[HDFS-7680] Support dataset-specific choice of short circuit implementation

• Type: Sub-task
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.0.0-alpha1
• Component/s: datanode, hdfs-client
• Assignee: Joe Pallas
• Reporter: Joe Pallas
• Created: 26/Jan/15 20:08
• Updated: 12/May/16 18:14

Description

As described in HDFS-5194, the current support for short circuit reading is tightly coupled to the default Dataset implementation. Since alternative implementations of the FsDatasetSpi may use a different short circuit pathway, there needs to be a way for the client to acquire the right kind of BlockReader.

Reviewing some considerations:
Today, there is only one dataset per datanode (with multiple volumes). Is that likely to change?
Can there be multiple datanodes local to a client? (definition of "local" might depend on dataset implementation)
Is it okay to assume that the client and datanode share configuration? More broadly, how should the client discover the appropriate short-circuit implementation?

Issue Links

• duplicates: HDFS-8924 Add pluggable interface for reading replicas in DFSClient