[HDFS-12861] Track speed in DFSClient

• Type: Improvement
• Status: Patch Available
• Priority: Major
• Resolution: Unresolved
• Assignee: María Fernanda Borge
• Reporter: Íñigo Goiri
• Created: 28/Nov/17 03:02
• Updated: 15/Dec/20 13:06

Description

Sometimes we get slow jobs because of the access to HDFS. However, is hard to tell what is the actual speed. We propose to add a log line with something like:

2017-11-19 09:55:26,309 INFO [main] hdfs.DFSClient: blk_1107222019_38144502 READ 129500B in 7ms 17.6MB/s
2017-11-27 19:01:04,141 INFO [DataStreamer for file /hdfs-federation/stats/2017/11/27/1511838000000.json] hdfs.DFSClient: blk_1135792057_86833357 WRITE 131072B in 10ms 12.5MB/s
2017-11-27 19:01:14,219 INFO [DataStreamer for file /hdfs-federation/stats/2017/11/27/1511838000000.json] hdfs.DFSClient: blk_1135792069_86833369 WRITE 131072B in 12ms 10.4MB/s
2017-11-27 19:01:24,282 INFO [DataStreamer for file /hdfs-federation/stats/2017/11/27/1511838000000.json] hdfs.DFSClient: blk_1135792081_86833381 WRITE 131072B in 11ms 11.4MB/s
2017-11-27 19:01:34,330 INFO [DataStreamer for file /hdfs-federation/stats/2017/11/27/1511838000000.json] hdfs.DFSClient: blk_1135792093_86833393 WRITE 131072B in 11ms 11.4MB/s
2017-11-27 19:01:44,408 INFO [DataStreamer for file /hdfs-federation/stats/2017/11/27/1511838000000.json] hdfs.DFSClient: blk_1135792105_86833405 WRITE 131072B in 11ms 11.4MB/s

Issue Links

• relates to: HDFS-13053 Track time to process packet in Datanode