Triggering action alerts across jobs and queries

You can trigger action alerts, that are defined by you, across your workload applications, jobs, and queries whilst they are running with the Cloudera Observability On-Premises Auto Actions feature. When a workload application, job, or query matches the action's defined threshold value, the auto action event is triggered. For example, you may have a scenario where too much memory is being allocated to specific jobs and you would like to take an action before a problem occurs, such as avoiding memory exhaustion. In this case, you can create an auto action that triggers a notification alert when a job is identified as having an over-allocation of memory. You can then either manually take steps to alleviate the problem or include the Kill action option that stops the job in question.

Considerations and Limitations

The following describes consideration decisions and limitations when using Auto Actions:
  • Terminating a workload application, job, or query could impact other workloads. Especially when another workload is dependent on the results of the terminated workload application, job, or query. Before triggering a Kill type action, Cloudera recommends using the Notification only action alert until you have verified that no issues will arise.
  • By default, the Cloudera Observability On-Premises UI limits the amount of displayed audit events to 500 and sorts them in ascending order (newest time stamp first). To display older audit events, change the date range duration and/or the time range duration from the time-range list on the Auto Actions Events page.
  • Too Fast To Collect: The minimum polling interval is one minute. If you have jobs or queries that overlap or start before the minimum polling interval is completed there may not be enough time for Cloudera Observability On-Premises to evaluate your auto action’s definition.

    For example, if Cloudera Observability On-Premises starts polling at 1:00:00 PM and polling finishes by 1:00:10 PM (10 seconds) and then a job starts at 1:00:12 PM and finishes before 1:01:00 PM, there is not enough of a time lapse for Cloudera Observability On-Premises to evaluate and trigger your action alert.

  • Too Fast to Kill: Under normal conditions the evaluation and invocation phases of an auto action is within the span of one minute. If you have jobs or queries whose run time is less than one minute, Cloudera Observability On-Premises may complete the evaluation phase but not have time to complete the invocation phase, such as terminating the job. Depending on the context of your auto action, this may or may not be an issue. But if, for example, you have a workload cluster that is dedicated for specific jobs and a rogue job is run before the action is triggered, then this could be an issue