src/pmie/pmie_check.sh

   1 #! /bin/sh
   2 #
   3 # Copyright (c) 2013-2016 Red Hat.
   4 # Copyright (c) 1998-2000,2003 Silicon Graphics, Inc.  All Rights Reserved.
   5 #
   6 # This program is free software; you can redistribute it and/or modify it
   7 # under the terms of the GNU General Public License as published by the
   8 # Free Software Foundation; either version 2 of the License, or (at your
   9 # option) any later version.
  10 #
  11 # This program is distributed in the hope that it will be useful, but
  12 # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  13 # or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  14 # for more details.
  15 #
  16 # Administrative script to check pmie processes are alive, and restart
  17 # them as required.
  18 #
  19
  20 # Get standard environment
  21 . $PCP_DIR/etc/pcp.env
  22 . $PCP_SHARE_DIR/lib/rc-proc.sh
  23
  24 PMIE="$PCP_BIN_DIR/pmie"
  25 PMIECONF="$PCP_BIN_DIR/pmieconf"
  26
  27 # error messages should go to stderr, not the GUI notifiers
  28 #
  29 unset PCP_STDERR
  30
  31 # added to handle problem when /var/log/pcp is a symlink, as first
  32 # reported by Micah_Altman@harvard.edu in Nov 2001
  33 #
  34 _unsymlink_path()
  35 {
  36     [ -z "$1" ] && return
  37     __d=`dirname $1`
  38     __real_d=`cd $__d 2>/dev/null && $PWDCMND`
  39     if [ -z "$__real_d" ]
  40     then
  41         echo $1
  42     else
  43         echo $__real_d/`basename $1`
  44     fi
  45 }
  46
  47 # constant setup
  48 #
  49 tmp=`mktemp -d /tmp/pcp.XXXXXXXXX` || exit 1
  50 status=0
  51 echo >$tmp/lock
  52 prog=`basename $0`
  53 PROGLOG=$PCP_LOG_DIR/pmie/$prog.log
  54 USE_SYSLOG=true
  55
  56 _cleanup()
  57 {
  58     $USE_SYSLOG && [ $status -ne 0 ] && \
  59     $PCP_SYSLOG_PROG -p daemon.error "$prog failed - see $PROGLOG"
  60     [ -s "$PROGLOG" ] || rm -f "$PROGLOG"
  61     lockfile=`cat $tmp/lock 2>/dev/null`
  62     rm -f "$lockfile"
  63     rm -rf $tmp
  64 }
  65 trap "_cleanup; exit \$status" 0 1 2 3 15
  66
  67 # control files for pmie administration ... edit the entries in this
  68 # file (and optional directory) to reflect your local configuration;
  69 # see also -c option below.
  70 #
  71 CONTROL=$PCP_PMIECONTROL_PATH
  72 CONTROLDIR=$PCP_PMIECONTROL_PATH.d
  73
  74 # determine path for pwd command to override shell built-in
  75 PWDCMND=`which pwd 2>/dev/null | $PCP_AWK_PROG '
  76 BEGIN           { i = 0 }
  77 / not in /      { i = 1 }
  78 / aliased to /  { i = 1 }
  79                 { if ( i == 0 ) print }
  80 '`
  81 [ -z "$PWDCMND" ] && PWDCMND=/bin/pwd
  82 eval $PWDCMND -P >/dev/null 2>&1
  83 [ $? -eq 0 ] && PWDCMND="$PWDCMND -P"
  84 here=`$PWDCMND`
  85
  86 # option parsing
  87 #
  88 SHOWME=false
  89 MV=mv
  90 RM=rm
  91 CP=cp
  92 KILL=pmsignal
  93 TERSE=false
  94 VERBOSE=false
  95 VERY_VERBOSE=false
  96 CHECK_RUNLEVEL=false
  97 START_PMIE=true
  98 STOP_PMIE=false
  99
 100 echo > $tmp/usage
 101 cat >> $tmp/usage << EOF
 102 Options:
 103   -c=FILE,--control=FILE  configuration of pmie instances to manage
 104   -l=FILE,--logfile=FILE  send important diagnostic messages to FILE
 105   -C                      query system service runlevel information
 106   -N,--showme             perform a dry run, showing what would be done
 107   -s,--stop               stop pmie processes instead of starting them
 108   -T,--terse              produce a terser form of output
 109   -V,--verbose            increase diagnostic verbosity
 110   --help
 111 EOF
 112
 113 ARGS=`pmgetopt --progname=$prog --config=$tmp/usage -- "$@"`
 114 [ $? != 0 ] && exit 1
 115
 116 eval set -- "$ARGS"
 117 while [ $# -gt 0 ]
 118 do
 119     case "$1"
 120     in
 121         -c)     CONTROL="$2"
 122                 CONTROLDIR="$2.d"
 123                 shift
 124                 ;;
 125         -C)     CHECK_RUNLEVEL=true
 126                 ;;
 127         -l)     PROGLOG="$2"
 128                 USE_SYSLOG=false
 129                 shift
 130                 ;;
 131         -N)     SHOWME=true
 132                 USE_SYSLOG=false
 133                 MV="echo + mv"
 134                 RM="echo + rm"
 135                 CP="echo + cp"
 136                 KILL="echo + kill"
 137                 ;;
 138         -s)     START_PMIE=false
 139                 STOP_PMIE=true
 140                 ;;
 141         -T)     TERSE=true
 142                 ;;
 143         -V)     if $VERBOSE
 144                 then
 145                     VERY_VERBOSE=true
 146                 else
 147                     VERBOSE=true
 148                 fi
 149                 ;;
 150         --)     shift
 151                 break
 152                 ;;
 153         -\?)    pmgetopt --usage --progname=$prog --config=$tmp/usage
 154                 status=1
 155                 exit
 156                 ;;
 157     esac
 158     shift
 159 done
 160
 161 if [ $# -ne 0 ]
 162 then
 163     pmgetopt --usage --progname=$prog --config=$tmp/usage
 164     status=1
 165     exit
 166 fi
 167
 168 # after argument checking, everything must be logged to ensure no mail is
 169 # accidentally sent from cron.  Close stdout and stderr, then open stdout
 170 # as our logfile and redirect stderr there too.
 171 #
 172 PROGLOGDIR=`dirname "$PROGLOG"`
 173 [ -d "$PROGLOGDIR" ] || mkdir -p -m 775 "$PROGLOGDIR" 2>/dev/null
 174 [ -f "$PROGLOG" ] && mv "$PROGLOG" "$PROGLOG.prev"
 175 exec 1>"$PROGLOG"
 176 exec 2>&1
 177
 178 _error()
 179 {
 180     echo "$prog: [$controlfile:$line]"
 181     echo "Error: $@"
 182     echo "... automated performance reasoning for host \"$host\" unchanged"
 183     touch $tmp/err
 184 }
 185
 186 _warning()
 187 {
 188     echo "$prog [$controlfile:$line]"
 189     echo "Warning: $@"
 190 }
 191
 192 _restarting()
 193 {
 194     $PCP_ECHO_PROG $PCP_ECHO_N "Restarting pmie for host \"$host\" ...""$PCP_ECHO_C"
 195 }
 196
 197 _lock()
 198 {
 199     # demand mutual exclusion
 200     #
 201     rm -f $tmp/stamp $tmp/out
 202     delay=200           # tenths of a second
 203     while [ $delay -ne 0 ]
 204     do
 205         if pmlock -v $logfile.lock >$tmp/out
 206         then
 207             echo $logfile.lock >$tmp/lock
 208             break
 209         else
 210             if [ ! -f $tmp/stamp ]
 211             then
 212                 touch -t `pmdate -30M %Y%m%d%H%M` $tmp/stamp
 213             fi
 214             if [ -n "`find $logfile.lock ! -newer $tmp/stamp -print 2>/dev/null`" ]
 215             then
 216                 _warning "removing lock file older than 30 minutes"
 217                 ls -l $logfile.lock
 218                 rm -f $logfile.lock
 219             fi
 220         fi
 221         pmsleep 0.1
 222         delay=`expr $delay - 1`
 223     done
 224
 225     if [ $delay -eq 0 ]
 226     then
 227         # failed to gain mutex lock
 228         #
 229         if [ -f $logfile.lock ]
 230         then
 231             _warning "is another PCP cron job running concurrently?"
 232             ls -l $logfile.lock
 233         else
 234             echo "$prog: `cat $tmp/out`"
 235         fi
 236         _warning "failed to acquire exclusive lock ($logfile.lock) ..."
 237         continue
 238     fi
 239 }
 240
 241 _unlock()
 242 {
 243     rm -f $logfile.lock
 244     echo >$tmp/lock
 245 }
 246
 247 _check_logfile()
 248 {
 249     if [ ! -f $logfile ]
 250     then
 251         echo "$prog: Error: cannot find pmie output file at \"$logfile\""
 252         if $TERSE
 253         then
 254             :
 255         else
 256             logdir=`dirname $logfile`
 257             echo "Directory (`cd $logdir; $PWDCMND`) contents:"
 258             LC_TIME=POSIX ls -la $logdir
 259         fi
 260     else
 261         echo "Contents of pmie output file \"$logfile\" ..."
 262         cat $logfile
 263     fi
 264 }
 265
 266 _check_pmie()
 267 {
 268     $VERBOSE && $PCP_ECHO_PROG $PCP_ECHO_N " [process $1] ""$PCP_ECHO_C"
 269
 270     # wait until pmie process starts, or exits
 271     #
 272     delay=5
 273     [ ! -z "$PMCD_CONNECT_TIMEOUT" ] && delay=$PMCD_CONNECT_TIMEOUT
 274     x=5
 275     [ ! -z "$PMCD_REQUEST_TIMEOUT" ] && x=$PMCD_REQUEST_TIMEOUT
 276
 277     # wait for maximum time of a connection and 20 requests
 278     #
 279     delay=`expr \( $delay + 20 \* $x \) \* 10`  # tenths of a second
 280     while [ $delay -ne 0 ]
 281     do
 282         if [ -f $logfile ]
 283         then
 284             # $logfile was previously removed, if it has appeared again then
 285             # we know pmie has started ... if not just sleep and try again
 286             #
 287             if ls "$PCP_TMP_DIR/pmie/$1" >$tmp/out 2>&1
 288             then
 289                 if grep "No such file or directory" $tmp/out >/dev/null
 290                 then
 291                     :
 292                 else
 293                     $VERBOSE && echo " done"
 294                     return 0
 295                 fi
 296             fi
 297
 298             _plist=`_get_pids_by_name pmie`
 299             _found=false
 300             for _p in `echo $_plist`
 301             do
 302                 [ $_p -eq $1 ] && _found=true
 303             done
 304
 305             if $_found
 306             then
 307                 # process still here, just hasn't created its status file
 308                 # yet, try again
 309                 :
 310             else
 311                 $VERBOSE || _restarting
 312                 echo " process exited!"
 313                 if $TERSE
 314                 then
 315                     :
 316                 else
 317                     echo "$prog: Error: failed to restart pmie"
 318                     echo "Current pmie processes:"
 319                     $PCP_PS_PROG $PCP_PS_ALL_FLAGS | tee $tmp/tmp | sed -n -e 1p
 320                     for _p in `echo $_plist`
 321                     do
 322                         sed -n -e "/^[ ]*[^ ]* [ ]*$_p /p" < $tmp/tmp
 323                     done
 324                     echo
 325                 fi
 326                 _check_logfile
 327                 return 1
 328             fi
 329         fi
 330         pmsleep 0.1
 331         delay=`expr $delay - 1`
 332         $VERBOSE && [ `expr $delay % 10` -eq 0 ] && \
 333                         $PCP_ECHO_PROG $PCP_ECHO_N ".""$PCP_ECHO_C"
 334     done
 335     $VERBOSE || _restarting
 336     echo " timed out waiting!"
 337     if $TERSE
 338     then
 339         :
 340     else
 341         sed -e 's/^/    /' $tmp/out
 342     fi
 343     _check_logfile
 344     return 1
 345 }
 346
 347 _get_configfile()
 348 {
 349     # extract the pmie configuration file (-c) from a list of arguments
 350     #
 351     echo $@ | sed -n \
 352         -e 's/^/ /' \
 353         -e 's/[         ][      ]*/ /g' \
 354         -e 's/-c /-c/' \
 355         -e 's/.* -c\([^ ]*\).*/\1/p'
 356 }
 357
 358 _configure_pmie()
 359 {
 360     # update a pmie configuration file if it should be created/modified
 361     #
 362     tmpconfig="$1.tmp"
 363     configfile="$1"
 364     isprimary="$3"
 365
 366     if [ -f "$configfile" ]
 367     then
 368         # look for "magic" string at start of file, and ensure we created it
 369         sed 1q "$configfile" | grep '^// pmieconf-pmie [0-9]' >/dev/null
 370         magic=$?
 371         grep '^// Auto-generated by pmieconf' "$configfile" >/dev/null
 372         owned=$?
 373         if [ $magic -eq 0 -a $owned -eq 0 ]
 374         then
 375             # pmieconf file, see if re-generation is needed
 376             cp "$configfile" "$tmpconfig"
 377             if $PMIECONF -f "$tmpconfig" -cF >$tmp/diag 2>&1
 378             then
 379                 [ $isprimary = y ] && $PMIECONF -f "$tmpconfig" modify primary enabled yes
 380                 grep -v "generated by pmieconf" "$configfile" >$tmp/old
 381                 grep -v "generated by pmieconf" "$tmpconfig" >$tmp/new
 382                 if ! diff $tmp/old $tmp/new >/dev/null
 383                 then
 384                     if [ -w "$configfile" ]
 385                     then
 386                         $VERBOSE && echo "Reconfigured: \"$configfile\" (pmieconf)"
 387                         eval $CP "$tmpconfig" "$configfile"
 388                     else
 389                         _warning "no write access to pmieconf file \"$configfile\", skip reconfiguration"
 390                         ls -l "$configfile"
 391                     fi
 392                 fi
 393             else
 394                 _warning "pmieconf failed to reconfigure \"$configfile\""
 395                 sed -e "s;$tmpconfig;$configfile;g" $tmp/diag
 396                 echo "=== start pmieconf file ==="
 397                 cat "$tmpconfig"
 398                 echo "=== end pmieconf file ==="
 399             fi
 400         fi
 401     elif [ ! -e "$configfile" ]
 402     then
 403         # file does not exist, generate it, if possible
 404         if $SHOWME
 405         then
 406             echo "+ $PMIECONF -f $configfile -cF"
 407         elif ! $PMIECONF -f "$configfile" -cF >$tmp/diag 2>&1
 408         then
 409             _warning "pmieconf failed to generate \"$configfile\""
 410             cat $tmp/diag
 411             echo "=== start pmieconf file ==="
 412             cat "$configfile"
 413             echo "=== end pmieconf file ==="
 414         else
 415             [ $isprimary = y ] && $PMIECONF -f "$configfile" modify primary enabled yes
 416             chown $PCP_USER:$PCP_GROUP "$configfile" >/dev/null 2>&1
 417         fi
 418     fi
 419 }
 420
 421 QUIETLY=false
 422 if [ $CHECK_RUNLEVEL = true ]
 423 then
 424     # determine whether to start pmie based on runlevel settings - we
 425     # need to do this when running unilaterally from cron, else we'll
 426     # always start pmie up (even when we shouldn't).
 427     #
 428     QUIETLY=true
 429     if is_chkconfig_on pmie
 430     then
 431         START_PMIE=true
 432     else
 433         START_PMIE=false
 434     fi
 435 fi
 436
 437 if [ $STOP_PMIE = true ]
 438 then
 439     # if pmie has never been started, there's no work to do to stop it
 440     [ ! -d "$PCP_TMP_DIR/pmie" ] && exit
 441     $QUIETLY || $PCP_BINADM_DIR/pmpost "stop pmie from $prog"
 442 elif [ $START_PMIE = false ]
 443 then
 444     exit
 445 fi
 446
 447 if [ ! -f "$CONTROL" ]
 448 then
 449     echo "$prog: Error: cannot find control file ($CONTROL)"
 450     status=1
 451     exit
 452 fi
 453
 454 # note on control file format version
 455 #   1.0 was the first release, and did not include the primary field
 456 #        [this is the default for backwards compatibility]
 457 #   1.1 adds the primary field (ala pmlogger control file) indicating
 458 #        localhost-specific rules should be enabled
 459 #
 460 version=''
 461
 462 rm -f $tmp/err $tmp/pmies
 463
 464 _parse_control()
 465 {
 466     controlfile="$1"
 467     line=0
 468
 469     if echo "$controlfile" | grep -q -e '\.rpmsave' -e '\.rpmnew'
 470     then
 471         _warning "ignored backup control file \"$controlfile\""
 472         return
 473     fi
 474
 475     sed -e "s;PCP_LOG_DIR;$PCP_LOG_DIR;g" $controlfile | \
 476     while read host primary socks logfile args
 477     do
 478         # start in one place for each iteration (beware relative paths)
 479         cd "$here"
 480         line=`expr $line + 1`
 481
 482         case "$host"
 483         in
 484             \#*|'')     # comment or empty
 485                 continue
 486                 ;;
 487             \$*)        # in-line variable assignment
 488                 $SHOWME && echo "# $host $primary $socks $logfile $args"
 489                 cmd=`echo "$host $primary $socks $logfile $args" \
 490                      | sed -n \
 491                          -e "/='/s/\(='[^']*'\).*/\1/" \
 492                          -e '/="/s/\(="[^"]*"\).*/\1/' \
 493                          -e '/=[^"'"'"']/s/[;&<>|].*$//' \
 494                          -e '/^\\$[A-Za-z][A-Za-z0-9_]*=/{
 495 s/^\\$//
 496 /^\([A-Za-z][A-Za-z0-9_]*\)=/s//export \1; \1=/p
 497 }'`
 498                 if [ -z "$cmd" ]
 499                 then
 500                     # in-line command, not a variable assignment
 501                     _warning "in-line command is not a variable assignment, line ignored"
 502                 else
 503                     case "$cmd"
 504                     in
 505                         'export PATH;'*)
 506                             _warning "cannot change \$PATH, line ignored"
 507                             ;;
 508                         'export IFS;'*)
 509                             _warning "cannot change \$IFS, line ignored"
 510                             ;;
 511                         *)
 512                             $SHOWME && echo "+ $cmd"
 513                             echo eval $cmd >>$tmp/cmd
 514                             eval $cmd
 515                             ;;
 516                     esac
 517                 fi
 518                 continue
 519                 ;;
 520         esac
 521
 522         # set the version and other variables
 523         #
 524         if [ -f $tmp/cmd ]
 525         then
 526             . $tmp/cmd
 527             if grep 'version=' $tmp/cmd >/dev/null
 528             then
 529                 case "$version"
 530                 in
 531                     1.0|1.1)
 532                         ;;
 533                     *)
 534                         _error "bad version ($version) in control file"
 535                         return
 536                         ;;
 537                 esac
 538             fi
 539         fi
 540
 541         if [ -z "$version" ]
 542         then
 543             _warning "missing \$version, assuming version 1.0 control format"
 544             version=1.0
 545         fi
 546         if [ "$version" = "1.0" ]
 547         then
 548             # handle backwards compatibility
 549             # one less field and primary defaults to "n" for version 1.0
 550             #
 551             if [ -n "$logfile" ]
 552             then
 553                 args="$logfile $args"
 554             else
 555                 # missing "args" ... this is bad, but will be reported below
 556                 # ... guard avoids setting "args" to " " which would defeat
 557                 # the test below
 558                 #
 559                 args=""
 560             fi
 561             logfile="$socks"
 562             socks="$primary"
 563             primary=n
 564         fi
 565
 566         if [ -z "$primary" -o -z "$socks" -o -z "$logfile" -o -z "$args" ]
 567         then
 568             _error "insufficient fields in control file record"
 569             continue
 570         fi
 571         if [ "$primary" != y -a "$primary" != n ]
 572         then
 573             _error "primary field in control file record must be y or n, not \"$primary\""
 574             continue
 575         fi
 576         if [ "$socks" != y -a "$socks" != n ]
 577         then
 578             _error "socks field in control file record must be y or n, not \"$socks\""
 579             continue
 580         fi
 581
 582         # args should begin with a hyphen
 583         #
 584         case "$args"
 585         in
 586             -*)
 587                 ;;
 588             *)
 589                 _error "args field in control file must begin with a hyphen not \"$args\""
 590                 continue
 591                 ;;
 592         esac
 593
 594         # substitute LOCALHOSTNAME marker in this config line
 595         # (differently for logfile and pcp -h HOST arguments)
 596         #
 597         logfilehost=`hostname || echo localhost`
 598         logfile=`echo $logfile | sed -e "s;LOCALHOSTNAME;$logfilehost;"`
 599         logfile=`_unsymlink_path $logfile`
 600         [ $primary = y -o "x$host" = xLOCALHOSTNAME ] && host=local:
 601
 602         dir=`dirname $logfile`
 603         $VERY_VERBOSE && echo "Check pmie -h $host -l $logfile ..."
 604
 605         # make sure output directory exists
 606         #
 607         if [ ! -d "$dir" ]
 608         then
 609             mkdir -p -m 755 "$dir" >$tmp/err 2>&1
 610             if [ ! -d "$dir" ]
 611             then
 612                 cat $tmp/err
 613                 _error "cannot create directory ($dir) for pmie log file"
 614                 continue
 615             fi
 616         fi
 617
 618         # and the user pcp can write there
 619         #
 620         chown $PCP_USER:$PCP_GROUP "$dir" >/dev/null 2>&1
 621
 622         # and the logfile is writeable, if it exists
 623         #
 624         [ -f "$logfile" ] && chown $PCP_USER:$PCP_GROUP "$logfile" >/dev/null 2>&1
 625
 626         if cd "$dir"
 627         then
 628             :
 629         else
 630             _error "cannot chdir to directory ($dir) for pmie log file"
 631             continue
 632         fi
 633         dir=`$PWDCMND`
 634         $SHOWME && echo "+ cd $dir"
 635
 636         if [ ! -w "$dir" ]
 637         then
 638             _warning "no write access in $dir, skip lock file processing"
 639             ls -ld "$dir"
 640         else
 641             _lock
 642         fi
 643
 644         # match $logfile from control file to running pmies
 645         pid=""
 646         for pidfile in $PCP_TMP_DIR/pmie/[0-9]*
 647         do
 648             [ "$pidfile" = "$PCP_TMP_DIR/pmie/[0-9]*" ] && continue
 649             $VERY_VERBOSE && $PCP_ECHO_PROG $PCP_ECHO_N "... try $pidfile: ""$PCP_ECHO_C"
 650
 651             p_id=`echo $pidfile | sed -e 's,.*/,,'`
 652             p_logfile=""
 653             p_pmcd_host=""
 654
 655             # throw away stderr in case $pidfile has been removed by now
 656             eval `$PCP_BINADM_DIR/pmiestatus $pidfile 2>/dev/null | $PCP_AWK_PROG '
 657 NR == 2 { printf "p_logfile=\"%s\"\n", $0; next }
 658 NR == 3 { printf "p_pmcd_host=\"%s\"\n", $0; next }
 659         { next }'`
 660
 661             p_logfile=`_unsymlink_path $p_logfile`
 662             if [ "$p_logfile" != $logfile ]
 663             then
 664                 $VERY_VERBOSE && echo "different logfile, skip"
 665                 $VERY_VERBOSE && echo "  $p_logfile differs to $logfile"
 666             elif _get_pids_by_name pmie | grep "^$p_id\$" >/dev/null
 667             then
 668                 $VERY_VERBOSE && echo "pmie process $p_id identified, OK"
 669                 pid=$p_id
 670                 break
 671             else
 672                 $VERY_VERBOSE && echo "pmie process $p_id not running, skip"
 673                 $VERY_VERBOSE && _get_pids_by_name pmie
 674             fi
 675         done
 676
 677         if $VERY_VERBOSE
 678         then
 679             if [ -z "$pid" ]
 680             then
 681                 echo "No current pmie process exists for:"
 682             else
 683                 echo "Found pmie process $pid monitoring:"
 684             fi
 685             echo "    host = $host"
 686             echo "    log file = $logfile"
 687         fi
 688
 689         if [ -z "$pid" -a $START_PMIE = true ]
 690         then
 691             configfile=`_get_configfile $args`
 692             if [ ! -z "$configfile" ]
 693             then
 694                 # if this is a relative path and not relative to cwd,
 695                 # substitute in the default pmie search location.
 696                 #
 697                 if [ ! -f "$configfile" -a "`basename $configfile`" = "$configfile" ]
 698                 then
 699                     configfile="$PCP_VAR_DIR/config/pmie/$configfile"
 700                 fi
 701
 702                 # check configuration file exists and is up to date
 703                 _configure_pmie "$configfile" "$host" "$primary"
 704             fi
 705
 706             args="-h $host -l $logfile $args"
 707
 708             $VERBOSE && _restarting
 709
 710             sock_me=''
 711             if [ "$socks" = y ]
 712             then
 713                 # only check for pmsocks if it's specified in the control file
 714                 have_pmsocks=false
 715                 if which pmsocks >/dev/null 2>&1
 716                 then
 717                     # check if pmsocks has been set up correctly
 718                     if pmsocks ls >/dev/null 2>&1
 719                     then
 720                         have_pmsocks=true
 721                     fi
 722                 fi
 723
 724                 if $have_pmsocks
 725                 then
 726                     sock_me="pmsocks "
 727                 else
 728                     echo "$prog: Warning: no pmsocks available, would run without"
 729                     sock_me=""
 730                 fi
 731             fi
 732
 733             [ -f "$logfile" ] && eval $MV -f "$logfile" "$logfile.prior"
 734
 735             if $SHOWME
 736             then
 737                 $VERBOSE && echo
 738                 echo "+ ${sock_me}$PMIE -b $args"
 739                 _unlock
 740                 continue
 741             else
 742                 # since this is launched as a sort of daemon, any output should
 743                 # go on pmie's stderr, i.e. $logfile ... use -b for this
 744                 #
 745                 $VERY_VERBOSE && ( echo; $PCP_ECHO_PROG $PCP_ECHO_N "+ ${sock_me}$PMIE -b $args""$PCP_ECHO_C"; echo "..." )
 746                 $PCP_BINADM_DIR/pmpost "start pmie from $prog for host $host"
 747                 ${sock_me}$PMIE -b $args &
 748                 pid=$!
 749             fi
 750
 751             # wait for pmie to get started, and check on its health
 752             _check_pmie $pid
 753
 754         elif [ ! -z "$pid" -a $STOP_PMIE = true ]
 755         then
 756             # Send pmie a SIGTERM, which is noted as a pending shutdown.
 757             # Add pid to list of pmies sent SIGTERM - may need SIGKILL later.
 758             #
 759             $VERY_VERBOSE && echo "+ $KILL -s TERM $pid"
 760             eval $KILL -s TERM $pid
 761             $PCP_ECHO_PROG $PCP_ECHO_N "$pid ""$PCP_ECHO_C" >> $tmp/pmies
 762         fi
 763
 764         _unlock
 765     done
 766 }
 767
 768 _parse_control $CONTROL
 769 append=`ls $CONTROLDIR 2>/dev/null | LC_COLLATE=POSIX sort`
 770 for controlfile in $append
 771 do
 772     _parse_control $CONTROLDIR/$controlfile
 773 done
 774
 775 # check all the SIGTERM'd pmies really died - if not, use a bigger hammer.
 776 #
 777 if $SHOWME
 778 then
 779     :
 780 elif [ $STOP_PMIE = true -a -s $tmp/pmies ]
 781 then
 782     pmielist=`cat $tmp/pmies`
 783     if $PCP_PS_PROG -p "$pmielist" >/dev/null 2>&1
 784     then
 785         $VERY_VERBOSE && ( echo; $PCP_ECHO_PROG $PCP_ECHO_N "+ $KILL -KILL `cat $tmp/pmies` ...""$PCP_ECHO_C" )
 786         eval $KILL -s KILL $pmielist >/dev/null 2>&1
 787         delay=30        # tenths of a second
 788         while $PCP_PS_PROG -f -p "$pmielist" >$tmp/alive 2>&1
 789         do
 790             if [ $delay -gt 0 ]
 791             then
 792                 pmsleep 0.1
 793                 delay=`expr $delay - 1`
 794                 continue
 795             fi
 796             echo "$prog: Error: pmie process(es) will not die"
 797             cat $tmp/alive
 798             status=1
 799             break
 800         done
 801     fi
 802 fi
 803
 804 [ -f $tmp/err ] && status=1
 805 exit