[ML] Job in Index Feature: TransportJobTaskAction no longer detects unknown jobs

[davidkyle]

TransportJobTaskAction checks the cluster state for the job configuration before looking for the job's persistent task. If the config cannot be found a missing jobs exception is thrown so all actions that extend TransportJobTaskAction automatically get job exists checking. With the change to move the config the job check has to be removed so in the case of an unknown job the error will be cannot perform action X as the job is not open, then of course opening the job would return the job not found error.

Discuss
Error messages are not part of the rest specification so is this considered a breaking change. Is the new behaviour acceptable or can anyone see a workaround?

The actions that extend TransportJobTaskAction and hence are affected are:

• TransportFlushJobAction
• TransportForecastJobAction
• TransportPostDataAction
• TransportUpdateProcessAction (internal)
• TransportPersistJobAction (internal)
• TransportKillProcessAction (internal)

[elasticmachine]

Pinging @elastic/ml-core

[droberts195]

I think it would be nice if the 3 actions that are designed for external use (TransportFlushJobAction, TransportForecastJobAction and TransportPostDataAction) would correctly report the job didn't exist.

I think there is a way to do this. TransportJobTaskAction could take an optional JobConfigProvider. The 3 internal actions would pass null for this; the 3 external actions would pass the one obtained by dependency injection. Then, on finding the job task didn't exist, instead of immediately returning a "job not open" error message, the TransportJobTaskAction could try to get the job config from the JobConfigProvider if one was provided. Any exception produced by that would be passed to the final listener, or if getting the config was successful then the "job not open" exception would be raised as now.