charts.d.plugin and tc-qos-helper.sh optimization to avoid frequent forks due to date and sleep

4 files changed, 129 insertions, 79 deletions

diff --git a/charts.d/example.chart.sh b/charts.d/example.chart.sh
index 1f020dee27..93f1cf4fdd 100755
--- a/charts.d/example.chart.sh
+++ b/charts.d/example.chart.sh
@@ -48,8 +48,8 @@ example_get() {
 	else
 		example_count=$((1 + (RANDOM * 5 / 32767) ))
 
-		[ $example_last -gt 16383 -a $example_value4 -gt 16383 ] && example_value4=$((value4 - 16383))
-		[ $example_last -le 16383 -a $example_value4 -lt 16383 ] && example_value4=$((value4 + 16383))
+		[ $example_last -gt 16383 -a $example_value4 -gt 16383 ] && example_value4=$((example_value4 - 16383))
+		[ $example_last -le 16383 -a $example_value4 -lt 16383 ] && example_value4=$((example_value4 + 16383))
 	fi
 	example_last=$example_value4
 
diff --git a/plugins.d/charts.d.plugin b/plugins.d/charts.d.plugin
index 9aaadc1685..df9998ecec 100755
--- a/plugins.d/charts.d.plugin
+++ b/plugins.d/charts.d.plugin
@@ -48,10 +48,6 @@ require_cmd awk || exit 1
 # insternal defaults
 # netdata exposes a few environment variables for us
 
-pause_method="sleep" # use either "suspend" or "sleep"
-                     # DO NOT USE SUSPEND - LINUX WILL SUSPEND NETDATA TOO
-                     # THE WHOLE PROCESS GROUP - NOT JUST THE SHELL
-
 pluginsd="${NETDATA_PLUGINS_DIR}"
 [ -z "$pluginsd" ] && pluginsd="$( dirname $PROGRAM_FILE )"
 
@@ -154,6 +150,24 @@ done
 
 
 # -----------------------------------------------------------------------------
+# loop control
+
+# default sleep function
+LOOPSLEEPMS_HIGHRES=0
+now_ms=
+current_time_ms_default() {
+    now_ms="$(date +'%s')000"
+}
+current_time_ms="current_time_ms_default"
+current_time_ms_accuracy=1
+mysleep="sleep"
+
+# if found and included, this file overwrites loopsleepms()
+# and current_time_ms() with a high resolution timer function
+# for precise looping.
+. "$pluginsd/loopsleepms.sh.inc"
+
+# -----------------------------------------------------------------------------
 # load my configuration
 
 if [ -f "$myconfig" ]
@@ -172,13 +186,6 @@ else
     echo >&2 "$PROGRAM_NAME: configuration file '$myconfig' not found. Using defaults."
 fi
 
-if [ "$pause_method" = "suspend" ]
-then
-    # enable bash job control
-    # this is required for suspend to work
-    set -m
-fi
-
 # we check for the timeout command, after we load our
 # configuration, so that the user may overwrite the
 # timeout command we use, providing a function that
@@ -205,27 +212,6 @@ fi
 
 
 # -----------------------------------------------------------------------------
-# loop control
-
-# default sleep function
-LOOPSLEEPMS_HIGHRES=0
-loopsleepms() {
-    [ "$1" = "tellwork" ] && shift
-    sleep $1
-}
-
-now_ms=
-current_time_ms() {
-    now_ms="$(date +'%s')000"
-}
-
-# if found and included, this file overwrites loopsleepms()
-# and current_time_ms() with a high resolution timer function
-# for precise looping.
-. "$pluginsd/loopsleepms.sh.inc"
-
-
-# -----------------------------------------------------------------------------
 # library functions
 
 fixid() {
@@ -537,7 +523,7 @@ global_update() {
         next_ms x seconds millis
 
     # return the current time in ms in $now_ms
-    current_time_ms
+    ${current_time_ms}
 
     exit_at=$(( now_ms + (restart_timeout * 1000) ))
 
@@ -562,16 +548,16 @@ global_update() {
         next_charts=()
 
         # return the current time in ms in $now_ms
-        current_time_ms
+        ${current_time_ms}
 
         for chart in "${now_charts[@]}"
         do
-            # echo >&2 "DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}"
+            #echo >&2 "                              DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}"
             if [ ${now_ms} -ge ${charts_next_update[$chart]} ]
             then
                 last_ms=${charts_last_update[$chart]}
                 dt=$(( (now_ms - last_ms) ))
-                # echo >&2 "DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}, dt: ${dt}"
+                #echo >&2 "                              DEBUG: chart: $chart last: ${charts_last_update[$chart]}, next: ${charts_next_update[$chart]}, now: ${now_ms}, dt: ${dt}"
 
                 charts_last_update[$chart]=${now_ms}
 
@@ -590,11 +576,12 @@ global_update() {
                 fi
 
                 exec_start_ms=$now_ms
+                #echo >&2 "                              EXEC: $chart$charts_update $dt"
                 $chart$charts_update $dt
                 ret=$?
 
                 # return the current time in ms in $now_ms
-                current_time_ms; exec_end_ms=$now_ms
+                ${current_time_ms}; exec_end_ms=$now_ms
 
                 echo "BEGIN netdata.plugin_chartsd_$chart $dt"
                 echo "SET run_time = $(( exec_end_ms - exec_start_ms ))"
@@ -620,27 +607,29 @@ global_update() {
             fi
         done
 
-        if [ "$pause_method" = "suspend" ]
-        then
-            echo "STOPPING_WAKE_ME_UP_PLEASE"
-            suspend || ( echo >&2 "$PROGRAM_NAME: suspend returned error $?, falling back to sleep."; loopsleepms $debug_time $update_every $time_divisor)
-        else
-            # wait the time you are required to
-            next_ms=$((now_ms + (update_every * 1000 * 100) ))
-            for x in "${charts_next_update[@]}"; do [ ${x} -lt ${next_ms} ] && next_ms=${x}; done
-            next_ms=$((next_ms - now_ms))
+        # wait the time you are required to
+        next_ms=$((now_ms + (update_every * 1000 * 100) ))
+        for x in "${charts_next_update[@]}"; do [ ${x} -lt ${next_ms} ] && next_ms=${x}; done
+        next_ms=$((next_ms - now_ms))
 
-            if [ ${LOOPSLEEPMS_HIGHRES} -eq 1 -a ${next_ms} -gt 0 ]
+        if [ ${LOOPSLEEPMS_HIGHRES} -eq 1 -a ${next_ms} -gt 0 ]
+            then
+            next_ms=$(( next_ms + current_time_ms_accuracy ))
+            seconds=$(( next_ms / 1000 ))
+            millis=$(( next_ms % 1000 ))
+            if [ ${millis} -lt 10  ]
                 then
-                seconds=$(( next_ms / 1000 ))
-                millis=$(( next_ms % 1000 ))
-                [ ${millis} -lt 10  ] && millis="0${millis}"
-                [ ${millis} -lt 100 ] && millis="0${millis}"
-                [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: sleeping for ${seconds}.${millis} seconds."
-                sleep ${seconds}.${millis}
-            else
-                sleep $update_every
+                millis="00${millis}"
+            elif [ ${millis} -lt 100 ]
+                then
+                millis="0${millis}"
             fi
+
+            [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: sleeping for ${seconds}.${millis} seconds."
+            ${mysleep} ${seconds}.${millis}
+        else
+            [ $debug -eq 1 ] && echo >&2 "$PROGRAM_NAME: sleeping for ${update_every} seconds."
+            ${mysleep} $update_every
         fi
 
         test ${now_ms} -ge ${exit_at} && exit 0
diff --git a/plugins.d/loopsleepms.sh.inc b/plugins.d/loopsleepms.sh.inc
index 02ab694d26..2a81152ea8 100644
--- a/plugins.d/loopsleepms.sh.inc
+++ b/plugins.d/loopsleepms.sh.inc
@@ -1,10 +1,5 @@
 # no need for shebang - this file is included from other scripts
 
-# this function is used to sleep a fraction of a second
-# it calculates the difference between every time is called
-# and tries to align the sleep time to give you exactly the
-# loop you need.
-
 LOOPSLEEP_DATE="$(which date)"
 if [ -z "$LOOPSLEEP_DATE" ]
     then
@@ -12,17 +7,13 @@ if [ -z "$LOOPSLEEP_DATE" ]
     exit 1
 fi
 
-LOOPSLEEPMS_LASTRUN=0
-LOOPSLEEPMS_LASTSLEEP=0
-LOOPSLEEPMS_LASTWORK=0
+# -----------------------------------------------------------------------------
+# use the date command as a high resolution timer
 
+now_ms=
 LOOPSLEEPMS_HIGHRES=1
 test "$($LOOPSLEEP_DATE +%N)" = "%N" && LOOPSLEEPMS_HIGHRES=0
-
-now_ms=
-current_time_ms() {
-    # if high resolution is not supported
-    # just sleep the time requested, in seconds
+current_time_ms_from_date() {
     if [ $LOOPSLEEPMS_HIGHRES -eq 0 ]
     then
         now_ms="$($LOOPSLEEP_DATE +'%s')000"
@@ -31,6 +22,83 @@ current_time_ms() {
     fi
 }
 
+# -----------------------------------------------------------------------------
+# use /proc/uptime as a high resolution timer
+
+current_time_ms_from_date
+current_time_ms_from_uptime_started="${now_ms}"
+current_time_ms_from_uptime_first=0
+current_time_ms_from_uptime() {
+    local up rest arr n
+
+    read up rest </proc/uptime
+    arr=(${up//./ })
+
+    if [ ${#arr[1]} -lt 1 ]
+        then
+        n="${arr[0]}000"
+    elif [ ${#arr[1]} -lt 2 ]
+        then
+        n="${arr[0]}${arr[1]}00"
+    elif [ ${#arr[1]} -lt 3 ]
+        then
+        n="${arr[0]}${arr[1]}0"
+    else
+        n="${arr[0]}${arr[1]}"
+    fi
+
+    now_ms=$((current_time_ms_from_uptime_started - current_time_ms_from_uptime_first + n))
+
+    if [ "${now_ms}" -lt "${current_time_ms_from_uptime_started}" ]
+        then
+        echo >&2 "$0: Cannot use current_time_ms_from_uptime() - falling back to current_time_ms_from_date()."
+        current_time_ms="current_time_ms_from_date"
+        current_time_ms_from_date
+        current_time_ms_accuracy=1
+    fi
+}
+current_time_ms_from_uptime
+current_time_ms_from_uptime_first="$((now_ms - current_time_ms_from_uptime_started))"
+current_time_ms="current_time_ms_from_uptime"
+current_time_ms_accuracy=10
+if [ "${current_time_ms_from_uptime_first}" -eq 0 ]
+    then
+    echo >&2 "$0: Invalid setup for current_time_ms_from_uptime() - falling back to current_time_ms_from_date()."
+    current_time_ms="current_time_ms_from_date"
+    current_time_ms_accuracy=1
+fi
+
+# -----------------------------------------------------------------------------
+# use read with timeout for sleep
+
+mysleep="mysleep_read"
+
+mysleep_fifo="${NETDATA_CACHE_DIR-/tmp}/.netdata_bash_sleep_timer_fifo"
+[ -e "${mysleep_fifo}" ] && rm "${mysleep_fifo}"
+mkfifo "${mysleep_fifo}" || mysleep="sleep"
+
+mysleep_read() {
+    read -t "${1}" <>"${mysleep_fifo}"
+    ret=$?
+    if [ $ret -le 128 ]
+        then
+        echo >&2 "$0: Cannot use read for sleeping (return code ${ret})."
+        mysleep="sleep"
+        ${mysleep} "${1}"
+    fi
+}
+
+
+# -----------------------------------------------------------------------------
+# this function is used to sleep a fraction of a second
+# it calculates the difference between every time is called
+# and tries to align the sleep time to give you exactly the
+# loop you need.
+
+LOOPSLEEPMS_LASTRUN=0
+LOOPSLEEPMS_LASTSLEEP=0
+LOOPSLEEPMS_LASTWORK=0
+
 loopsleepms() {
     local tellwork=0 t="$1" div s m now mstosleep
 
@@ -68,7 +136,7 @@ loopsleepms() {
     # echo "# last loop's work took $LOOPSLEEPMS_LASTWORK ms"
 
     # calculate ms to sleep
-    mstosleep=$(( t - LOOPSLEEPMS_LASTWORK ))
+    mstosleep=$(( t - LOOPSLEEPMS_LASTWORK + current_time_ms_accuracy ))
     # echo "# mstosleep is $mstosleep ms"
 
     # if we are too slow, sleep some time
@@ -81,7 +149,7 @@ loopsleepms() {
 
     # echo "# sleeping $s.$m"
     # echo
-    sleep $s.$m
+    ${mysleep} $s.$m
 
     # keep the values we need
     # for our next run
diff --git a/plugins.d/tc-qos-helper.sh b/plugins.d/tc-qos-helper.sh
index bff5217d28..9caef85f72 100755
--- a/plugins.d/tc-qos-helper.sh
+++ b/plugins.d/tc-qos-helper.sh
@@ -27,16 +27,9 @@ if [ -f "${config_dir}/tc-qos-helper.conf" ]
     source "${config_dir}/tc-qos-helper.conf"
 fi
 
-# default time function
-now_ms=
-current_time_ms() {
-    now_ms="$(date +'%s')000"
-}
-
 # default sleep function
 LOOPSLEEPMS_LASTWORK=0
 loopsleepms() {
-    [ "$1" = "tellwork" ] && shift
     sleep $1
 }