[ https://issues.apache.org/jira/browse/GEODE-1761?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15428606#comment-15428606 ]
ASF subversion and git services commented on GEODE-1761: -------------------------------------------------------- Commit cea5535cf2e0de1ee1ef8ea902ea44a2f48b4fd5 in incubator-geode's branch refs/heads/develop from [~bschuchardt] [ https://git-wip-us.apache.org/repos/asf?p=incubator-geode.git;h=cea5535 ] GEODE-1761 Clients don't fail back when servers are bounced Servers will send a refresh hint to clients if they detect that a request had to be send to a different server who owned the primary bucket affected by the operation. Clients should always refresh when this happens unless they have connection-pool size constraints that force them to use non-optimal servers. Client-side operation classes have been modified to initiate the refresh. I've added code in the meta-data service class to avoid performing multiple concurrent refreshes on the same region. On the server-side I've cleaned up some of the network-hop detection code to stop using hard-coded integers and to consolidate some of the code that resets the ThreadLocals being used to record network- hops detected. > Clients don't fail back to servers containing non-redundant partitioned > regions when they are bounced and rebalanced > -------------------------------------------------------------------------------------------------------------------- > > Key: GEODE-1761 > URL: https://issues.apache.org/jira/browse/GEODE-1761 > Project: Geode > Issue Type: Bug > Components: client/server > Reporter: Bruce Schuchardt > > Clients don't fail back to servers containing non-redundant partitioned > regions when they are bounced and rebalanced. > When one of the servers is stopped, the client switches to the server(s) > containing the new primary. When the server is restarted and rebalanced > (which recovers primaryness), the clients do not switch back even though the > server continually sends the metadata refresh byte on each operation. > If the partitioned region has a redundant copy, clients do switch back. -- This message was sent by Atlassian JIRA (v6.3.4#6332)