[HELIX-281] Need to optimize the scenario that multiple Message change can stop external view update

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Assignee: Kanak Biscuitwala
• Reporter: Shi Lu
• Created: 25/Oct/13 21:27
• Updated: 22/Jan/14 23:22
• Resolved: 22/Jan/14 23:22

Description

We have seen the following behavior in the relay cluster startup:

controller is full time doing onMessage() callback, although all the current state has already been updated. As all state transitions are done, all the Message update callback in controller side becomes NO-OP but stops the current state from being updated. As there are so many onMessage change callbacks, ExternalView does not get updated in hours. This is going to be a blocking perf issue in big clusters.

One work-around is to switch the controller if the current controller is in this Message update loop.

Issue Links

• is duplicated by: HELIX-349 Coalesce different types of callbacks
• relates to: HELIX-349 Coalesce different types of callbacks