Alexey Scherbakov created IGNITE-14517:
------------------------------------------
Summary: Inconsistent behavior of the method
NetworkCluster#allMembers
Key: IGNITE-14517
URL: https://issues.apache.org/jira/browse/IGNITE-14517
Project: Ignite
Issue Type: Bug
Reporter: Alexey Scherbakov
This method reports invalid number of alive nodes when a node is stopped
gracefully.
The scenario:
# Start the cluster of 3 nodes: n1, n2, n3.
# Request n2 NetworkCluster#allMembers. It will return 3.
# Stop node n1.
# Request n2 NetworkCluster#allMembers again. It will return 3, but should
return 2.
Here is a failed test from my working branch [1][2]
I've looked into scalecube code and found out the node is stuck in the
io.scalecube.cluster.membership.MembershipProtocolImpl#membershipTable having
MemberStatus.LEAVING state.
The possible fix would avoid using cluster.members at all and instead use
events to maintain local topology. Events seem to work fine.
[1]
https://github.com/gridgain/apache-ignite-3/blob/ignite-13885/modules/network/src/integrationTest/java/org/apache/ignite/network/scalecube/ITScaleCubeNetworkClusterMessagingTest.java
[2]
https://ci.ignite.apache.org/viewLog.html?buildId=5963158&tab=buildResultsDiv&buildTypeId=ignite3_Tests_IntegrationTests&branch_ignite3_Tests=pull%2F78
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
