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 # convert any floating point number
 239 # to integer, give a multiplier
 240 # the result is stored in ${FLOAT2INT_RESULT}
 241 # so that no fork is necessary
 242 # the multiplier must be a power of 10
 243 float2int() {
 244         local f="$1" m="$2" a b l
 245
 246         #echo >&2 "f='${f}', m='${m}'"
 247
 248         # the length of the multiplier - 1
 249         l=$[ ${#m} - 1 ]
 250
 251         # split the floating point number
 252         # in integer (a) and decimal (b)
 253         a=${f/.*/}
 254         b=${f/*./}
 255
 256         # prepend a zero to the integer part
 257         # if it is missing
 258         [ -z "${a}" ] && a="0"
 259         # strip leading zeros - base 10 convertion
 260         a=$[10#$a]
 261
 262         # check the length of the decimal part
 263         # against the length of the multiplier
 264         if [ ${#b} -gt ${l} ]
 265                 then
 266                 # too many digits - take the most significant
 267                 b=${b:0:${l}}
 268         elif [ ${#b} -lt ${l} ]
 269                 then
 270                 # too few digits - pad with zero on the right
 271                 local z="00000000000000000000000" r=$[l - ${#b}]
 272                 b="${b}${z:0:${r}}"
 273         fi
 274         # strip leading zeros - base 10 convertion
 275         b=$[10#$b]
 276
 277         # store the result
 278         FLOAT2INT_RESULT=$[ (a * m) + b ]
 279         #echo >&2 "FLOAT2INT_RESULT='${FLOAT2INT_RESULT}'"
 280 }
 281
 282
 283 # -----------------------------------------------------------------------------
 284 # charts check functions
 285
 286 all_charts() {
 287         cd "$chartsd"
 288         [ $? -ne 0 ] && echo >&2 "$PROGRAM_NAME: Cannot cd to $chartsd" && return 1
 289
 290         ls *.chart.sh | sed "s/\.chart\.sh$//g"
 291 }
 292
 293 all_enabled_charts() {
 294         local charts= enabled=
 295
 296         # find all enabled charts
 297
 298         for chart in $( all_charts )
 299         do
 300                 eval "enabled=\$$chart"
 301                 if [ -z "${enabled}" ]
 302                         then
 303                         enabled="${enable_all_charts}"
 304                 fi
 305
 306                 if [ ! "${enabled}" = "yes" ]
 307                 then
 308                         echo >&2 "$PROGRAM_NAME: '$chart' is NOT enabled. Add a line with $chart=yes in $myconfig to enable it (or remove the line that disables it)."
 309                 else
 310                         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' is enabled."
 311                         local charts="$charts $chart"
 312                 fi
 313         done
 314
 315         local charts2=
 316         for chart in $charts
 317         do
 318                 # check the enabled charts
 319                 local check="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_check()" )"
 320                 if [ -z "$check" ]
 321                 then
 322                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_check() function. Disabling it."
 323                         continue
 324                 fi
 325
 326                 local create="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_create()" )"
 327                 if [ -z "$create" ]
 328                 then
 329                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_create() function. Disabling it."
 330                         continue
 331                 fi
 332
 333                 local update="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_update()" )"
 334                 if [ -z "$update" ]
 335                 then
 336                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_update() function. Disabling it."
 337                         continue
 338                 fi
 339
 340                 # check its config
 341                 #if [ -f "$confd/$chart.conf" ]
 342                 #then
 343                 #       if [ ! -z "$( cat "$confd/$chart.conf" | sed "s/^ \+//g" | grep -v "^$" | grep -v "^#" | grep -v "^$chart$charts_undescore" )" ]
 344                 #       then
 345                 #               echo >&2 "$PROGRAM_NAME: chart's $chart config $confd/$chart.conf should only have lines starting with $chart$charts_undescore . Disabling it."
 346                 #               continue
 347                 #       fi
 348                 #fi
 349
 350                 #if [ $dryrunner -eq 1 ]
 351                 #       then
 352                 #       "$pluginsd/charts.d.dryrun-helper.sh" "$chart" "$chartsd/$chart.chart.sh" "$confd/$chart.conf" >/dev/null
 353                 #       if [ $? -ne 0 ]
 354                 #       then
 355                 #               echo >&2 "$PROGRAM_NAME: chart's $chart did not pass the dry run check. This means it uses global variables not starting with $chart. Disabling it."
 356                 #               continue
 357                 #       fi
 358                 #fi
 359
 360                 local charts2="$charts2 $chart"
 361         done
 362
 363         echo $charts2
 364         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: enabled charts: $charts2"
 365 }
 366
 367
 368 # -----------------------------------------------------------------------------
 369 # load the charts
 370
 371 suffix_update_every="_update_every"
 372 active_charts=
 373 for chart in $( all_enabled_charts )
 374 do
 375         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: loading chart: '$chartsd/$chart.chart.sh'"
 376         . "$chartsd/$chart.chart.sh"
 377
 378         if [ -f "$confd/$chart.conf" ]
 379         then
 380                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: loading chart options: '$confd/$chart.conf'"
 381                 . "$confd/$chart.conf"
 382         else
 383                 echo >&2 "$PROGRAM_NAME: $chart: configuration file '$confd/$chart.conf' not found. Using defaults."
 384         fi
 385
 386         eval "dt=\$$chart$suffix_update_every"
 387         dt=$(( dt + 1 - 1 )) # make sure it is a number
 388         if [ $dt -lt $update_every ]
 389         then
 390                 eval "$chart$suffix_update_every=$update_every"
 391         fi
 392
 393         $chart$charts_check
 394         if [ $? -eq 0 ]
 395         then
 396                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' activated"
 397                 active_charts="$active_charts $chart"
 398         else
 399                 echo >&2 "$PROGRAM_NAME: chart '$chart' check() function reports failure."
 400         fi
 401 done
 402 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: activated charts: $active_charts"
 403
 404
 405 # -----------------------------------------------------------------------------
 406 # check overwrites
 407
 408 # enable work time reporting
 409 debug_time=
 410 test $debug -eq 1 && debug_time=tellwork
 411
 412 # if we only need a specific chart, remove all the others
 413 if [ ! -z "${chart_only}" ]
 414 then
 415         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: requested to run only for: '${chart_only}'"
 416         check_charts=
 417         for chart in $active_charts
 418         do
 419                 if [ "$chart" = "$chart_only" ]
 420                 then
 421                         check_charts="$chart"
 422                         break
 423                 fi
 424         done
 425         active_charts="$check_charts"
 426 fi
 427 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: activated charts: $active_charts"
 428
 429 # stop if we just need a pre-check
 430 if [ $check -eq 1 ]
 431 then
 432         echo >&2 "CHECK RESULT"
 433         echo >&2 "Will run the charts: $active_charts"
 434         exit 0
 435 fi
 436
 437 # -----------------------------------------------------------------------------
 438 # create temp dir
 439
 440 TMP_DIR=
 441 chartsd_cleanup() {
 442         cd /tmp
 443         if [ ! -z "$TMP_DIR" -a -d "$TMP_DIR" ]
 444         then
 445                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: cleaning up temporary directory $TMP_DIR ..."
 446                 rm -rf "$TMP_DIR"
 447         fi
 448         exit 0
 449 }
 450 trap chartsd_cleanup EXIT
 451 trap chartsd_cleanup SIGHUP
 452 trap chartsd_cleanup INT
 453
 454 if [ $UID = "0" ]
 455 then
 456         TMP_DIR="$( mktemp -d /var/run/netdata-${PROGRAM_NAME}-XXXXXXXXXX )"
 457 else
 458         TMP_DIR="$( mktemp -d /tmp/.netdata-${PROGRAM_NAME}-XXXXXXXXXX )"
 459 fi
 460
 461 cd "$TMP_DIR" || exit 1
 462
 463 # -----------------------------------------------------------------------------
 464 # create charts
 465
 466 run_charts=
 467 for chart in $active_charts
 468 do
 469         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: Calling '$chart$charts_create()'..."
 470         $chart$charts_create
 471         if [ $? -eq 0 ]
 472         then
 473                 run_charts="$run_charts $chart"
 474                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' has initialized."
 475         else
 476                 echo >&2 "$PROGRAM_NAME: chart '$chart' function '$chart$charts_create()' reports failure."
 477         fi
 478 done
 479 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: run_charts='$run_charts'"
 480
 481
 482 # -----------------------------------------------------------------------------
 483 # update dimensions
 484
 485 if [ -z "$run_charts" ]
 486         then
 487         echo >&2 "$PROGRAM_NAME: No charts to collect data from."
 488         echo "DISABLE"
 489         exit 1
 490 fi
 491
 492 declare -A charts_last_update=() charts_update_every=() charts_next_update=() charts_run_counter=()
 493 global_update() {
 494         local exit_at \
 495                 c=0 dt ret last_ms exec_start_ms exec_end_ms \
 496                 chart now_charts=() next_charts=($run_charts)
 497
 498         # return the current time in ms in $now_ms
 499         current_time_ms
 500
 501         exit_at=$(( now_ms + (restart_timeout * 1000) ))
 502
 503         for chart in $run_charts
 504         do
 505                 eval "charts_update_every[$chart]=\$$chart$suffix_update_every"
 506                 test -z "${charts_update_every[$chart]}" && charts_update_every[$charts]=$update_every
 507                 charts_last_update[$chart]=$((now_ms - (now_ms % (charts_update_every[$chart] * 1000) ) ))
 508                 charts_next_update[$chart]=$(( charts_last_update[$chart] + (charts_update_every[$chart] * 1000) ))
 509                 charts_run_counter[$chart]=0
 510
 511                 echo "CHART netdata.plugin_chartsd_$chart '' 'Execution time for $chart plugin' 'milliseconds / run' netdata netdata area 90000 ${charts_update_every[$chart]}"
 512                 echo "DIMENSION run_time 'run time' absolute 1 1"
 513         done
 514
 515         # the main loop
 516         while [ 1 ]
 517         do
 518                 c=$((c + 1))
 519                 now_charts=("${next_charts[@]}")
 520                 next_charts=()
 521
 522                 # return the current time in ms in $now_ms
 523                 current_time_ms
 524
 525                 for chart in "${now_charts[@]}"
 526                 do
 527                         # echo >&2 "DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}"
 528                         if [ ${now_ms} -ge ${charts_next_update[$chart]} ]
 529                         then
 530                                 last_ms=${charts_last_update[$chart]}
 531                                 dt=$(( (now_ms - last_ms) ))
 532                                 # echo >&2 "DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}, dt: ${dt}"
 533
 534                                 charts_last_update[$chart]=${now_ms}
 535
 536                                 while [ ${charts_next_update[$chart]} -lt ${now_ms} ]
 537                                 do
 538                                         charts_next_update[$chart]=$(( charts_next_update[$chart] + (charts_update_every[$chart] * 1000) ))
 539                                 done
 540
 541                                 # the first call should not give a duration
 542                                 # so that netdata calibrates to current time
 543                                 dt=$(( dt * 1000 ))
 544                                 charts_run_counter[$chart]=$(( charts_run_counter[$chart] + 1 ))
 545                                 if [ ${charts_run_counter[$chart]} -eq 1 ]
 546                                         then
 547                                         dt=
 548                                 fi
 549
 550                                 exec_start_ms=$now_ms
 551                                 $chart$charts_update $dt
 552                                 ret=$?
 553
 554                                 # return the current time in ms in $now_ms
 555                                 current_time_ms; exec_end_ms=$now_ms
 556
 557                                 echo "BEGIN netdata.plugin_chartsd_$chart $dt"
 558                                 if [ $ret -eq 0 ]
 559                                 then
 560                                         echo "SET run_time = $(( exec_end_ms - exec_start_ms ))"
 561                                         next_charts+=($chart)
 562                                 else
 563                                         echo "SET run_time = $(( (exec_end_ms - exec_start_ms) * -1 ))"
 564                                         echo >&2 "$PROGRAM_NAME: chart '$chart' update() function reports failure. Disabling it."
 565                                 fi
 566                                 echo "END"
 567                         else
 568                                 next_charts+=($chart)
 569                         fi
 570                 done
 571
 572                 if [ "$pause_method" = "suspend" ]
 573                 then
 574                         echo "STOPPING_WAKE_ME_UP_PLEASE"
 575                         suspend || ( echo >&2 "$PROGRAM_NAME: suspend returned error $?, falling back to sleep."; loopsleepms $debug_time $update_every $time_divisor)
 576                 else
 577                         # wait the time you are required to
 578                         #loopsleepms $debug_time $update_every $time_divisor
 579                         if [ ${LOOPSLEEPMS_HIGHRES} -eq 1 ]
 580                                 then
 581                                 sleep 0.2
 582                         else
 583                                 sleep 1
 584                         fi
 585                 fi
 586
 587                 test ${now_ms} -ge ${exit_at} && exit 0
 588         done
 589 }
 590
 591 global_update