[jboss-jira] [JBoss JIRA] (WFLY-11535) Cluster fails to merge if instances started simultaneously

[Ian Rodgers (Jira)]

    [ https://issues.jboss.org/browse/WFLY-11535?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13693032#comment-13693032 ] 

Ian Rodgers edited comment on WFLY-11535 at 2/8/19 8:09 AM:
------------------------------------------------------------

Have successfully recreated this issue, with Keycloak taken out of the equation.

We are now using the Wildfly 14 official docker image.

Attached is the latest standalone-ha file, the only change being a tweak to use JDBC_PING within JGroups as per our original config. It is the standalone file 

If the 4 instances are started in a staggered fashion then they correctly form one cluster of 4.
If they are started simultaneously then they form 4 individual clusters.

This was not an issue when we had Keycloak 4.5 running on Wildfly 13.

Incidentally we noticed there was a modcluster definition in the Wildfly distribution standalone file, but not our original Keycloak standalone. We don't think it is necessary because we are using tcp rather than multicast. In any case we get the same outcome with and without modcluster present. 


was (Author: ianrodgers):
Have successfully recreated this issue, with Keycloak taken out of the equation.

We are now using the Wildfly 14 official docker image.

Attached is the latest standalone-ha file, the only change being a tweak to use JDBC_PING within JGroups as per our original config.

If the 4 instances are started in a staggered fashion then they correctly form one cluster of 4.
If they are started simultaneously then they form 4 individual clusters.

This was not an issue when we had Keycloak 4.5 running on Wildfly 13.

Incidentally we noticed there was a modcluster definition in the standalone file and thought that this might be connected to the problem. We get the same outcome with and without modcluster present. 

> Cluster fails to merge if instances started simultaneously
> ----------------------------------------------------------
>
>                 Key: WFLY-11535
>                 URL: https://issues.jboss.org/browse/WFLY-11535
>             Project: WildFly
>          Issue Type: Bug
>          Components: Clustering
>    Affects Versions: 14.0.0.Final
>         Environment: Keycloak 4.6.0 official docker image, AWS ECS cluster
>            Reporter: Ian Rodgers
>            Assignee: Paul Ferraro
>            Priority: Major
>         Attachments: standalone-ha.xml, standalone-ha.xml
>
>
> Our Keycloak docker cluster has four instances, clustered using Jgroups/Infinispan as per the standalone-ha.xml. If you start them all simultaneously the "Receive new cluster" logs indicate four separate clusters, each with a single member. They never get merged into the proper single cluster of four members. It seems to be the merging that has changed. The application then fails (we are not using sticky sessions, and each member is ignorant of the sessions on the other members).
> We can only start the cluster by first starting one instance, then when it is running, starting the other three. The logs then indicate the creation of a single cluster which subsequent instances join.
> This is consistent behaviour, and when we revert back to v.4.5.0, the issue goes away, Unfortunately we need 4.6.0 for an important fix.
> On 4.5.0 we get the message "Received new, MERGED cluster view for channel ejb: MergeView::" when it detects a number of subgroups to merge. This never appears in 4.6.0.



--
This message was sent by Atlassian Jira
(v7.12.1#712002)