[HDFS-15599] RBF: Add API to expose resolved destinations (namespace) in Router

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Assignee: Qifan Shi
• Reporter: Fengnan Li
• Created: 25/Sep/20 00:07
• Updated: 19/May/21 20:14

Description

We have seen quite often requests like where a path in Router is actually pointed. Two main use cases are:

1) Calculate the HDFS capacity usage allocation of all Hive tables, whose have onboarded to Router.

2) A failure prevention method for cross-cluster rename. First check the source HDFS location and dest HDFS location, and then issue a distcp cmd if possible to avoid the Exception.

Inside Router, the function getLocationsForPath does the work but it is internal only and not visible to Clients.

RouterAdmin has getMountTableEntries but this is a cast of Mount table without any resolving.

 

We are proposing adding such an API, and there are two ways:

1) Adding this API in RouterRpcServer, which requires the change in ClientNameNodeProtocol to include this new API. 

2) Adding this API in RouterAdminServer, which requires the a protocol between Client and the admin server.

 

There is one existing resolvePath in FileSystem which can be used to implement this call from client side.