[KAFKA-5879] Controller should read the latest IsrChangeNotification znodes when handling IsrChangeNotification event

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.0.0
• Assignee: Dong Lin
• Reporter: Dong Lin
• Reviewer: Ismael Juma
• Created: 13/Sep/17 00:27
• Updated: 05/Oct/17 00:39
• Resolved: 05/Oct/17 00:38

Description

Currently controller can be very inefficient in handling IsrChangeNotification event because it may need to access znode O(n^2) times to handle O( n ) IsrChangeNotification znodes.

For example, say there are 100 IsrChangeNotification nodes added to the zookeeper. This will generate 100 IsrChangeNotification events with children [1], [1, 2], [1, 2, 3], ... [1, 2, .. 100]. Let's say the controller now needs to handle the IsrChangeNotification event with children [1, 2, ... 100]. Controller will read zookeeper 100 times, delete these 100 znodes, which further generates 100 IsrChangeNotification events with children count from 0 to 99.

The main cause of the problem is that, controller will attempt to access zookeeper n times, where n is the count of the children at the time the IsrChangeNotification event is generated, even though there is no IsrChangeNotification znodes in the zookeeper.

Issue Links

• links to: GitHub Pull Request #3840