Cleaning up metastore clients

[Raghu Murthy <rm...@facebook.com>]

While reviewing HIVE-736, I noticed that there are two metastore clients -
HiveMetaStoreClient and IMetaStoreClient. I thought we cleaned that up after
we got rid of the file based metastore.

It seems like Hive.java, HiveMetaStoreClient.java, and IMetaStoreClient.java
have very similar method calls. It is not clear to me if these classes need
to exist at all given that thrift already provides a client interface. Why
are we adding intelligence to the thrift clients? Does it make sense to just
create a simple metastore client (like HiveClient for the hiveserver) and
let all intelligence be in HMSHandler?

raghu

Re: Cleaning up metastore clients

[Prasad Chakka <pc...@facebook.com>]

I don't remember why I left them out when I was removing the file store. You can try doing that, there shouldn't be any problems.

I  also feel that hive.ql.metadata package should be moved to hive.metadata or hive.metadata.ql so that hive_metastore.jar can be used independent of Hive.

Prasad


________________________________
From: Raghu Murthy <rm...@facebook.com>
Reply-To: <hi...@hadoop.apache.org>
Date: Sat, 8 Aug 2009 14:31:57 -0700
To: <hi...@hadoop.apache.org>
Subject: Cleaning up metastore clients

While reviewing HIVE-736, I noticed that there are two metastore clients -
HiveMetaStoreClient and IMetaStoreClient. I thought we cleaned that up after
we got rid of the file based metastore.

It seems like Hive.java, HiveMetaStoreClient.java, and IMetaStoreClient.java
have very similar method calls. It is not clear to me if these classes need
to exist at all given that thrift already provides a client interface. Why
are we adding intelligence to the thrift clients? Does it make sense to just
create a simple metastore client (like HiveClient for the hiveserver) and
let all intelligence be in HMSHandler?

raghu