webhcat was built to be stateless and should work with multiple instances. Hive's thrift metastore service (used by HCatalog) is mostly stateless. If you are running it with the Kerberos security turned on then the security credentials are kept in memory. This means that sessions cannot be switched from one server to the other. When running in unsecure mode I am not aware of any state issues.
Alan. On Apr 18, 2013, at 10:27 AM, Nasron Cheong wrote: > We are looking into running multiple instances of HCatalog/WebHCat on > different servers to achieve some sort of redundancy. From some tests it > appears that multiple nodes seem to be synchronized through Zookeeper, and > since data is stored in HDFS each node doesn't rely on its own state. > > So if a HCatalog node goes down we can just point to the other and continue. > > If we did go with this approach, are there any gotchas to look out for? > Documentation on similar setup(s) seem lacking. > > - Ron
