[HDFS-917] Write pipeline heartbeat interval should be determined by client timeout, not DN

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 0.22.0
• Component/s: hdfs-client
• Hadoop Flags: Incompatible change
• Assignee: Unassigned
• Reporter: Todd Lipcon
• Created: 23/Jan/10 01:42
• Updated: 04/Feb/10 00:39

Description

For the dfs client timeouts to be more effective, they should be communicated to the DN on the start of a write, so that the DN can send heartbeats at a fast enough rate. Currently, the pipeline heartbeats are half the datanode's socket timeout, so a DFS client with a very low timeout will fail if it's not continuously writing application data.

Issue Links

• relates to: HDFS-915 Hung DN stalls write pipeline for far longer than its timeout