[SAMZA-2305] Stream processor should ensure previous container is stopped during a rebalance

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Docs Text: https://docs.google.com/document/d/1gWjKsbNoRwuSexjAxgqw_1R9MaJEckMIZY0cX2_NoaY/edit?usp=sharing
• Assignee: Bharath Kumarasubramanian
• Reporter: Bharath Kumarasubramanian
• Created: 20/Aug/19 21:07
• Updated: 05/Feb/20 21:02
• Resolved: 21/Jan/20 19:23

Description

In standalone, when we have a rebalance before the container has started successfully with the previous job model, OnJobModelExpired does a no-op. It happens since the stream processor doesn’t transition to RUNNING until the container callback gets invoked. OnJobModelExpired, assumes the processor is still in the REBALANCE phase and does a no-op.

 

Often times, containers take longer time to start. Especially stateful containers that have large state to restore. In those scenarios, if a rebalance happens before the container has started successfully, processors in the quorum will run with inconsistent job model.

Issue Links

• links to: GitHub Pull Request #1213