[
https://issues.apache.org/jira/browse/IGNITE-2356?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15859313#comment-15859313
]
Vladimir Ozerov commented on IGNITE-2356:
-----------------------------------------
[~tledkov-gridgain], my comments:
1) Please make sure that existing node is never stopped irrespective of Hadoop
configuration. Currently it is possible due to call to {{Ignition.getOrStart}},
because you do not know whether node was started bu you or not.
2) I would create separate mutex for synchronization instead of using
{{refCnts}} monitor.
3) Let's move {{HadoopIgfsInProcWithIgniteRefsCount}} to separate class to
maintain clean abstractions. By the way, do we really need this class? It seems
that all new logic fits well into existing {{HadoopIgfsInProc}} class.
> IGFS client should be able to failover in case of server crash.
> ---------------------------------------------------------------
>
> Key: IGNITE-2356
> URL: https://issues.apache.org/jira/browse/IGNITE-2356
> Project: Ignite
> Issue Type: Sub-task
> Components: IGFS
> Affects Versions: ignite-1.4
> Reporter: Vladimir Ozerov
> Assignee: Taras Ledkov
> Fix For: 2.0
>
>
> IGFS client (IgniteHadoopFileSystem) communicates IGFS over endpoint - either
> TCP or shmem.
> Only single endpoint can be specified. As such, should the server went down,
> IgntieHadoopFileSystem (either new or existing) is no longer operational.
> We need to let user specify several endpoints and failover/balance between
> them.
> Look at Hadoop HA first to get an ideas on how to configure multiple
> addresses.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
