David Mollitor created HDFS-14641:
-------------------------------------
Summary: Possible Intermediate Block Report Race Condition
Key: HDFS-14641
URL: https://issues.apache.org/jira/browse/HDFS-14641
Project: Hadoop HDFS
Issue Type: Bug
Affects Versions: 3.2.0
Reporter: David Mollitor
I am doing some playing around in the DataNode and if I set a small delay
(100ms) before sending the intermediate block reports, the file files to
transfer.
This is a one-node cluster with a tiny file. Here is the output from the
NameNode:
{code:none}
2019-07-09 20:45:37,198 INFO namenode.FSNamesystem: BLOCK* blk_1073741829_1005
is COMMITTED but not COMPLETE(numNodes= 0 < minimum = 1) in file
/tmp/sent.12.data._COPYING_
2019-07-09 20:45:37,606 INFO namenode.FSNamesystem: BLOCK* blk_1073741829_1005
is COMMITTED but not COMPLETE(numNodes= 0 < minimum = 1) in file
/tmp/sent.12.data._COPYING_
2019-07-09 20:45:38,413 INFO namenode.FSNamesystem: BLOCK* blk_1073741829_1005
is COMMITTED but not COMPLETE(numNodes= 1 >= minimum = 1) in file
/tmp/sent.12.data._COPYING_
2019-07-09 20:45:40,023 INFO namenode.FSNamesystem: BLOCK* blk_1073741829_1005
is COMMITTED but not COMPLETE(numNodes= 1 >= minimum = 1) in file
/tmp/sent.12.data._COPYING_
{code}
If I setup a thread that watches for IBRs and send them to the NN immediately,
the file transfers successfully. There may be a race condition here.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-dev-h...@hadoop.apache.org
