Lucas Bradstreet
Lucas Bradstreet
This fix prevents poll! from being called once the plugin is marked with completed?. The downside will be when completed? is an expensive call, and thus this fix should be...
In cases where max-peers is already allocated on all tasks, more than one job should be allowed to run. We received a report that this is not the case.
State machine should back-off until condition has passed, but it should do so in the task state machine itself, where it can heartbeat.
It'd be great if we could report metrics from a coordinator, as the coordinator is an important part of the system.
Ensures heartbeating continues during the recovery process. TODO: add timeout to recover process by recording System/nanoTime at start so that peers can't get stuck forever.
As we perform two checkpoints, we are only recording checkpoint_size_Value for the output task checkpoint, and not for the windowed checkpoint_size_Value.
This line should return the state without advancing: https://github.com/onyx-platform/onyx/blob/master/src/onyx/peer/read_batch.clj#L34 allowing the normal state machine idle cycle to take place.