- Created FailureDetectorStrategy base trait.
- Created FailureDetectorPuppetStrategy.
- Created AccrualFailureDetectorStrategy.
- Created two versions of LeaderDowningNodeThatIsUnreachableMultiJvmSpec
- LeaderDowningNodeThatIsUnreachableWithFailureDetectorPuppet
- LeaderDowningNodeThatIsUnreachableWithAccrualFailureDetector
- Added AccrualFailureDetectorStrategy to all the remaining tests - will be split up into two versions shortly.
Signed-off-by: Jonas Bonér <jonas@jonasboner.com>
* Failure detector didn't use hearbeat 1 and 2
* Included heartbeat 2 in ordinary stats
* For heartbeat 1 use guess stats,
important so that connections with only one heartbeat
becomes unavailble, the guess corresponds to 1 second interval
which results in phi > 8 after 18 seconds
* Improved AccrualFailureDetectorSpec
- Renamed all 'frequency' to 'interval'
- Split up NodeJoinAndUpSpec and into NodeJoinSpec and NodeUpSpec.
- Split up MembershipChangeListenerJoinAndUpSpec and into MembershipChangeListenerJoinSpec and MembershipChangeListenerUpSpec.
- Added utility method 'startClusterNode()'
- Fixed race in register listener and telling node to leave
- Removed 'after' blocks
- Cleaned up unused code
- Improved comments
Signed-off-by: Jonas Bonér <jonas@jonasboner.com>
* It's no problem using after, but scalatest will output the test method
as completed (green) before running after, so it looks confusing in the logs
* Using unique barrier names adds extra traceability in case of failures.
* The only allowed member states for a leader are up or leaving
* Added above check in assertLeader so that we always verify that
* More usage of assertLeader in the tests
