[HIVE-13160] HS2 unable to load UDFs on startup when HMS is not ready

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.2.1
• Fix Version/s: 2.1.0
• Component/s: HiveServer2
• Assignee: Aihua Xu
• Reporter: Eric Lin
• Created: 26/Feb/16 00:02
• Updated: 09/Sep/16 15:03
• Resolved: 01/Mar/16 14:30

Description

The error looks like this:

2016-02-18 14:43:54,251 INFO  hive.metastore: [main]: Trying to connect to metastore with URI thrift://host-10-17-81-201.coe.cloudera.com:9083
2016-02-18 14:48:54,692 WARN  hive.metastore: [main]: Failed to connect to the MetaStore Server...
2016-02-18 14:48:54,692 INFO  hive.metastore: [main]: Waiting 1 seconds before next connection attempt.
2016-02-18 14:48:55,692 INFO  hive.metastore: [main]: Trying to connect to metastore with URI thrift://host-10-17-81-201.coe.cloudera.com:9083
2016-02-18 14:53:55,800 WARN  hive.metastore: [main]: Failed to connect to the MetaStore Server...
2016-02-18 14:53:55,800 INFO  hive.metastore: [main]: Waiting 1 seconds before next connection attempt.
2016-02-18 14:53:56,801 INFO  hive.metastore: [main]: Trying to connect to metastore with URI thrift://host-10-17-81-201.coe.cloudera.com:9083
2016-02-18 14:58:56,967 WARN  hive.metastore: [main]: Failed to connect to the MetaStore Server...
2016-02-18 14:58:56,967 INFO  hive.metastore: [main]: Waiting 1 seconds before next connection attempt.
2016-02-18 14:58:57,994 WARN  hive.ql.metadata.Hive: [main]: Failed to register all functions.
java.lang.RuntimeException: Unable to instantiate org.apache.hadoop.hive.ql.metadata.SessionHiveMetaStoreClient
        at org.apache.hadoop.hive.metastore.MetaStoreUtils.newInstance(MetaStoreUtils.java:1492)
        at org.apache.hadoop.hive.metastore.RetryingMetaStoreClient.<init>(RetryingMetaStoreClient.java:64)
        at org.apache.hadoop.hive.metastore.RetryingMetaStoreClient.getProxy(RetryingMetaStoreClient.java:74)
        at org.apache.hadoop.hive.ql.metadata.Hive.createMetaStoreClient(Hive.java:2915)
.......
016-02-18 14:58:57,997 INFO  hive.metastore: [main]: Trying to connect to metastore with URI thrift://host-10-17-81-201.coe.cloudera.com:9083
2016-02-18 15:03:58,094 WARN  hive.metastore: [main]: Failed to connect to the MetaStore Server...
2016-02-18 15:03:58,095 INFO  hive.metastore: [main]: Waiting 1 seconds before next connection attempt.
2016-02-18 15:03:59,095 INFO  hive.metastore: [main]: Trying to connect to metastore with URI thrift://host-10-17-81-201.coe.cloudera.com:9083
2016-02-18 15:08:59,203 WARN  hive.metastore: [main]: Failed to connect to the MetaStore Server...
2016-02-18 15:08:59,203 INFO  hive.metastore: [main]: Waiting 1 seconds before next connection attempt.
2016-02-18 15:09:00,203 INFO  hive.metastore: [main]: Trying to connect to metastore with URI thrift://host-10-17-81-201.coe.cloudera.com:9083
2016-02-18 15:14:00,304 WARN  hive.metastore: [main]: Failed to connect to the MetaStore Server...
2016-02-18 15:14:00,304 INFO  hive.metastore: [main]: Waiting 1 seconds before next connection attempt.
2016-02-18 15:14:01,306 INFO  org.apache.hive.service.server.HiveServer2: [main]: Shutting down HiveServer2
2016-02-18 15:14:01,308 INFO  org.apache.hive.service.server.HiveServer2: [main]: Exception caught when calling stop of HiveServer2 before retrying start
java.lang.NullPointerException
        at org.apache.hive.service.server.HiveServer2.stop(HiveServer2.java:283)
        at org.apache.hive.service.server.HiveServer2.startHiveServer2(HiveServer2.java:351)
        at org.apache.hive.service.server.HiveServer2.access$400(HiveServer2.java:69)
        at org.apache.hive.service.server.HiveServer2$StartOptionExecutor.execute(HiveServer2.java:545)

And then none of the functions will be available for use as HS2 does not re-register them after HMS is up and ready.

This is not desired behaviour, we shouldn't allow HS2 to be in a servicing state if function list is not ready. Or, maybe instead of initialize the function list when HS2 starts, try to load the function list when each Hive session is created. Of course we can have a cache of function list somewhere for better performance, but we would better decouple it from class Hive.