watchdog.yaml

# Template          : Quick watchdog to ensure processes are running
# Author            : Nuno Aguiar
# How to run        : ojob watchdog.yaml (in crontab every x minutes)
# oPack dependencies: latest ojob-common;

include:
  - oJobWatchDog.yaml

ojob:
  logToFile   :
    logFolder            : /my/dir/watchdog.logs
    HKhowLongAgoInMinutes: 10080                  # keep logs for 7 days
    setLogOff            : true
  logToConsole: false
  logJobs     : false
  logArgs     : false
  unique      :
    pidFile     : /my/dir/watchdog.pid
    killPrevious: true
  checkStall  :
    everySeconds    : 1
    killAfterSeconds: 60

todo:
  - Watchdog for process A
  #- Watchdog for process B

jobs:
  - name: Watchdog for process A
    to  :
      - oJob WatchDog
    args:
      nameOfDog   : Process A dog
      quiet       : false
      #jobToNotify : Barking to someone
      cmdToStart  : "/usr/bin/java -Xmx64m /my/processes/A/start.sh"
      workDirStart: /my/processes/A

      checks      :
        pid   :
          file  : /my/processes/A/a.pid

        log   :
          folder  : /my/processes/A/log
          fileRE  : "A-.+\.log"
          histFile: /my/dir/.watchdog.A.history  # where to keep track of what was previously saw on the log file
          stringRE:
            - java\.lang\.OutOfMemory

        custom:
          exec: |
            try {
              // my custom code to check if A is working
              return true;
            } catch(e) {
              logErr("Error: " + String(e));
              return false;
            }

  #- name: Watchdog for process B
  #  to  :
  #    - oJob WatchDog
  #  args:
  #    nameOfDog   : Process B dog
  #    quiet       : false
  #    #jobToNotify : Call someone
  #    cmdToStart  : "/usr/bin/java -Xmx64m /my/processes/B/start.sh"
  #    workDirStart: /my/processes/B
  #
  #    checks      :
  #      pid   :
  #        file  : /my/processes/B/b.pid
  #
  #      custom:
  #        exec: |
  #          try {
  #            // my custom code to check if B is working
  #            return true;
  #          } catch(e) {
  #            logErr("Error: " + String(e));
  #            return false;
  #          }