On Aug 13, 2010, at 10:19 AM, Pradeep Kamath wrote: > Hi, > Reading through http://wiki.apache.org/hadoop/Hive/Locking, it appears that > locking will be implemented using Zookeeper in the Query language - just > wanted to confirm that the metastore APIs are not being modified and that > there is no information about existing locks in the metastore - is this > correct? If so, won't a thrift API call circumvent existing locks and gain > read/write access? Did I miss something?
That's correct. The locking is all from the Hive client side. We want the locks to be released automatically if a client dies, and doing it from within the metastore thrift server wouldn't give us that. You can also circumvent the locks by going directly to HDFS through any non-Hive means...we can't stop that. JVS
