[ https://issues.apache.org/jira/browse/HIVE-10507?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14518154#comment-14518154 ]
Hari Sankar Sivarama Subramaniyan commented on HIVE-10507: ---------------------------------------------------------- The test failures are unrelated to this change. Thanks Hari > Expose RetryingMetastoreClient to other external users of metastore client > like Flume and Storm. > ------------------------------------------------------------------------------------------------- > > Key: HIVE-10507 > URL: https://issues.apache.org/jira/browse/HIVE-10507 > Project: Hive > Issue Type: Bug > Reporter: Hari Sankar Sivarama Subramaniyan > Assignee: Hari Sankar Sivarama Subramaniyan > Attachments: HIVE-10507.1.patch > > > HiveMetastoreClient is now being relied upon by external clients like Flume > and Storm for streaming. > When the thrift connection between MetaStoreClient and the meta store is > broken (due to intermittent network issues or restarting of metastore) the > Metastore does not handle the connection error and automatically re-establish > the connection. Currently the client process needs to be restarted to > re-establish the connection. > The request here is consider supporting the following behavior: For each API > invocation on the MetastoreClient, it should try to restablish the connection > (if needed) once. And if that does not work out then throw a specific > exception indicating the same. The client could then handle the issue by > retrying the same API after some delay. By catching the specific connection > exception, the client could decide how many times to retry before aborting. > Hive does this internally using RetryingMetastoreClient. This jira is suppose > to expose this mechanism to other users of that interface. This is useful for > users of this interface, and from metastore HA point of view. -- This message was sent by Atlassian JIRA (v6.3.4#6332)