Zoltán Borók-Nagy created HIVE-25347:
----------------------------------------
Summary: RetryingMetastoreClients cannot retry
Key: HIVE-25347
URL: https://issues.apache.org/jira/browse/HIVE-25347
Project: Hive
Issue Type: Bug
Reporter: Zoltán Borók-Nagy
Even if the connection is broken, RetryingMetastoreClient doesn't reconnect to
HMS when the followings are true:
* metastore.client.socket.lifetime has default value 0, which means "infinite
lifetime"
* Non-retryable method is invoked, e.g.
['lock()'|https://github.com/apache/hive/blob/a75b8680214c490be6b092b4fa5f790ae7c2e5ce/standalone-metastore/metastore-common/src/main/java/org/apache/hadoop/hive/metastore/IMetaStoreClient.java#L3413]
If we only invoke non-retryable methods then RetryingMetastoreClient will never
reconnect to HMS, therefore all the RPCs will fail.
It's because RetryingMetastoreClient only reconnects on the second attempt
(which will never happen for non-retryable methods), or if the connection
lifetime has expired (by default connections don't expire):
https://github.com/apache/hive/blob/a75b8680214c490be6b092b4fa5f790ae7c2e5ce/standalone-metastore/metastore-common/src/main/java/org/apache/hadoop/hive/metastore/RetryingMetaStoreClient.java#L183
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
