[SLING-5195] HeartbeatHandler should check for topology changes independently of long-during session.saves

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Discovery Impl 1.1.8
• Fix Version/s: Discovery Impl 1.2.0
• Assignee: Stefan Egli
• Reporter: Stefan Egli
• Created: 26/Oct/15 12:50
• Updated: 12/Nov/15 11:20
• Resolved: 02/Nov/15 14:47

Description

When running discovery.impl in a DocumentNodeStore cluster that is under heavy load and thus produces large delays, the HeartbeatHandler can sometimes be blocked in session.save for non-trivial amounts of time - longer than the heartbeat timeout for example. This has the effect that both the lastHeartbeat property is no longer updated but also that the topology is not checked and updated towards the listeners. These two facts should be separated within the HeartbeatHandler though.

Issue Links

• is blocked by: SLING-5200 introduce simulation of repository delays in discovery tests
• is related to: SLING-5216 Concurrent creation of voting can lead to continuous re-voting
• is related to: SLING-4640 Possibility of duplicate leaders w/discovery.impl on eventually consistent repo
• is related to: SLING-5243 discovery.impl.cluster.SingleInstanceTest.testBootstrap failed on jenkins