BOSH release for Telegraf

.final_builds/jobs/telegraf-system/index.yml

@@ -0,0 +1,7 @@
+---
+builds:
+  16318fd8b9dea2a69c153199dec6e382c035e233:
+    version: 16318fd8b9dea2a69c153199dec6e382c035e233
+    sha1: a8b114a4f220ad570299318f5cece01ffbebd8c0
+    blobstore_id: 26a67da1-70ab-4ad7-92c3-922ed66d0e69
+format-version: '2'

.final_builds/packages/telegraf/index.yml

@@ -0,0 +1,7 @@
+---
+builds:
+  11d238e06a154b9a8539712c54432ffc968bf7c5:
+    version: 11d238e06a154b9a8539712c54432ffc968bf7c5
+    sha1: a63307c67af09d5a936c8f0180521727ad3ffa14
+    blobstore_id: ace8c27e-4c19-4f0e-80c8-58787ec9225f
+format-version: '2'

.gitignore

@@ -0,0 +1,18 @@
+config/dev.yml
+config/private.yml
+config/settings.yml
+releases/*.tgz
+releases/**/*.tgz
+dev_releases
+.blobs
+blobs
+.dev_builds
+.idea
+.DS_Store
+.final_builds/jobs/**/*.tgz
+.final_builds/packages/**/*.tgz
+*.swp
+*~
+
+/deployments
+.torus.json

config/blobs.yml

@@ -0,0 +1,5 @@
+---
+telegraf/telegraf-1.2.1_linux_amd64.tar.gz:
+  object_id: e61a55cf-53f9-46d3-9a02-279e82f424e1
+  sha: 5c61cabe4ac5c5fde972262194dd26e561cd8579
+  size: 9635266

config/final.yml

@@ -0,0 +1,6 @@
+---
+blobstore:
+  provider: s3
+  options:
+    bucket_name: bosh-release-telegraf
+final_name: c-telegraf

jobs/telegraf-system/monit

@@ -0,0 +1,5 @@
+check process telegraf-system
+  with pidfile /var/vcap/sys/run/telegraf-system/telegraf-system.pid
+  start program "/var/vcap/jobs/telegraf-system/bin/telegraf-system_ctl start"
+  stop program "/var/vcap/jobs/telegraf-system/bin/telegraf-system_ctl stop"
+  group vcap

jobs/telegraf-system/spec

@@ -0,0 +1,25 @@
+---
+name: telegraf-system
+
+templates:
+  bin/telegraf-system_ctl: bin/telegraf-system_ctl
+  bin/pid_utils.sh: bin/pid_utils.sh
+  config/telegraf.conf.erb: config/telegraf.conf
+
+packages:
+- telegraf
+
+properties:
+  telegraf.tags:
+   description: Telegraf global tags as a dictionary
+   default: {}
+
+  telegraf.influxdb.url:
+    description: InfluxDB url
+    default: http://localhost:8086
+  telegraf.influxdb.database:
+    description: InfluxDB database
+    default: system
+  telegraf.influxdb.retention_policy:
+    description: InfluxDB database retention policy
+    default: autogen

jobs/telegraf-system/templates/bin/pid_utils.sh

@@ -0,0 +1,186 @@
+SCRIPT=$(basename $0)
+mkdir -p /var/vcap/sys/log/monit
+
+exec 1>> /var/vcap/sys/log/monit/$SCRIPT.log
+exec 2>> /var/vcap/sys/log/monit/$SCRIPT.err.log
+
+echo "------------ `basename $0` $* at `date` --------------" | tee /dev/stderr
+
+function pid_is_running() {
+  declare pid="$1"
+  ps -p "${pid}" >/dev/null 2>&1
+}
+
+# pid_guard
+#
+# @param pidfile
+# @param name [String] an arbitrary name that might show up in STDOUT on errors
+#
+# Run this before attempting to start new processes that may use the same :pidfile:.
+# If an old process is running on the pid found in the :pidfile:, exit 1. Otherwise,
+# remove the stale :pidfile: if it exists.
+#
+function pid_guard() {
+  declare pidfile="$1" name="$2"
+
+  echo "------------ STARTING $(basename "$0") at $(date) --------------" | tee /dev/stderr
+
+  if [ ! -f "${pidfile}" ]; then
+    return 0
+  fi
+
+  local pid
+  pid=$(head -1 "${pidfile}")
+
+  if pid_is_running "${pid}"; then
+    echo "${name} is already running, please stop it first"
+    exit 1
+  fi
+
+  echo "Removing stale pidfile"
+  rm "${pidfile}"
+}
+
+# wait_pid_death
+#
+# @param pid
+# @param timeout
+#
+# Watch a :pid: for :timeout: seconds, waiting for it to die.
+# If it dies before :timeout:, exit 0. If not, exit 1.
+#
+# Note that this should be run in a subshell, so that the current
+# shell does not exit.
+#
+function wait_pid_death() {
+  declare pid="$1" timeout="$2"
+
+  local countdown
+  countdown=$(( timeout * 10 ))
+
+  while true; do
+    if ! pid_is_running "${pid}"; then
+      return 0
+    fi
+
+    if [ ${countdown} -le 0 ]; then
+      return 1
+    fi
+
+    countdown=$(( countdown - 1 ))
+    sleep 0.1
+  done
+}
+
+# kill_and_wait
+#
+# @param pidfile
+# @param timeout [default 25s]
+#
+# For a pid found in :pidfile:, send a `kill -6`, then wait for :timeout: seconds to
+# see if it dies on its own. If not, send it a `kill -9`. If the process does die,
+# exit 0 and remove the :pidfile:. If after all of this, the process does not actually
+# die, exit 1.
+#
+# Note:
+# Monit default timeout for start/stop is 30s
+# Append 'with timeout {n} seconds' to monit start/stop program configs
+#
+function kill_and_wait() {
+  declare pidfile="$1" timeout="${2:-25}" sigkill_on_timeout="${3:-1}"
+
+  if [ ! -f "${pidfile}" ]; then
+    echo "Pidfile ${pidfile} doesn't exist"
+    exit 0
+  fi
+
+  local pid
+  pid=$(head -1 "${pidfile}")
+
+  if [ -z "${pid}" ]; then
+    echo "Unable to get pid from ${pidfile}"
+    exit 1
+  fi
+
+  if ! pid_is_running "${pid}"; then
+    echo "Process ${pid} is not running"
+    rm -f "${pidfile}"
+    exit 0
+  fi
+
+  echo "Killing ${pidfile}: ${pid} "
+  kill "${pid}"
+
+  if ! wait_pid_death "${pid}" "${timeout}"; then
+    if [ "${sigkill_on_timeout}" = "1" ]; then
+      echo "Kill timed out, using kill -9 on ${pid}"
+      kill -9 "${pid}"
+      sleep 0.5
+    fi
+  fi
+
+  if pid_is_running "${pid}"; then
+    echo "Timed Out"
+    exit 1
+  else
+    echo "Stopped"
+    rm -f "${pidfile}"
+  fi
+}
+
+check_mount() {
+  opts=$1
+  exports=$2
+  mount_point=$3
+
+  if grep -qs $mount_point /proc/mounts; then
+    echo "Found NFS mount $mount_point"
+  else
+    echo "Mounting NFS..."
+    mount $opts $exports $mount_point
+    if [ $? != 0 ]; then
+      echo "Cannot mount NFS from $exports to $mount_point, exiting..."
+      exit 1
+    fi
+  fi
+}
+
+# Check the syntax of a sudoers file.
+check_sudoers() {
+  /usr/sbin/visudo -c -f "$1"
+}
+
+# Check the syntax of a sudoers file and if it's ok install it.
+install_sudoers() {
+  src="$1"
+  dest="$2"
+
+  check_sudoers "$src"
+
+  if [ $? -eq 0 ]; then
+    chown root:root "$src"
+    chmod 0440 "$src"
+    cp -p "$src" "$dest"
+  else
+    echo "Syntax error in sudoers file $src"
+    exit 1
+  fi
+}
+
+# Add a line to a file if it is not already there.
+file_must_include() {
+  file="$1"
+  line="$2"
+
+  # Protect against empty $file so it doesn't wait for input on stdin.
+  if [ -n "$file" ]; then
+    grep --quiet "$line" "$file" || echo "$line" >> "$file"
+  else
+    echo 'File name is required'
+    exit 1
+  fi
+}
+
+running_in_container() {
+  grep -q -E '/instance|/docker/' /proc/self/cgroup
+}

jobs/telegraf-system/templates/bin/telegraf-system_ctl

@@ -0,0 +1,43 @@
+#!/bin/bash -eu
+
+JOB_NAME=telegraf-system
+
+JOB_DIR=/var/vcap/jobs/$JOB_NAME
+RUN_DIR=/var/vcap/sys/run/$JOB_NAME
+LOG_DIR=/var/vcap/sys/log/$JOB_NAME
+PIDFILE=$RUN_DIR/$JOB_NAME.pid
+
+source $JOB_DIR/bin/pid_utils.sh
+
+case $1 in
+  start)
+    pid_guard $PIDFILE $JOB_NAME
+
+    mkdir -p $RUN_DIR
+    mkdir -p $LOG_DIR
+
+    PATH=/var/vcap/packages/telegraf/bin:$PATH
+
+    # check configuration
+    echo "checking telegraf configuration..."
+    telegraf --config $JOB_DIR/config/telegraf.conf --test > /dev/null
+    echo "telegraf configuration OK"
+    echo
+
+    # store pid in $PIDFILE
+    echo $$ > $PIDFILE
+
+    exec telegraf \
+      --config $JOB_DIR/config/telegraf.conf \
+      >>$LOG_DIR/$JOB_NAME.log 2>&1
+
+    ;;
+
+  stop)
+    kill_and_wait $PIDFILE
+    ;;
+
+  *)
+    echo "Usage: ctl {start|stop}"
+    ;;
+esac

jobs/telegraf-system/templates/config/telegraf.conf.erb

@@ -0,0 +1,34 @@
+[global_tags]
+  service = "<%= spec.deployment.sub(/-[0-9a-f]{20}$/, '') %>"
+  node_type = "<%= spec.name %>"
+  instance = "<%= spec.index %>"
+  <% if_p("telegraf.tags") do |tags| tags.each do |tag, value| %>
+  <%= tag %> = "<%= value %>"
+  <% end; end %>
+
+[agent]
+  interval = "1m"
+  flush_interval = "1m"
+  flush_jitter = "15s"
+
+[[outputs.influxdb]]
+  urls = ["<%= p("telegraf.influxdb.url") %>"]
+  database = "<%= p("telegraf.influxdb.database") %>"
+  retention_policy = "<%= p("telegraf.influxdb.retention_policy") %>"
+
+[[inputs.cpu]]
+  percpu = false
+  totalcpu = true
+[[inputs.disk]]
+  ignore_fs = ["tmpfs", "devtmpfs", "none"]
+[[inputs.diskio]]
+  devices = ["sda", "sdb", "sdc"]
+[[inputs.kernel]]
+[[inputs.mem]]
+[[inputs.processes]]
+[[inputs.swap]]
+[[inputs.system]]
+[[inputs.net]]
+[[inputs.netstat]]
+[[inputs.nstat]]
+  dump_zeros = true

packages/telegraf/packaging

@@ -0,0 +1,9 @@
+#!/bin/sh
+
+set -e
+set -u
+
+tar xfz telegraf/telegraf-*.tar.gz \
+  --strip-components 3 \
+  --wildcards \
+  -C $BOSH_INSTALL_TARGET './*/usr/bin/telegraf'

packages/telegraf/spec

@@ -0,0 +1,4 @@
+---
+name: telegraf
+files:
+- telegraf/*

releases/c-telegraf/c-telegraf-1.yml

@@ -0,0 +1,16 @@
+---
+packages:
+- name: telegraf
+  version: 11d238e06a154b9a8539712c54432ffc968bf7c5
+  fingerprint: 11d238e06a154b9a8539712c54432ffc968bf7c5
+  sha1: a63307c67af09d5a936c8f0180521727ad3ffa14
+  dependencies: []
+jobs:
+- name: telegraf-system
+  version: 16318fd8b9dea2a69c153199dec6e382c035e233
+  fingerprint: 16318fd8b9dea2a69c153199dec6e382c035e233
+  sha1: a8b114a4f220ad570299318f5cece01ffbebd8c0
+commit_hash: 80f31f96
+uncommitted_changes: true
+name: c-telegraf
+version: '1'

releases/c-telegraf/index.yml

@@ -0,0 +1,5 @@
+---
+builds:
+  a8b815c2-9a70-4860-b921-1a07538305cc:
+    version: '1'
+format-version: '2'