140
141    C) # syntaxError "obsoleteOption" $usageMsg "-$OPT"
142       [[ -n $Cflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
143       Cflag="-$OPT"
144       nodesetId="$OPTARG"
145       [[ -n $aflag || -n $Wflag || -n $wflag || -n $nflag || -n $Nflag ]] &&  \
146         syntaxError "invalidCombination"  \
147           $usageMsg $Cflag $aflag $Wflag $wflag $nflag $Nflag
148       ;;
149
150    e) [[ -n $eflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
151       eflag="-$OPT"
152       estimatedStartupTime=$(checkIntRange estimatedStartupTime "$OPTARG")
153       [[ $? -ne 0 ]] && cleanupAndExit
154       ;;
155
156    E) # more than one -E option is allowed
157       envString="${envString} -$OPT $OPTARG"
158       ;;
159
160    f) [[ -n $fflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
161       fflag="-$OPT"
162       [[ $osName != Linux ]] && syntaxError "invalidOption" $usageMsg "-$OPT"
163       ;;
164
165    G) [[ -n $Gflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
166       Gflag="-$OPT"
167       ;;
168
169    n) [[ -n $nflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
170       nflag="-$OPT"
171       nodenums="$OPTARG"
172       [[ -n $Cflag || -n $Wflag || -n $aflag || -n $Nflag ]] &&  \
173         syntaxError "invalidCombination"  \
174           $usageMsg $nflag $Cflag $Wflag $aflag $Nflag
175       ;;
176
177    N) [[ -n $Nflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
178       Nflag="-$OPT"
179       nodenames="$OPTARG"
180       [[ -n $Cflag || -n $Wflag || -n $aflag || -n $nflag || -n $wflag ]] &&  \
181         syntaxError "invalidCombination"  \
182           $usageMsg $Nflag $Cflag $Wflag $aflag $nflag $wflag
183       ;;
184
185    t) [[ -n $tflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
186       tflag="-$OPT $OPTARG"
187       ;;
188
189    T) [[ -n $Tflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
190       Tflag="-$OPT"
191       ;;
192
193    w) [[ -n $wflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
194       wflag="-$OPT"
195       nodenames="$OPTARG"
196       [[ -n $Cflag || -n $Wflag || -n $aflag || -n $Nflag ]] &&  \
197         syntaxError "invalidCombination"  \
198           $usageMsg $wflag $Cflag $Wflag $aflag $Nflag
199       ;;
200
201    W) [[ -n $Wflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT"
202       Wflag="-$OPT"
203       wcoll="$OPTARG"
204       [[ -n $Cflag || -n $aflag || -n $wflag || -n $nflag || -n $Nflag ]] &&  \
205         syntaxError "invalidCombination"  \
206           $usageMsg $Wflag $Cflag $aflag $wflag $nflag $Nflag
207       ;;
208
209    :) syntaxError "missingValue" $usageMsg $OPTARG
210       ;;
211
212    +[aCeEfnNtTwW])
213       syntaxError "invalidOption" $usageMsg "$OPT"
214       ;;
215
216    *) syntaxError "invalidOption" $usageMsg $OPTARG
217       ;;
218
219  esac
220done
221
222shift OPTIND-1
223[[ $# != 0 ]] && syntaxError "extraArg" $usageMsg $1
224
225# Rather than fail, convert the obsolete -C option to -a.
226[[ -n $Cflag ]] && aflag="-a"
227
228
229########################################################################
230# Set up trap exception handling and call the gpfsInit function.
231# It will ensure that the local copy of the mmsdrfs and the rest of the
232# GPFS system files are up-to-date.  There is no need to lock the sdr.
233########################################################################
234trap pretrap2 HUP INT QUIT KILL
235gpfsInitOutput=$(gpfsInit nolock)
236setGlobalVar $? $gpfsInitOutput
237
238
239#######################################################
240# Create a file containing all of the specified nodes.
241#######################################################
242if [[ -n $aflag ]]
243then
244  # Get a list of the nodes.
245  getNodeList $REL_HOSTNAME_Field $GLOBAL_ID $mmsdrfsFile > $nodefile
246
247elif [[ -n $Nflag ]]
248then
249  # Convert the passed data into a file containing admin node names.
250  createVerifiedNodefile $nodenames $REL_HOSTNAME_Field no $nodefile
251  [[ $? -ne 0 ]] && cleanupAndExit
252
253elif [[ -n $Wflag ]]
254then
255  # Verify the input file is readable.
256  if [[ ! -f $Wcoll || ! -r $Wcoll ]]
257  then
258    printErrorMsg 43 $mmcmd $Wcoll
259    cleanupAndExit
260  fi
261
262  # Filter out comment lines and localhost entries.
263  $grep -v -e "localhost" -e "^#" "$Wcoll" > $tmpfile
264
265  # Convert any entries in the node file into admin node names.
266  if [[ -s $tmpfile ]]
267  then
268    createVerifiedNodefile $tmpfile $REL_HOSTNAME_Field no $nodefile
269    [[ $? -ne 0 ]] && cleanupAndExit
270  else
271    # No node names were specified.
272    printErrorMsg 328 $mmcmd $Wcoll
273    cleanupAndExit
274  fi
275
276else
277  # Either no option was specified, or we have some combination of -w and -n.
278
279  # Convert the node names list (if any) into a file.
280  $rm -f $tmpfile
281  if [[ -n $nodenames ]]
282  then
283    for i in $(print $nodenames | $tr "," " ")
284    do
285      print -- "$i" >> $tmpfile
286    done
287  fi
288
289  # Append the node number list (if any) to the node file.
290  if [[ -n $nodenums ]]
291  then
292    for i in $(print $nodenums | $tr "," " ")
293    do
294      print -- "$i" >> $tmpfile
295    done
296  fi
297
298  # Convert the entries in the node file into admin node names.
299  if [[ -s $tmpfile ]]
300  then
301    createVerifiedNodefile $tmpfile $REL_HOSTNAME_Field no $nodefile
302    [[ $? -ne 0 ]] && cleanupAndExit
303  fi
304
305fi  # end of if [[ -n $aflag ]]
306
307
308# Ensure we have the proper credentials.
309[[ $getCredCalled = no ]] && getCred
310
311
312##############################################################
313# If starting GPFS on a large number of nodes, lock the sdr.
314# The goal is to prevent the config data from changing and to
315# inform the rest of the nodes that they can trust their data
316# without having to check with the server nodes first.
317##############################################################
318[[ -s $nodefile ]] && nodes=$($cat $nodefile | $wc -l)
319if [[ -z $eflag && $nodes -gt 32 ||
320      -n $eflag && $estimatedStartupTime -gt 0 ]]
321then
322  # If the estimated startup time is given by the user,
323  # use the value from the command line.  Otherwise, pick
324  # a number based on the number of nodes to start.
325  if [[ -z $eflag ]]
326  then
327    if [[ $nodes -lt 128 ]]
328    then
329      estimatedStartupTime=90
330    elif [[ $nodes -lt 256 ]]
331    then
332      estimatedStartupTime=150
333    else
334      estimatedStartupTime=240
335    fi
336
337    # Add additional time for each file system that will be mounted.
338    fsToMount=$($cat $startupMountFile 2>/dev/null | $wc -l )
339    [[ $fsToMount -gt 0 ]] &&  \
340      (( estimatedStartupTime = estimatedStartupTime + 10 * $fsToMount ))
341  fi  # end of if [[ -z $eflag ]]
342
343  # Create a special lock id.
344  expLockId="mmSdrLockExp:$ourNodeName:$estimatedStartupTime"
345
346  # Try to obtain the sdr lock.
347  gpfsInitOutput=$(gpfsInit $expLockId 2>/dev/null)
348  rc=$?
349  if [[ $rc -eq 0 ]]
350  then
351    # We got the lock.  Parse the output to get the latest
352    # generation number and corresponding timestamp.
353    setGlobalVar $rc $gpfsInitOutput
354
355    # Create the expiration token to be passed to the nodes.
356    currentTime=$($perl -e 'print time')
357    (( expirationTime = currentTime + estimatedStartupTime ))
358    expirationData="mmSdrLockExp:$sdrGenNumber:$sdrGenTimestamp:$expirationTime"
359
360    # Start the background process that will free the lock.
361    # Reset the sdrLocked var to prevent the unlocking that
362    # would otherwise take place as part of cleanupAndExit.
363    sdrLocked=no
364    $mmcommon expirationDataCleanup $expirationData unlock >/dev/null 2>&1 &
365
366    # Add an option letter in front of the string.
367    expirationData="-e $expirationData"
368
369  else
370    # gpfsInit failed and we did not get the lock.
371    # Ignore the error and continue with the regular processing.
372    expirationData=""
373  fi  # end of if [[ $rc -eq 0 ]]
374
375else
376  # Do not do anything special if starting just a few nodes,
377  # or the user explicitly requested no locking (-e 0 on the command line).
378  expirationData=""
379fi  # end of if [[ $nodes -gt 32 ]]
380
381
382####################################################################
383# If daemon tracing is specified, decide what its scope should be.
384####################################################################
385if [[ -n $Tflag ]]
386then
387  if [[ -n $Gflag ]]
388  then
389    Tflag="-T global"
390  else
391    Tflag="-T local"
392  fi
393elif [[ -n $Gflag ]]
394then
395  Tflag="-T global"
396else
397  Tflag=""
398fi  # end of if [[ -n $Tflag ]]
399
400
401############################################
402# Start GPFS daemon on the specified nodes.
403############################################
404
405# Starting GPFS ...
406printInfoMsg 392 "$(date)" $mmcmd
407
408if [[ ! -s $nodefile ]]
409then
410  # The request is to start the local daemon only.
411  $mmremote startSubsys $fflag $Tflag $tflag $envString
412  rc=$?
413else
414  # The request is to start the daemon on a number of nodes.
415  $mmcommon onall $nodefile $unreachedNodes  \
416    startSubsys $expirationData $fflag $Tflag $tflag $envString
417  rc=$?
418fi
419
420# If any nodes could not be reached, tell the user which ones.
421if [[ -s $unreachedNodes ]]
422then
423  # The following nodes could not be reached: . . .
424  printErrorMsg 270 $mmcmd
425  $cat $unreachedNodes 1>&2
426fi
427
428cleanupAndExit $rc
429
Note: See TracBrowser for help on using the repository browser.