[infinispan-issues] [JBoss JIRA] (ISPN-6740) Client topologies not updated when cache topology loaded from persistent state

[Galder Zamarreño (JIRA)]

     [ https://issues.jboss.org/browse/ISPN-6740?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Galder Zamarreño updated ISPN-6740:
-----------------------------------
    Description: 
Infinispan Caches now support storing persistent views. When these are loaded, these might be loaded with topology ID 0:

{code}
2016-05-31 10:20:04,254 INFO  [org.infinispan.globalstate.impl.GlobalStateManagerImpl] (MSC service thread 1-3) 
ISPN000389: Loaded global state, version=9.0.0-SNAPSHOT timestamp=2016-05-30T12:03:33.822Z
....
2016-05-31 10:20:07,867 TRACE [org.infinispan.statetransfer.StateTransferManagerImpl] (MSC service thread 1-6) 
Installing new cache topology CacheTopology{id=0, rebalanceId=0, currentCH=DefaultConsistentHash{ns=20, owners = (3)[node5: 7+5, node4: 7+7, node6: 6+8]}, 
pendingCH=null, unionCH=null, actualMembers=[node5, node4, node6], persistentUUIDs=[
bb76729d-2b30-4e54-8108-4ac1db9a04cf, bb76729d-2b30-4e54-8108-4ac1db9a04cf, bb76729d-2b30-4e54-8108-4ac1db9a04cf]} on cache default
{code}

If there's no further view changes, the topology ID will remain 0. When a Hot Rod client first connects, it sends its view topology as 0 so that it receives a newly installed topology, but if the topology is already 0 in the server, the server won't send the installed topology, even though it should be newer than having no topology.

We should start numbering topologies in server starting from 1 instead. That avoids this issue. This is easier than forcing clients to send -1 as initial topology because the topology ID is currently defined as VInt that can only 0 or positive number.

Also, some extra log messages indicating that the cache topology installed comes from persisted state would be handy for debugging.

  was:
Infinispan Caches now support storing persistent views. When these are loaded, these might be loaded with topology ID 0:

{code}
2016-05-31 10:20:04,254 INFO  [org.infinispan.globalstate.impl.GlobalStateManagerImpl] (MSC service thread 1-3) ISPN000389: Loaded global state, version=9.0.0-SNAPSHOT timestamp=2016-05-30T12:03:33.822Z
....
2016-05-31 10:20:07,867 TRACE [org.infinispan.statetransfer.StateTransferManagerImpl] (MSC service thread 1-6) Installing new cache topology CacheTopology{id=0, rebalanceId=0, currentCH=DefaultConsistentHash{ns=20, owners = (3)[node5: 7+5, node4: 7+7, node6: 6+8]}, pendingCH=null, unionCH=null, actualMembers=[node5, node4, node6], persistentUUIDs=[bb76729d-2b30-4e54-8108-4ac1db9a04cf, bb76729d-2b30-4e54-8108-4ac1db9a04cf, bb76729d-2b30-4e54-8108-4ac1db9a04cf]} on cache default
{code}

If there's no further view changes, the topology ID will remain 0. When a Hot Rod client first connects, it sends its view topology as 0 so that it receives a newly installed topology, but if the topology is already 0 in the server, the server won't send the installed topology, even though it should be newer than having no topology.

We should start numbering topologies in server starting from 1 instead. That avoids this issue. This is easier than forcing clients to send -1 as initial topology because the topology ID is currently defined as VInt that can only 0 or positive number.

Also, some extra log messages indicating that the cache topology installed comes from persisted state would be handy for debugging.



> Client topologies not updated when cache topology loaded from persistent state
> ------------------------------------------------------------------------------
>
>                 Key: ISPN-6740
>                 URL: https://issues.jboss.org/browse/ISPN-6740
>             Project: Infinispan
>          Issue Type: Bug
>          Components: Remote Protocols, State Transfer
>    Affects Versions: 9.0.0.Alpha2, 8.2.2.Final
>            Reporter: Galder Zamarreño
>            Assignee: Dan Berindei
>             Fix For: 9.0.0.Alpha3, 9.0.0.Final
>
>
> Infinispan Caches now support storing persistent views. When these are loaded, these might be loaded with topology ID 0:
> {code}
> 2016-05-31 10:20:04,254 INFO  [org.infinispan.globalstate.impl.GlobalStateManagerImpl] (MSC service thread 1-3) 
> ISPN000389: Loaded global state, version=9.0.0-SNAPSHOT timestamp=2016-05-30T12:03:33.822Z
> ....
> 2016-05-31 10:20:07,867 TRACE [org.infinispan.statetransfer.StateTransferManagerImpl] (MSC service thread 1-6) 
> Installing new cache topology CacheTopology{id=0, rebalanceId=0, currentCH=DefaultConsistentHash{ns=20, owners = (3)[node5: 7+5, node4: 7+7, node6: 6+8]}, 
> pendingCH=null, unionCH=null, actualMembers=[node5, node4, node6], persistentUUIDs=[
> bb76729d-2b30-4e54-8108-4ac1db9a04cf, bb76729d-2b30-4e54-8108-4ac1db9a04cf, bb76729d-2b30-4e54-8108-4ac1db9a04cf]} on cache default
> {code}
> If there's no further view changes, the topology ID will remain 0. When a Hot Rod client first connects, it sends its view topology as 0 so that it receives a newly installed topology, but if the topology is already 0 in the server, the server won't send the installed topology, even though it should be newer than having no topology.
> We should start numbering topologies in server starting from 1 instead. That avoids this issue. This is easier than forcing clients to send -1 as initial topology because the topology ID is currently defined as VInt that can only 0 or positive number.
> Also, some extra log messages indicating that the cache topology installed comes from persisted state would be handy for debugging.



--
This message was sent by Atlassian JIRA
(v6.4.11#64026)