ultimate_cron.nagios.inc in Ultimate Cron 8

File

ultimate_cron.nagios.inc

<?php

/**
 * Implements hook_nagios_info().
 */
function ultimate_cron_nagios_info() {
  return array(
    'name' => t('Ultimate Cron Monitoring'),
    'id' => 'ULTIMATE_CRON',
  );
}

/**
 * Implementation of hook_nagios().
 */
function ultimate_cron_nagios($check = 'nagios') {
  $status = array();
  foreach (ultimate_cron_nagios_functions() as $function => $description) {
    if (variable_get('ultimate_cron_nagios_func_' . $function, TRUE) && ($check == 'nagios' || $check == $function)) {
      $func = $function . '_check';
      $result = $func();
      $status[$result['key']] = $result['data'];
    }
  }
  return $status;
}

/**
 * Implementation of hook_nagios_settings().
 */
function ultimate_cron_nagios_settings() {
  $form = array();
  foreach (ultimate_cron_nagios_functions() as $function => $description) {
    $var = 'ultimate_cron_nagios_func_' . $function;
    $form[$var] = array(
      '#type' => 'checkbox',
      '#title' => $function,
      '#default_value' => variable_get($var, TRUE),
      '#description' => $description,
    );
  }
  $group = 'thresholds';
  $form[$group] = array(
    '#type' => 'fieldset',
    '#collapsible' => TRUE,
    '#collapsed' => FALSE,
    '#title' => t('Thresholds'),
    '#description' => t('Thresholds for reporting critical alerts to Nagios.'),
  );
  $form[$group]['ultimate_cron_nagios_running_threshold'] = array(
    '#type' => 'textfield',
    '#title' => t('Running jobs count'),
    '#default_value' => variable_get('ultimate_cron_nagios_running_threshold', 50),
    '#description' => t('Issue a critical alert when more than this number of jobs are running. Default is 50.'),
  );
  $form[$group]['ultimate_cron_nagios_failed_threshold'] = array(
    '#type' => 'textfield',
    '#title' => t('Failed jobs count'),
    '#default_value' => variable_get('ultimate_cron_nagios_failed_threshold', 10),
    '#description' => t('Issue a critical alert when more than this number of jobs failed their last run. Default is 10.'),
  );
  $form[$group]['ultimate_cron_nagios_longrunning_threshold'] = array(
    '#type' => 'textfield',
    '#title' => t('Long running jobs'),
    '#default_value' => variable_get('ultimate_cron_nagios_longrunning_threshold', 0),
    '#description' => t('Issue a critical alert when more than this number of jobs are running longer than usual. Default is 0.'),
  );
  return $form;
}

/**
 * Implementation of hook_nagios_checks().
 */
function ultimate_cron_nagios_checks() {
  return ultimate_cron_nagios_functions();
}

/**
 * Implementation of drush hook_nagios_check().
 */
function ultimate_cron_nagios_check($function) {

  // We don't bother to check if the function has been enabled by the user.
  // Since this runs via drush, web security is not an issue.
  $func = $function . '_check';
  $result = $func();
  $status[$result['key']] = $result['data'];
  return $status;
}

/************** HELPER FUNCTIONS ***********************************/

/**
 * Return a list of nagios check functions 
 * @see ultimate_cron_nagios()
 */
function ultimate_cron_nagios_functions() {
  return array(
    'ultimate_cron_running' => t('Check number of currently running jobs'),
    'ultimate_cron_failed' => t('Check the number of jobs that failed last run'),
    'ultimate_cron_longrunning' => t('Check the number of jobs that are running longer than usual'),
  );
}

/**
 * Get information about running jobs - currently running or failed.
 *
 * @staticvar array $overview
 * @param string $mode Which mode to get info about; 'running' or 'errors'
 * @return int 
 */
function ultimate_cron_nagios_get_job_info($mode = 'running') {

  // Ensure valid mode
  if (!in_array($mode, array(
    'running',
    'errors',
  ))) {
    $mode = 'running';
  }
  static $overview = array();
  if (!isset($overview[$mode])) {
    $overview[$mode] = 0;

    // Get hooks and their data
    $data = _ultimate_cron_preload_cron_data();
    $hooks = ultimate_cron_get_hooks();
    $modules = array();
    foreach ($hooks as $function => $hook) {
      if (!$module || $module == $hook['module']) {
        $hook['settings'] = $data[$function]['settings'] + $hook['settings'];
        $hook['background_process'] = $data[$function]['background_process'];
        $hook['log'] = ultimate_cron_get_log($function);

        // Setup process
        if ($hook['background_process']) {
          $overview['running']++;
        }
        $log = $hook['log'];
        if (isset($log['status']) && !$log['status']) {
          $overview['errors']++;
        }
      }
    }
  }
  return $overview[$mode];
}

/*************** NAGIOS CHECK FUNCTIONS ********************************/

/**
 * Check number of running jobs.
 * 
 * @return array
 */
function ultimate_cron_running_check() {
  $running = ultimate_cron_nagios_get_job_info('running');
  $threshold = variable_get('ultimate_cron_nagios_running_threshold', 50);
  if (count($running) > $threshold) {
    $data = array(
      'status' => NAGIOS_STATUS_CRITICAL,
      'type' => 'state',
      'text' => t('@jobs currently running - it is more than @threshold', array(
        '@jobs' => $running,
        '@threshold' => $threshold,
      )),
    );
  }
  else {
    $data = array(
      'status' => NAGIOS_STATUS_OK,
      'type' => 'state',
      'text' => t('@jobs currently running', array(
        '@jobs' => $running,
      )),
    );
  }
  return array(
    'key' => 'ULTIMATE_CRON_RUNNING',
    'data' => $data,
  );
}

/**
 * Check number of jobs that failed last run.
 * 
 * @return array
 */
function ultimate_cron_failed_check() {
  $failed = ultimate_cron_nagios_get_job_info('errors');
  $threshold = variable_get('ultimate_cron_nagios_failed_threshold', 10);
  if (count($failed) > $threshold) {
    $data = array(
      'status' => NAGIOS_STATUS_CRITICAL,
      'type' => 'state',
      'text' => t('@jobs failed their last run - it is more than @threshold', array(
        '@jobs' => $failed,
        '@threshold' => $threshold,
      )),
    );
  }
  else {
    $data = array(
      'status' => NAGIOS_STATUS_OK,
      'type' => 'state',
      'text' => t('@jobs failed their last run', array(
        '@jobs' => $failed,
      )),
    );
  }
  return array(
    'key' => 'ULTIMATE_CRON_FAILED',
    'data' => $data,
  );
}

/**
 * Check number of jobs running longer than usual.
 * 
 * @return array
 * 
 * @todo Implement the logic
 */
function ultimate_cron_longrunning_check() {
  $longrunning = 0;

  // Get running jobs
  // Find out how long they have been running
  // Calculate average run time per job (over a threshold? E.g. queues run very fast if there is nothing to process)
  // If
  $threshold = variable_get('ultimate_cron_nagios_longrunning_threshold', 0);
  if ($longrunning > $threshold) {
    $data = array(
      'status' => NAGIOS_STATUS_CRITICAL,
      'type' => 'state',
      'text' => t('@jobs jobs are running longer than usual - it is more than @threshold', array(
        '@jobs' => $longrunning,
        '@threshold' => $threshold,
      )),
    );
  }
  else {
    $data = array(
      'status' => NAGIOS_STATUS_OK,
      'type' => 'state',
      'text' => t('@jobs jobs are running longer than usual', array(
        '@jobs' => $longrunning,
      )),
    );
  }
  return array(
    'key' => 'ULTIMATE_CRON_LONGRUNNING',
    'data' => $data,
  );
}

Functions

Name	Description
ultimate_cron_failed_check	Check number of jobs that failed last run.
ultimate_cron_longrunning_check	Check number of jobs running longer than usual.
ultimate_cron_nagios	Implementation of hook_nagios().
ultimate_cron_nagios_check	Implementation of drush hook_nagios_check().
ultimate_cron_nagios_checks	Implementation of hook_nagios_checks().
ultimate_cron_nagios_functions	Return a list of nagios check functions
ultimate_cron_nagios_get_job_info	Get information about running jobs - currently running or failed.
ultimate_cron_nagios_info	Implements hook_nagios_info().
ultimate_cron_nagios_settings	Implementation of hook_nagios_settings().
ultimate_cron_running_check	Check number of running jobs.