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=${minimum_update_frequency}        # this will be overwritten by the command line
  65
  66 # work around for non BASH shells
  67 charts_create="_create"
  68 charts_update="_update"
  69 charts_check="_check"
  70 charts_undescore="_"
  71
  72 # when making iterations, charts.d can loop more frequently
  73 # to prevent plugins missing iterations.
  74 # this is a percentage relative to update_every to align its
  75 # iterations.
  76 # The minimum is 10%, the maximum 100%.
  77 # So, if update_every is 1 second and time_divisor is 50,
  78 # charts.d will iterate every 500ms.
  79 # Charts will be called to collect data only if the time
  80 # passed since the last time the collected data is equal or
  81 # above their update_every.
  82 time_divisor=50
  83
  84 # number of seconds to run without restart
  85 # after this time, charts.d.plugin will exit
  86 # netdata will restart it
  87 restart_timeout=$((3600 * 4))
  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 m="$2" a b l v=($1)
 245         f=${v[0]}
 246
 247         # echo >&2 "value='${1}' f='${f}', m='${m}'"
 248
 249         # the length of the multiplier - 1
 250         l=$(( ${#m} - 1 ))
 251
 252         # check if the number is in scientific notation
 253         if [[ ${f} =~ ^[[:space:]]*(-)?[0-9.]+(e|E)(\+|-)[0-9]+ ]]
 254                 then
 255                 # convert it to decimal
 256                 # unfortunately, this fork cannot be avoided
 257                 # if you know of a way to avoid it, please let me know
 258                 f=$(printf "%0.${l}f" ${f})
 259         fi
 260
 261         # split the floating point number
 262         # in integer (a) and decimal (b)
 263         a=${f/.*/}
 264         b=${f/*./}
 265
 266         # if the integer part is missing
 267         # set it to zero
 268         [ -z "${a}" ] && a="0"
 269
 270         # strip leading zeros from the integer part
 271         # base 10 convertion
 272         a=$((10#$a))
 273
 274         # check the length of the decimal part
 275         # against the length of the multiplier
 276         if [ ${#b} -gt ${l} ]
 277                 then
 278                 # too many digits - take the most significant
 279                 b=${b:0:${l}}
 280
 281         elif [ ${#b} -lt ${l} ]
 282                 then
 283                 # too few digits - pad with zero on the right
 284                 local z="00000000000000000000000" r=$((l - ${#b}))
 285                 b="${b}${z:0:${r}}"
 286         fi
 287
 288         # strip leading zeros from the decimal part
 289         # base 10 convertion
 290         b=$((10#$b))
 291
 292         # store the result
 293         FLOAT2INT_RESULT=$(( (a * m) + b ))
 294         #echo >&2 "FLOAT2INT_RESULT='${FLOAT2INT_RESULT}'"
 295 }
 296
 297
 298 # -----------------------------------------------------------------------------
 299 # charts check functions
 300
 301 all_charts() {
 302         cd "$chartsd"
 303         [ $? -ne 0 ] && echo >&2 "$PROGRAM_NAME: Cannot cd to $chartsd" && return 1
 304
 305         ls *.chart.sh | sed "s/\.chart\.sh$//g"
 306 }
 307
 308 all_enabled_charts() {
 309         local charts= enabled=
 310
 311         # find all enabled charts
 312
 313         for chart in $( all_charts )
 314         do
 315                 eval "enabled=\$$chart"
 316                 if [ -z "${enabled}" ]
 317                         then
 318                         enabled="${enable_all_charts}"
 319                 fi
 320
 321                 if [ ! "${enabled}" = "yes" ]
 322                 then
 323                         echo >&2 "$PROGRAM_NAME: '$chart' is NOT enabled. Add a line with $chart=yes in $myconfig to enable it (or remove the line that disables it)."
 324                 else
 325                         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' is enabled."
 326                         local charts="$charts $chart"
 327                 fi
 328         done
 329
 330         local charts2=
 331         for chart in $charts
 332         do
 333                 # check the enabled charts
 334                 local check="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_check()" )"
 335                 if [ -z "$check" ]
 336                 then
 337                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_check() function. Disabling it."
 338                         continue
 339                 fi
 340
 341                 local create="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_create()" )"
 342                 if [ -z "$create" ]
 343                 then
 344                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_create() function. Disabling it."
 345                         continue
 346                 fi
 347
 348                 local update="$( cat "$chartsd/$chart.chart.sh" | sed "s/^ \+//g" | grep "^$chart$charts_update()" )"
 349                 if [ -z "$update" ]
 350                 then
 351                         echo >&2 "$PROGRAM_NAME: chart '$chart' does not seem to have a $chart$charts_update() function. Disabling it."
 352                         continue
 353                 fi
 354
 355                 # check its config
 356                 #if [ -f "$confd/$chart.conf" ]
 357                 #then
 358                 #       if [ ! -z "$( cat "$confd/$chart.conf" | sed "s/^ \+//g" | grep -v "^$" | grep -v "^#" | grep -v "^$chart$charts_undescore" )" ]
 359                 #       then
 360                 #               echo >&2 "$PROGRAM_NAME: chart's $chart config $confd/$chart.conf should only have lines starting with $chart$charts_undescore . Disabling it."
 361                 #               continue
 362                 #       fi
 363                 #fi
 364
 365                 #if [ $dryrunner -eq 1 ]
 366                 #       then
 367                 #       "$pluginsd/charts.d.dryrun-helper.sh" "$chart" "$chartsd/$chart.chart.sh" "$confd/$chart.conf" >/dev/null
 368                 #       if [ $? -ne 0 ]
 369                 #       then
 370                 #               echo >&2 "$PROGRAM_NAME: chart's $chart did not pass the dry run check. This means it uses global variables not starting with $chart. Disabling it."
 371                 #               continue
 372                 #       fi
 373                 #fi
 374
 375                 local charts2="$charts2 $chart"
 376         done
 377
 378         echo $charts2
 379         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: enabled charts: $charts2"
 380 }
 381
 382
 383 # -----------------------------------------------------------------------------
 384 # load the charts
 385
 386 suffix_update_every="_update_every"
 387 active_charts=
 388 for chart in $( all_enabled_charts )
 389 do
 390         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: loading chart: '$chartsd/$chart.chart.sh'"
 391         . "$chartsd/$chart.chart.sh"
 392
 393         if [ -f "$confd/$chart.conf" ]
 394         then
 395                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: loading chart options: '$confd/$chart.conf'"
 396                 . "$confd/$chart.conf"
 397         else
 398                 echo >&2 "$PROGRAM_NAME: $chart: configuration file '$confd/$chart.conf' not found. Using defaults."
 399         fi
 400
 401         eval "dt=\$$chart$suffix_update_every"
 402         dt=$(( dt + 1 - 1 )) # make sure it is a number
 403         if [ $dt -lt $update_every ]
 404         then
 405                 eval "$chart$suffix_update_every=$update_every"
 406         fi
 407
 408         $chart$charts_check
 409         if [ $? -eq 0 ]
 410         then
 411                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' activated"
 412                 active_charts="$active_charts $chart"
 413         else
 414                 echo >&2 "$PROGRAM_NAME: chart '$chart' check() function reports failure."
 415         fi
 416 done
 417 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: activated charts: $active_charts"
 418
 419
 420 # -----------------------------------------------------------------------------
 421 # check overwrites
 422
 423 # enable work time reporting
 424 debug_time=
 425 test $debug -eq 1 && debug_time=tellwork
 426
 427 # if we only need a specific chart, remove all the others
 428 if [ ! -z "${chart_only}" ]
 429 then
 430         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: requested to run only for: '${chart_only}'"
 431         check_charts=
 432         for chart in $active_charts
 433         do
 434                 if [ "$chart" = "$chart_only" ]
 435                 then
 436                         check_charts="$chart"
 437                         break
 438                 fi
 439         done
 440         active_charts="$check_charts"
 441 fi
 442 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: activated charts: $active_charts"
 443
 444 # stop if we just need a pre-check
 445 if [ $check -eq 1 ]
 446 then
 447         echo >&2 "CHECK RESULT"
 448         echo >&2 "Will run the charts: $active_charts"
 449         exit 0
 450 fi
 451
 452 # -----------------------------------------------------------------------------
 453 # create temp dir
 454
 455 TMP_DIR=
 456 chartsd_cleanup() {
 457         cd /tmp
 458         if [ ! -z "$TMP_DIR" -a -d "$TMP_DIR" ]
 459         then
 460                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: cleaning up temporary directory $TMP_DIR ..."
 461                 rm -rf "$TMP_DIR"
 462         fi
 463         exit 0
 464 }
 465 trap chartsd_cleanup EXIT
 466 trap chartsd_cleanup SIGHUP
 467 trap chartsd_cleanup INT
 468
 469 if [ $UID = "0" ]
 470 then
 471         TMP_DIR="$( mktemp -d /var/run/netdata-${PROGRAM_NAME}-XXXXXXXXXX )"
 472 else
 473         TMP_DIR="$( mktemp -d /tmp/.netdata-${PROGRAM_NAME}-XXXXXXXXXX )"
 474 fi
 475
 476 cd "$TMP_DIR" || exit 1
 477
 478 # -----------------------------------------------------------------------------
 479 # create charts
 480
 481 run_charts=
 482 for chart in $active_charts
 483 do
 484         [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: Calling '$chart$charts_create()'..."
 485         $chart$charts_create
 486         if [ $? -eq 0 ]
 487         then
 488                 run_charts="$run_charts $chart"
 489                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: '$chart' has initialized."
 490         else
 491                 echo >&2 "$PROGRAM_NAME: chart '$chart' function '$chart$charts_create()' reports failure."
 492         fi
 493 done
 494 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: run_charts='$run_charts'"
 495
 496
 497 # -----------------------------------------------------------------------------
 498 # update dimensions
 499
 500 if [ -z "$run_charts" ]
 501         then
 502         echo >&2 "$PROGRAM_NAME: No charts to collect data from."
 503         echo "DISABLE"
 504         exit 1
 505 fi
 506
 507 declare -A charts_last_update=() charts_update_every=() charts_next_update=() charts_run_counter=() charts_serial_failures=()
 508 global_update() {
 509         local exit_at \
 510                 c=0 dt ret last_ms exec_start_ms exec_end_ms \
 511                 chart now_charts=() next_charts=($run_charts) \
 512                 next_ms x seconds millis
 513
 514         # return the current time in ms in $now_ms
 515         current_time_ms
 516
 517         exit_at=$(( now_ms + (restart_timeout * 1000) ))
 518
 519         for chart in $run_charts
 520         do
 521                 eval "charts_update_every[$chart]=\$$chart$suffix_update_every"
 522                 test -z "${charts_update_every[$chart]}" && charts_update_every[$charts]=$update_every
 523                 charts_last_update[$chart]=$((now_ms - (now_ms % (charts_update_every[$chart] * 1000) ) ))
 524                 charts_next_update[$chart]=$(( charts_last_update[$chart] + (charts_update_every[$chart] * 1000) ))
 525                 charts_run_counter[$chart]=0
 526                 charts_serial_failures[$chart]=0
 527
 528                 echo "CHART netdata.plugin_chartsd_$chart '' 'Execution time for $chart plugin' 'milliseconds / run' charts.d netdata.plugin_charts area 145000 ${charts_update_every[$chart]}"
 529                 echo "DIMENSION run_time 'run time' absolute 1 1"
 530         done
 531
 532         # the main loop
 533         while [ "${#next_charts[@]}" -gt 0 ]
 534         do
 535                 c=$((c + 1))
 536                 now_charts=("${next_charts[@]}")
 537                 next_charts=()
 538
 539                 # return the current time in ms in $now_ms
 540                 current_time_ms
 541
 542                 for chart in "${now_charts[@]}"
 543                 do
 544                         # echo >&2 "DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}"
 545                         if [ ${now_ms} -ge ${charts_next_update[$chart]} ]
 546                         then
 547                                 last_ms=${charts_last_update[$chart]}
 548                                 dt=$(( (now_ms - last_ms) ))
 549                                 # echo >&2 "DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}, dt: ${dt}"
 550
 551                                 charts_last_update[$chart]=${now_ms}
 552
 553                                 while [ ${charts_next_update[$chart]} -lt ${now_ms} ]
 554                                 do
 555                                         charts_next_update[$chart]=$(( charts_next_update[$chart] + (charts_update_every[$chart] * 1000) ))
 556                                 done
 557
 558                                 # the first call should not give a duration
 559                                 # so that netdata calibrates to current time
 560                                 dt=$(( dt * 1000 ))
 561                                 charts_run_counter[$chart]=$(( charts_run_counter[$chart] + 1 ))
 562                                 if [ ${charts_run_counter[$chart]} -eq 1 ]
 563                                         then
 564                                         dt=
 565                                 fi
 566
 567                                 exec_start_ms=$now_ms
 568                                 $chart$charts_update $dt
 569                                 ret=$?
 570
 571                                 # return the current time in ms in $now_ms
 572                                 current_time_ms; exec_end_ms=$now_ms
 573
 574                                 echo "BEGIN netdata.plugin_chartsd_$chart $dt"
 575                                 echo "SET run_time = $(( exec_end_ms - exec_start_ms ))"
 576                                 echo "END"
 577
 578                                 if [ $ret -eq 0 ]
 579                                 then
 580                                         charts_serial_failures[$chart]=0
 581                                         next_charts+=($chart)
 582                                 else
 583                                         charts_serial_failures[$chart]=$(( charts_serial_failures[$chart] + 1 ))
 584
 585                                         if [ charts_serial_failures[$chart] -gt 10 ]
 586                                                 then
 587                                                 echo >&2 "$PROGRAM_NAME: chart '$chart' update() function reported failure ${charts_serial_failures[$chart]} times. Disabling it."
 588                                         else
 589                                                 echo >&2 "$PROGRAM_NAME: chart '$chart' update() function reports failure. Will keep trying for a while."
 590                                                 next_charts+=($chart)
 591                                         fi
 592                                 fi
 593                         else
 594                                 next_charts+=($chart)
 595                         fi
 596                 done
 597
 598                 if [ "$pause_method" = "suspend" ]
 599                 then
 600                         echo "STOPPING_WAKE_ME_UP_PLEASE"
 601                         suspend || ( echo >&2 "$PROGRAM_NAME: suspend returned error $?, falling back to sleep."; loopsleepms $debug_time $update_every $time_divisor)
 602                 else
 603                         # wait the time you are required to
 604                         next_ms=$((now_ms + (update_every * 1000 * 100) ))
 605                         for x in "${charts_next_update[@]}"; do [ ${x} -lt ${next_ms} ] && next_ms=${x}; done
 606                         next_ms=$((next_ms - now_ms))
 607
 608                         if [ ${LOOPSLEEPMS_HIGHRES} -eq 1 -a ${next_ms} -gt 0 ]
 609                                 then
 610                                 seconds=$(( next_ms / 1000 ))
 611                                 millis=$(( next_ms % 1000 ))
 612                                 [ ${millis} -lt 10  ] && millis="0${millis}"
 613                                 [ ${millis} -lt 100 ] && millis="0${millis}"
 614                                 [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: sleeping for ${seconds}.${millis} seconds."
 615                                 sleep ${seconds}.${millis}
 616                         else
 617                                 sleep $update_every
 618                         fi
 619                 fi
 620
 621                 test ${now_ms} -ge ${exit_at} && exit 0
 622         done
 623
 624         echo >&2 "$PROGRAM_NAME: Nothing left to do. Disabling charts.d.plugin."
 625         echo "DISABLE"
 626 }
 627
 628 global_update