[jira] [Closed] (LENS-1286) Handle Lens server restart

["Divakar Tomar (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/LENS-1286?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Divakar Tomar closed LENS-1286.
-------------------------------

verified

> Handle Lens server restart
> --------------------------
>
>                 Key: LENS-1286
>                 URL: https://issues.apache.org/jira/browse/LENS-1286
>             Project: Apache Lens
>          Issue Type: Sub-task
>          Components: server
>            Reporter: Lavkesh Lahngir
>            Assignee: Lavkesh Lahngir
>             Fix For: 2.7
>
>         Attachments: restart.patch
>
>
> Scheduled events which are generated by AlarmService for scheduling are processed by a SchedulerEventListener thread.
> These steps are performed after receiving a scheduled event:
> 1. Get the job stored in the SchedulerDAO
> 2. Create a session with job owner and fill the session conf.
> 3. Create the instance if it is not a rerun otherwise get the instance from the store. 
> 4. Create a Run instance. 
> 5. Store instance (if not present) and run to the store.
> 6. Launch the query.
> 7. Update the run with the latest information.
> 8. If there is a failure in query launch then update the run status to be failed.
> While EventListener is processing the event, it can be closed during the shutdown of scheduler service. Closing the thread can cause operations to be incomplete.
> There are these stages where the thread can be interrupted:
> 1. Shutdown just after the alarm Event is fired and received:
> In this case, we won't have any entry for the instance run in the store. This will cause a hole in the list of instances.
> 2. Shutdown before query launch. 
> In this case, at least we will have an entry in the table and status as NEW.
> 3. Shutdown after query launch. 
> In this case, the status won't be updated, and even if the query is launched we will have no information about the handle. When this happens the 
> scheduled status will remain NEW
> Of all the scenarios above, the first one is kind of annoying because there is no way to generate the event from Alarm Service again if it is already triggered. 
> Second and third scenarios can be handled manually by killing the instance and re-running it.
> Apart from the scheduler event listener thread, we have another thread which processes the query end event which works as follows:
> 1. Receive the event and check if this is a scheduled query.
> 2. Check the final status of the query and set it as the instance status.
> If a shutdown happens after receiving the query end event, it will cause the status to remain as RUNNING.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)