Staffan Friberg created HDFS-10563:
--------------------------------------
Summary: Block reports could be silently dropped by NN
Key: HDFS-10563
URL: https://issues.apache.org/jira/browse/HDFS-10563
Project: Hadoop HDFS
Issue Type: Bug
Components: namenode
Affects Versions: 3.0.0-beta1
Reporter: Staffan Friberg
Reading through the block reporting code I think I've spotted a case when block
reports can silently be dropped and leave thread waiting indefinitely on a
FutureTask that never will be executed.
The BlockReportProcessingThread.enqueue method doesn't return any status on if
the enqueuing of the task was successful and does not handle the case when the
queue is full and offer return false.
Going back through the call stack to BlockManager.runBlockOp, which indirectly
calls enqueue with a FutureTask and then proceeds to do get() om the task.
So if the internal queue in the BlockReportingProcessingThread is full, the BR
would never be handled and the thread queuing the task would wait indefinitely
on the FutureTask that will never be executed.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-dev-h...@hadoop.apache.org
