plugins.d/charts.d.plugin

   1 #!/bin/bash
   2
   3 PROGRAM_FILE="$0"
   4 PROGRAM_NAME="$(basename $0)"
   5 PROGRAM_NAME="${PROGRAM_NAME/.plugin}"
   6
   7 # if you need to run parallel charts.d processes
   8 # just link this files with a different name
   9 # in the same directory, with a .plugin suffix
  10 # netdata will start multiple of them
  11 # each will have a different config file
  12
  13 echo >&2 "$PROGRAM_NAME: started from '$PROGRAM_FILE' with options: $*"
  14
  15 if [ $(( ${BASH_VERSINFO[0]} )) -lt 4 ]
  16 then
  17         echo >&2
  18         echo >&2 "$PROGRAM_NAME: ERROR"
  19         echo >&2 "BASH version 4 or later is required."
  20         echo >&2 "You are running version: ${BASH_VERSION}"
  21         echo >&2 "Please upgrade."
  22         echo >&2
  23         exit 1
  24 fi
  25
  26 # check a few commands
  27 require_cmd() {
  28         which "$1" >/dev/null
  29         if [ $? -ne 0 ]
  30                 then
  31                 echo >&2 "$PROGRAM_NAME: ERROR: Command '$1' is not found in the system path."
  32                 return 1
  33         fi
  34         return 0
  35 }
  36
  37 require_cmd date || exit 1
  38 require_cmd sed || exit 1
  39 require_cmd basename || exit 1
  40 require_cmd dirname || exit 1
  41 require_cmd cat || exit 1
  42 require_cmd grep || exit 1
  43 require_cmd egrep || exit 1
  44 require_cmd mktemp || exit 1
  45 require_cmd awk || exit 1
  46
  47 # -----------------------------------------------------------------------------
  48 # insternal defaults
  49 # netdata exposes a few environment variables for us
  50
  51 pause_method="sleep" # use either "suspend" or "sleep"
  52                      # DO NOT USE SUSPEND - LINUX WILL SUSPEND NETDATA TOO
  53                      # THE WHOLE PROCESS GROUP - NOT JUST THE SHELL
  54
  55 pluginsd="${NETDATA_PLUGINS_DIR}"
  56 [ -z "$pluginsd" ] && pluginsd="$( dirname $PROGRAM_FILE )"
  57
  58 confd="${NETDATA_CONFIG_DIR-/etc/netdata}"
  59 chartsd="$pluginsd/../charts.d"
  60
  61 myconfig="$confd/$PROGRAM_NAME.conf"
  62
  63 minimum_update_frequency="${NETDATA_UPDATE_EVERY-1}"
  64 update_every=${minimum_update_frequency}        # this will be overwritten by the command line
  65
  66 # work around for non BASH shells
  67 charts_create="_create"
  68 charts_update="_update"
  69 charts_check="_check"
  70 charts_undescore="_"
  71
  72 # when making iterations, charts.d can loop more frequently
  73 # to prevent plugins missing iterations.
  74 # this is a percentage relative to update_every to align its
  75 # iterations.
  76 # The minimum is 10%, the maximum 100%.
  77 # So, if update_every is 1 second and time_divisor is 50,
  78 # charts.d will iterate every 500ms.
  79 # Charts will be called to collect data only if the time
  80 # passes since the last time the collected data is equal or
  81 # above their update_every.
  82 time_divisor=50
  83
  84 # number of seconds to run without restart
  85 # after this time, charts.d.plugin will exit
  86 # netdata will restart it
  87 restart_timeout=1800
  88
  89 # check if the charts.d plugins are using global variables
  90 # they should not.
  91 # It does not currently support BASH v4 arrays, so it is
  92 # disabled
  93 dryrunner=0
  94
  95 # check for timeout command
  96 check_for_timeout=1
  97
  98 # the default enable/disable value for all charts
  99 enable_all_charts="yes"
 100
 101 # -----------------------------------------------------------------------------
 102 # parse parameters
 103
 104 debug=0
 105 check=0
 106 chart_only=
 107 while [ ! -z "$1" ]
 108 do
 109         if [ "$1" = "check" ]
 110         then
 111                 check=1
 112                 shift
 113                 continue
 114         fi
 115
 116         if [ "$1" = "debug" -o "$1" = "all" ]
 117         then
 118                 debug=1
 119                 shift
 120                 continue
 121         fi
 122
 123         if [ -f "$chartsd/$1.chart.sh" ]
 124         then
 125                 debug=1
 126                 chart_only="$( echo $1.chart.sh | sed "s/\.chart\.sh$//g" )"
 127                 shift
 128                 continue
 129         fi
 130
 131         if [ -f "$chartsd/$1" ]
 132         then
 133                 debug=1
 134                 chart_only="$( echo $1 | sed "s/\.chart\.sh$//g" )"
 135                 shift
 136                 continue
 137         fi
 138
 139         # number check
 140         n="$1"
 141         x=$(( n ))
 142         if [ "$x" = "$n" ]
 143         then
 144                 shift
 145                 update_every=$x
 146                 [ $update_every -lt $minimum_update_frequency ] && update_every=$minimum_update_frequency
 147                 continue
 148         fi
 149
 150         echo >&2 "Cannot understand parameter $1. Aborting."
 151         echo "DISABLE"
 152         exit 1
 153 done
 154
 155
 156 # -----------------------------------------------------------------------------
 157 # load my configuration
 158
 159 if [ -f "$myconfig" ]
 160         then
 161         . "$myconfig"
 162         if [ $? -ne 0 ]
 163         then
 164                 echo >&2 "$PROGRAM_NAME: cannot load $myconfig"
 165                 echo "DISABLE"
 166                 exit 1
 167         fi
 168         time_divisor=$((time_divisor))
 169         [ $time_divisor -lt 10 ] && time_divisor=10
 170         [ $time_divisor -gt 100 ] && time_divisor=100
 171 else
 172         echo >&2 "$PROGRAM_NAME: configuration file '$myconfig' not found. Using defaults."
 173 fi
 174
 175 if [ "$pause_method" = "suspend" ]
 176 then
 177         # enable bash job control
 178         # this is required for suspend to work
 179         set -m
 180 fi
 181
 182 # we check for the timeout command, after we load our
 183 # configuration, so that the user may overwrite the
 184 # timeout command we use, providing a function that
 185 # can emulate the timeout command we need:
 186 # > timeout SECONDS command ...
 187 if [ $check_for_timeout -eq 1 ]
 188         then
 189         require_cmd timeout || exit 1
 190 fi
 191
 192 # -----------------------------------------------------------------------------
 193 # internal checks
 194
 195 # netdata passes the requested update frequency as the first argument
 196 update_every=$(( update_every + 1 - 1)) # makes sure it is a number
 197 test $update_every -eq 0 && update_every=1 # if it is zero, make it 1
 198
 199 # check the charts.d directory
 200 if [ ! -d "$chartsd" ]
 201         then
 202         echo >&2 "$PROGRAM_NAME: cannot find charts directory '$chartsd'"
 203         echo "DISABLE"
 204 fi
 205
 206
 207 # -----------------------------------------------------------------------------
 208 # loop control
 209
 210 # default sleep function
 211 LOOPSLEEPMS_HIGHRES=0
 212 loopsleepms() {
 213         [ "$1" = "tellwork" ] && shift
 214         sleep $1
 215 }
 216
 217 now_ms=
 218 current_time_ms() {
 219         now_ms="$(date +'%s')000"
 220 }
 221
 222 # if found and included, this file overwrites loopsleepms()
 223 # and current_time_ms() with a high resolution timer function
 224 # for precise looping.
 225 . "$pluginsd/loopsleepms.sh.inc"
 226
 227
 228 # -----------------------------------------------------------------------------
 229 # library functions
 230
 231 fixid() {
 232         echo "$*" |\
 233                 tr -c "[A-Z][a-z][0-9]" "_" |\
 234                 sed -e "s|^_\+||g" -e "s|_\+$||g" -e "s|_\+|_|g" |\
 235                 tr "[A-Z]" "[a-z]"
 236 }
 237
 238
 239 # -----------------------------------------------------------------------------
 240 # charts check functions
 241
 242 all_charts() {
 243         cd "$chartsd"
 244         [ $? -ne 0 ] && echo >&2 "$PROGRAM_NAME: Cannot cd to $chartsd" && return 1
 245
 246         ls *.chart.sh | sed "s/\.chart\.sh$//g"
 247 }
 248
 249 all_enabled_charts() {
 250         local charts= enabled=
 251
 252         # find all enabled charts
 253
 254         for chart in $( all_charts )
 255         do
 256                 eval "enabled=\$$chart"
 257                 if [ -z "${enabled}" ]
 258                         then
 259                         enabled="${enable_all_charts}"
 260                 fi
 261
 262                 if [ ! "${enabled}" = "yes" ]
 263                 then
 264                         echo >&2 "$PROGRAM_NAME: '$chart' is NOT enabled. Add a line with $chart=yes in $myconfig to enable it (or remove the line that disables it)."
 265                 else
 266                         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' is enabled."
 267                         local charts="$charts $chart"
 268                 fi
 269         done
 270
 271         local charts2=
 272         for chart in $charts
 273         do
 274                 # check the enabled charts
 275                 local check="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_check()" )"
 276                 if [ -z "$check" ]
 277                 then
 278                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_check() function. Disabling it."
 279                         continue
 280                 fi
 281
 282                 local create="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_create()" )"
 283                 if [ -z "$create" ]
 284                 then
 285                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_create() function. Disabling it."
 286                         continue
 287                 fi
 288
 289                 local update="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_update()" )"
 290                 if [ -z "$update" ]
 291                 then
 292                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_update() function. Disabling it."
 293                         continue
 294                 fi
 295
 296                 # check its config
 297                 #if [ -f "$confd/$chart.conf" ]
 298                 #then
 299                 #       if [ ! -z "$( cat "$confd/$chart.conf" | sed "s/^ \+//g" | grep -v "^$" | grep -v "^#" | grep -v "^$chart$charts_undescore" )" ]
 300                 #       then
 301                 #               echo >&2 "$PROGRAM_NAME: chart's $chart config $confd/$chart.conf should only have lines starting with $chart$charts_undescore . Disabling it."
 302                 #               continue
 303                 #       fi
 304                 #fi
 305
 306                 #if [ $dryrunner -eq 1 ]
 307                 #       then
 308                 #       "$pluginsd/charts.d.dryrun-helper.sh" "$chart" "$chartsd/$chart.chart.sh" "$confd/$chart.conf" >/dev/null
 309                 #       if [ $? -ne 0 ]
 310                 #       then
 311                 #               echo >&2 "$PROGRAM_NAME: chart's $chart did not pass the dry run check. This means it uses global variables not starting with $chart. Disabling it."
 312                 #               continue
 313                 #       fi
 314                 #fi
 315
 316                 local charts2="$charts2 $chart"
 317         done
 318
 319         echo $charts2
 320         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: enabled charts: $charts2"
 321 }
 322
 323
 324 # -----------------------------------------------------------------------------
 325 # load the charts
 326
 327 suffix_update_every="_update_every"
 328 active_charts=
 329 for chart in $( all_enabled_charts )
 330 do
 331         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: loading chart: '$chartsd/$chart.chart.sh'"
 332         . "$chartsd/$chart.chart.sh"
 333
 334         if [ -f "$confd/$chart.conf" ]
 335         then
 336                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: loading chart options: '$confd/$chart.conf'"
 337                 . "$confd/$chart.conf"
 338         else
 339                 echo >&2 "$PROGRAM_NAME: $chart: configuration file '$confd/$chart.conf' not found. Using defaults."
 340         fi
 341
 342         eval "dt=\$$chart$suffix_update_every"
 343         dt=$(( dt + 1 - 1 )) # make sure it is a number
 344         if [ $dt -lt $update_every ]
 345         then
 346                 eval "$chart$suffix_update_every=$update_every"
 347         fi
 348
 349         $chart$charts_check
 350         if [ $? -eq 0 ]
 351         then
 352                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' activated"
 353                 active_charts="$active_charts $chart"
 354         else
 355                 echo >&2 "$PROGRAM_NAME: chart '$chart' check() function reports failure."
 356         fi
 357 done
 358 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: activated charts: $active_charts"
 359
 360
 361 # -----------------------------------------------------------------------------
 362 # check overwrites
 363
 364 # enable work time reporting
 365 debug_time=
 366 test $debug -eq 1 && debug_time=tellwork
 367
 368 # if we only need a specific chart, remove all the others
 369 if [ ! -z "${chart_only}" ]
 370 then
 371         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: requested to run only for: '${chart_only}'"
 372         check_charts=
 373         for chart in $active_charts
 374         do
 375                 if [ "$chart" = "$chart_only" ]
 376                 then
 377                         check_charts="$chart"
 378                         break
 379                 fi
 380         done
 381         active_charts="$check_charts"
 382 fi
 383 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: activated charts: $active_charts"
 384
 385 # stop if we just need a pre-check
 386 if [ $check -eq 1 ]
 387 then
 388         echo >&2 "CHECK RESULT"
 389         echo >&2 "Will run the charts: $active_charts"
 390         exit 0
 391 fi
 392
 393 # -----------------------------------------------------------------------------
 394 # create temp dir
 395
 396 TMP_DIR=
 397 chartsd_cleanup() {
 398         cd /tmp
 399         if [ ! -z "$TMP_DIR" -a -d "$TMP_DIR" ]
 400         then
 401                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: cleaning up temporary directory $TMP_DIR ..."
 402                 rm -rf "$TMP_DIR"
 403         fi
 404         exit 0
 405 }
 406 trap chartsd_cleanup EXIT
 407 trap chartsd_cleanup SIGHUP
 408 trap chartsd_cleanup INT
 409
 410 if [ $UID = "0" ]
 411 then
 412         TMP_DIR="$( mktemp -d /var/run/netdata-${PROGRAM_NAME}-XXXXXXXXXX )"
 413 else
 414         TMP_DIR="$( mktemp -d /tmp/.netdata-${PROGRAM_NAME}-XXXXXXXXXX )"
 415 fi
 416
 417 cd "$TMP_DIR" || exit 1
 418
 419 # -----------------------------------------------------------------------------
 420 # create charts
 421
 422 run_charts=
 423 for chart in $active_charts
 424 do
 425         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: Calling '$chart$charts_create()'..."
 426         $chart$charts_create
 427         if [ $? -eq 0 ]
 428         then
 429                 run_charts="$run_charts $chart"
 430                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' has initialized."
 431         else
 432                 echo >&2 "$PROGRAM_NAME: chart '$chart' function '$chart$charts_create()' reports failure."
 433         fi
 434 done
 435 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: run_charts='$run_charts'"
 436
 437
 438 # -----------------------------------------------------------------------------
 439 # update dimensions
 440
 441 if [ -z "$run_charts" ]
 442         then
 443         echo >&2 "$PROGRAM_NAME: No charts to collect data from."
 444         echo "DISABLE"
 445         exit 1
 446 fi
 447
 448 declare -A charts_last_update=() charts_update_every=() charts_next_update=() chart_run_counter=()
 449 global_update() {
 450         local exit_at \
 451                 c=0 dt ret exec_start_ms exec_end_ms \
 452                 chart now_charts=() next_charts=($run_charts)
 453
 454         # return the current time in ms in $now_ms
 455         current_time_ms
 456
 457         exit_at=$(( now_ms + (restart_timeout * 1000) ))
 458
 459         for chart in $run_charts
 460         do
 461                 eval "charts_update_every[$chart]=\$$chart$suffix_update_every"
 462                 test -z "${charts_update_every[$chart]}" && charts_update_every[$charts]=$update_every
 463                 charts_last_update[$chart]=$((now_ms - (now_ms % (charts_update_every[$chart] * 1000) ) ))
 464                 charts_next_update[$chart]=$(( charts_last_update[$chart] + (charts_update_every[$chart] * 1000) ))
 465                 charts_run_counter[$chart]=0
 466
 467                 echo "CHART netdata.plugin_$chart '' 'Execution time for $chart plugin' 'milliseconds / run' netdata netdata area 90000 ${charts_update_every[$chart]}"
 468                 echo "DIMENSION run_time 'run time' absolute 1 1"
 469         done
 470
 471         # the main loop
 472         while [ 1 ]
 473         do
 474                 c=$((c + 1))
 475                 now_charts=("${next_charts[@]}")
 476                 next_charts=()
 477
 478                 # return the current time in ms in $now_ms
 479                 current_time_ms
 480
 481                 for chart in "${now_charts[@]}"
 482                 do
 483                         # echo >&2 "DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}"
 484                         if [ ${now_ms} -ge ${charts_next_update[$chart]} ]
 485                         then
 486                                 dt=$(( (now_ms - charts_last_update[$chart]) ))
 487                                 #echo >&2 "DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}, dt: ${dt}"
 488
 489                                 charts_last_update[$chart]=$now_ms
 490
 491                                 while [ ${charts_next_update[$chart]} -lt ${now_ms} ]
 492                                 do
 493                                         charts_next_update[$chart]=$(( charts_next_update[$chart] + (charts_update_every[$chart] * 1000) ))
 494                                 done
 495
 496                                 # the first call should not give a duration
 497                                 # so that netdata calibrates to current time
 498                                 charts_run_counter[$chart]=$(( charts_run_counter[$chart] + 1 ))
 499                                 dt=$(( dt * 1000 ))
 500                                 test ${charts_run_counter[$chart]} -eq 1 && dt=
 501
 502                                 exec_start_ms=$now_ms
 503                                 $chart$charts_update $dt
 504                                 ret=$?
 505
 506                                 # return the current time in ms in $now_ms
 507                                 current_time_ms; exec_end_ms=$now_ms
 508
 509                                 echo "BEGIN netdata.plugin_$chart $dt"
 510                                 if [ $ret -eq 0 ]
 511                                 then
 512                                         echo "SET run_time = $(( exec_end_ms - exec_start_ms ))"
 513                                         next_charts+=($chart)
 514                                 else
 515                                         echo "SET run_time = $(( (exec_end_ms - exec_start_ms) * -1 ))"
 516                                         echo >&2 "$PROGRAM_NAME: chart '$chart' update() function reports failure. Disabling it."
 517                                 fi
 518                                 echo "END"
 519                         else
 520                                 next_charts+=($chart)
 521                         fi
 522                 done
 523
 524                 if [ "$pause_method" = "suspend" ]
 525                 then
 526                         echo "STOPPING_WAKE_ME_UP_PLEASE"
 527                         suspend || ( echo >&2 "$PROGRAM_NAME: suspend returned error $?, falling back to sleep."; loopsleepms $debug_time $update_every $time_divisor)
 528                 else
 529                         # wait the time you are required to
 530                         #loopsleepms $debug_time $update_every $time_divisor
 531                         if [ ${LOOPSLEEPMS_HIGHRES} -eq 1 ]
 532                                 then
 533                                 sleep 0.2
 534                         else
 535                                 sleep 1
 536                         fi
 537                 fi
 538
 539                 test ${now_ms} -ge ${exit_at} && exit 0
 540         done
 541 }
 542
 543 global_update