0

We have clustered Quartz scheduler runner on a couple of application nodes. The application nodes need to be updated, and for high-availability reasons, the update is done as rolling update.

Together with the update, we need to add a new job, and that job needs to start running immediately - i.e. it can't wait until all nodes have been updated. The problem is that I can't control which node will run the new job, and if one of the old nodes runs the job, the job instantiation will faill (with a ClassNotFoundException), the trigger will be set to the state ERROR and the job won't run again.

One solution for this problem would be to do two updates: one to add the class in all nodes, and one to add the trigger. The main reason against this approach is that our ops procedures don't support this.

So is there also a way to schedule the new job and make it run reliably with a single update?

Community
  • 1
  • 1
oberlies
  • 11,503
  • 4
  • 63
  • 110

1 Answers1

1

I just tried it and it turned out that Quartz gets a ClassCastException while trying acquire the trigger. The exception is wrapped into a JobPersistenceException and the trigger is left in WAITING state.

So, although this could cause an error log entry in one of the old nodes, Quartz doesn't leave the trigger in a non-working state.

Sascha Scholz
  • 26
  • 1