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 # -----------------------------------------------------------------------------
  99 # parse parameters
 100
 101 debug=0
 102 check=0
 103 chart_only=
 104 while [ ! -z "$1" ]
 105 do
 106         if [ "$1" = "check" ]
 107         then
 108                 check=1
 109                 shift
 110                 continue
 111         fi
 112
 113         if [ "$1" = "debug" -o "$1" = "all" ]
 114         then
 115                 debug=1
 116                 shift
 117                 continue
 118         fi
 119
 120         if [ -f "$chartsd/$1.chart.sh" ]
 121         then
 122                 debug=1
 123                 chart_only="$( echo $1.chart.sh | sed "s/\.chart\.sh$//g" )"
 124                 shift
 125                 continue
 126         fi
 127
 128         if [ -f "$chartsd/$1" ]
 129         then
 130                 debug=1
 131                 chart_only="$( echo $1 | sed "s/\.chart\.sh$//g" )"
 132                 shift
 133                 continue
 134         fi
 135
 136         # number check
 137         n="$1"
 138         x=$(( n ))
 139         if [ "$x" = "$n" ]
 140         then
 141                 shift
 142                 update_every=$x
 143                 [ $update_every -lt $minimum_update_frequency ] && update_every=$minimum_update_frequency
 144                 continue
 145         fi
 146
 147         echo >&2 "Cannot understand parameter $1. Aborting."
 148         echo "DISABLE"
 149         exit 1
 150 done
 151
 152
 153 # -----------------------------------------------------------------------------
 154 # load my configuration
 155
 156 if [ -f "$myconfig" ]
 157         then
 158         . "$myconfig"
 159         if [ $? -ne 0 ]
 160         then
 161                 echo >&2 "$PROGRAM_NAME: cannot load $myconfig"
 162                 echo "DISABLE"
 163                 exit 1
 164         fi
 165         time_divisor=$((time_divisor))
 166         [ $time_divisor -lt 10 ] && time_divisor=10
 167         [ $time_divisor -gt 100 ] && time_divisor=100
 168 else
 169         echo >&2 "$PROGRAM_NAME: configuration file '$myconfig' not found. Using defaults."
 170 fi
 171
 172 if [ "$pause_method" = "suspend" ]
 173 then
 174         # enable bash job control
 175         # this is required for suspend to work
 176         set -m
 177 fi
 178
 179 # we check for the timeout command, after we load our
 180 # configuration, so that the user may overwrite the
 181 # timeout command we use, providing a function that
 182 # can emulate the timeout command we need:
 183 # > timeout SECONDS command ...
 184 if [ $check_for_timeout -eq 1 ]
 185         then
 186         require_cmd timeout || exit 1
 187 fi
 188
 189 # -----------------------------------------------------------------------------
 190 # internal checks
 191
 192 # netdata passes the requested update frequency as the first argument
 193 update_every=$(( update_every + 1 - 1)) # makes sure it is a number
 194 test $update_every -eq 0 && update_every=1 # if it is zero, make it 1
 195
 196 # check the charts.d directory
 197 if [ ! -d "$chartsd" ]
 198         then
 199         echo >&2 "$PROGRAM_NAME: cannot find charts directory '$chartsd'"
 200         echo "DISABLE"
 201 fi
 202
 203
 204 # -----------------------------------------------------------------------------
 205 # loop control
 206
 207 # default sleep function
 208 LOOPSLEEPMS_HIGHRES=0
 209 loopsleepms() {
 210         [ "$1" = "tellwork" ] && shift
 211         sleep $1
 212 }
 213
 214 now_ms=
 215 current_time_ms() {
 216         now_ms="$(date +'%s')000"
 217 }
 218
 219 # if found and included, this file overwrites loopsleepms()
 220 # and current_time_ms() with a high resolution timer function
 221 # for precise looping.
 222 . "$pluginsd/loopsleepms.sh.inc"
 223
 224
 225 # -----------------------------------------------------------------------------
 226 # library functions
 227
 228 fixid() {
 229         echo "$*" |\
 230                 tr -c "[A-Z][a-z][0-9]" "_" |\
 231                 sed -e "s|^_\+||g" -e "s|_\+$||g" -e "s|_\+|_|g" |\
 232                 tr "[A-Z]" "[a-z]"
 233 }
 234
 235
 236 # -----------------------------------------------------------------------------
 237 # charts check functions
 238
 239 all_charts() {
 240         cd "$chartsd"
 241         [ $? -ne 0 ] && echo >&2 "$PROGRAM_NAME: Cannot cd to $chartsd" && return 1
 242
 243         ls *.chart.sh | sed "s/\.chart\.sh$//g"
 244 }
 245
 246 all_enabled_charts() {
 247         local charts=
 248
 249         # find all enabled charts
 250
 251         for chart in $( all_charts )
 252         do
 253                 eval "enabled=\$$chart"
 254                 if [ "$enabled" = "no" ]
 255                 then
 256                         echo >&2 "$PROGRAM_NAME: '$chart' is NOT enabled. Add a line with $chart=yes in $myconfig to enable it (or remove the line that disables it)."
 257                 else
 258                         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' is enabled."
 259                         local charts="$charts $chart"
 260                 fi
 261         done
 262
 263         local charts2=
 264         for chart in $charts
 265         do
 266                 # check the enabled charts
 267                 local check="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_check()" )"
 268                 if [ -z "$check" ]
 269                 then
 270                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_check() function. Disabling it."
 271                         continue
 272                 fi
 273
 274                 local create="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_create()" )"
 275                 if [ -z "$create" ]
 276                 then
 277                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_create() function. Disabling it."
 278                         continue
 279                 fi
 280
 281                 local update="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_update()" )"
 282                 if [ -z "$update" ]
 283                 then
 284                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_update() function. Disabling it."
 285                         continue
 286                 fi
 287
 288                 # check its config
 289                 #if [ -f "$confd/$chart.conf" ]
 290                 #then
 291                 #       if [ ! -z "$( cat "$confd/$chart.conf" | sed "s/^ \+//g" | grep -v "^$" | grep -v "^#" | grep -v "^$chart$charts_undescore" )" ]
 292                 #       then
 293                 #               echo >&2 "$PROGRAM_NAME: chart's $chart config $confd/$chart.conf should only have lines starting with $chart$charts_undescore . Disabling it."
 294                 #               continue
 295                 #       fi
 296                 #fi
 297
 298                 #if [ $dryrunner -eq 1 ]
 299                 #       then
 300                 #       "$pluginsd/charts.d.dryrun-helper.sh" "$chart" "$chartsd/$chart.chart.sh" "$confd/$chart.conf" >/dev/null
 301                 #       if [ $? -ne 0 ]
 302                 #       then
 303                 #               echo >&2 "$PROGRAM_NAME: chart's $chart did not pass the dry run check. This means it uses global variables not starting with $chart. Disabling it."
 304                 #               continue
 305                 #       fi
 306                 #fi
 307
 308                 local charts2="$charts2 $chart"
 309         done
 310
 311         echo $charts2
 312         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: enabled charts: $charts2"
 313 }
 314
 315
 316 # -----------------------------------------------------------------------------
 317 # load the charts
 318
 319 suffix_update_every="_update_every"
 320 active_charts=
 321 for chart in $( all_enabled_charts )
 322 do
 323         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: loading chart: '$chartsd/$chart.chart.sh'"
 324         . "$chartsd/$chart.chart.sh"
 325
 326         if [ -f "$confd/$chart.conf" ]
 327         then
 328                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: loading chart options: '$confd/$chart.conf'"
 329                 . "$confd/$chart.conf"
 330         else
 331                 echo >&2 "$PROGRAM_NAME: $chart: configuration file '$confd/$chart.conf' not found. Using defaults."
 332         fi
 333
 334         eval "dt=\$$chart$suffix_update_every"
 335         dt=$(( dt + 1 - 1 )) # make sure it is a number
 336         if [ $dt -lt $update_every ]
 337         then
 338                 eval "$chart$suffix_update_every=$update_every"
 339         fi
 340
 341         $chart$charts_check
 342         if [ $? -eq 0 ]
 343         then
 344                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' activated"
 345                 active_charts="$active_charts $chart"
 346         else
 347                 echo >&2 "$PROGRAM_NAME: chart '$chart' check() function reports failure."
 348         fi
 349 done
 350 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: activated charts: $active_charts"
 351
 352
 353 # -----------------------------------------------------------------------------
 354 # check overwrites
 355
 356 # enable work time reporting
 357 debug_time=
 358 test $debug -eq 1 && debug_time=tellwork
 359
 360 # if we only need a specific chart, remove all the others
 361 if [ ! -z "${chart_only}" ]
 362 then
 363         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: requested to run only for: '${chart_only}'"
 364         check_charts=
 365         for chart in $active_charts
 366         do
 367                 if [ "$chart" = "$chart_only" ]
 368                 then
 369                         check_charts="$chart"
 370                         break
 371                 fi
 372         done
 373         active_charts="$check_charts"
 374 fi
 375 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: activated charts: $active_charts"
 376
 377 # stop if we just need a pre-check
 378 if [ $check -eq 1 ]
 379 then
 380         echo >&2 "CHECK RESULT"
 381         echo >&2 "Will run the charts: $active_charts"
 382         exit 0
 383 fi
 384
 385 # -----------------------------------------------------------------------------
 386 # create temp dir
 387
 388 TMP_DIR=
 389 chartsd_cleanup() {
 390         cd /tmp
 391         if [ ! -z "$TMP_DIR" -a -d "$TMP_DIR" ]
 392         then
 393                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: cleaning up temporary directory $TMP_DIR ..."
 394                 rm -rf "$TMP_DIR"
 395         fi
 396         exit 0
 397 }
 398 trap chartsd_cleanup EXIT
 399 trap chartsd_cleanup SIGHUP
 400 trap chartsd_cleanup INT
 401
 402 if [ $UID = "0" ]
 403 then
 404         TMP_DIR="$( mktemp -d /var/run/netdata-${PROGRAM_NAME}-XXXXXXXXXX )"
 405 else
 406         TMP_DIR="$( mktemp -d /tmp/.netdata-${PROGRAM_NAME}-XXXXXXXXXX )"
 407 fi
 408
 409 cd "$TMP_DIR" || exit 1
 410
 411 # -----------------------------------------------------------------------------
 412 # create charts
 413
 414 run_charts=
 415 for chart in $active_charts
 416 do
 417         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: Calling '$chart$charts_create()'..."
 418         $chart$charts_create
 419         if [ $? -eq 0 ]
 420         then
 421                 run_charts="$run_charts $chart"
 422                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' has initialized."
 423         else
 424                 echo >&2 "$PROGRAM_NAME: chart '$chart' function '$chart$charts_create()' reports failure."
 425         fi
 426 done
 427 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: run_charts='$run_charts'"
 428
 429
 430 # -----------------------------------------------------------------------------
 431 # update dimensions
 432
 433 if [ -z "$run_charts" ]
 434         then
 435         echo >&2 "$PROGRAM_NAME: No charts to collect data from."
 436         echo "DISABLE"
 437         exit 1
 438 fi
 439
 440 declare -A charts_last_update=() charts_update_every=() charts_next_update=() chart_run_counter=()
 441 global_update() {
 442         local exit_at \
 443                 c=0 dt ret exec_start_ms exec_end_ms \
 444                 chart now_charts=() next_charts=($run_charts)
 445
 446         # return the current time in ms in $now_ms
 447         current_time_ms
 448
 449         exit_at=$(( now_ms + (restart_timeout * 1000) ))
 450
 451         for chart in $run_charts
 452         do
 453                 eval "charts_update_every[$chart]=\$$chart$suffix_update_every"
 454                 test -z "${charts_update_every[$chart]}" && charts_update_every[$charts]=$update_every
 455                 charts_last_update[$chart]=$((now_ms - (now_ms % (charts_update_every[$chart] * 1000) ) ))
 456                 charts_next_update[$chart]=$(( charts_last_update[$chart] + (charts_update_every[$chart] * 1000) ))
 457                 charts_run_counter[$chart]=0
 458
 459                 echo "CHART netdata.plugin_$chart '' 'Execution time for $chart plugin' 'milliseconds / run' netdata netdata area 90000 ${charts_update_every[$chart]}"
 460                 echo "DIMENSION run_time 'run time' absolute 1 1"
 461         done
 462
 463         # the main loop
 464         while [ 1 ]
 465         do
 466                 c=$((c + 1))
 467                 now_charts=("${next_charts[@]}")
 468                 next_charts=()
 469
 470                 # return the current time in ms in $now_ms
 471                 current_time_ms
 472
 473                 for chart in "${now_charts[@]}"
 474                 do
 475                         # echo >&2 "DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}"
 476                         if [ ${now_ms} -ge ${charts_next_update[$chart]} ]
 477                         then
 478                                 dt=$(( (now_ms - charts_last_update[$chart]) ))
 479                                 #echo >&2 "DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}, dt: ${dt}"
 480
 481                                 charts_last_update[$chart]=$now_ms
 482
 483                                 while [ ${charts_next_update[$chart]} -lt ${now_ms} ]
 484                                 do
 485                                         charts_next_update[$chart]=$(( charts_next_update[$chart] + (charts_update_every[$chart] * 1000) ))
 486                                 done
 487
 488                                 # the first call should not give a duration
 489                                 # so that netdata calibrates to current time
 490                                 charts_run_counter[$chart]=$(( charts_run_counter[$chart] + 1 ))
 491                                 test ${charts_run_counter[$chart]} -eq 1 && dt=
 492                                 dt=$(( dt * 1000 ))
 493
 494                                 exec_start_ms=$now_ms
 495                                 $chart$charts_update $dt
 496                                 ret=$?
 497
 498                                 # return the current time in ms in $now_ms
 499                                 current_time_ms; exec_end_ms=$now_ms
 500
 501                                 echo "BEGIN netdata.plugin_$chart $dt"
 502                                 if [ $ret -eq 0 ]
 503                                 then
 504                                         echo "SET run_time = $(( exec_end_ms - exec_start_ms ))"
 505                                         next_charts+=($chart)
 506                                 else
 507                                         echo "SET run_time = $(( (exec_end_ms - exec_start_ms) * -1 ))"
 508                                         echo >&2 "$PROGRAM_NAME: chart '$chart' update() function reports failure. Disabling it."
 509                                 fi
 510                                 echo "END"
 511                         else
 512                                 next_charts+=($chart)
 513                         fi
 514                 done
 515
 516                 if [ "$pause_method" = "suspend" ]
 517                 then
 518                         echo "STOPPING_WAKE_ME_UP_PLEASE"
 519                         suspend || ( echo >&2 "$PROGRAM_NAME: suspend returned error $?, falling back to sleep."; loopsleepms $debug_time $update_every $time_divisor)
 520                 else
 521                         # wait the time you are required to
 522                         #loopsleepms $debug_time $update_every $time_divisor
 523                         if [ ${LOOPSLEEPMS_HIGHRES} -eq 1 ]
 524                                 then
 525                                 sleep 0.2
 526                         else
 527                                 sleep 1
 528                         fi
 529                 fi
 530
 531                 test ${now_ms} -ge ${exit_at} && exit 0
 532         done
 533 }
 534
 535 global_update