[
https://issues.apache.org/jira/browse/HDFS-101?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Todd Lipcon updated HDFS-101:
-----------------------------
Attachment: hdfs-101.tar.gz
Hi Hairong,
This doesn't seem to fix the issue for me. Here's the test setup:
- Cluster is with 3 DNs, running on EC2.
- On DN at 10.251.43.82, I have dfs.data.dir pointing at two volumes. One has
150M remaining. On the other nodes, there is plenty of disk space. I've set
dfs.du.reserved to 0 on the small-disk volume.
- From a separate node, I try to upload a 100M file. As soon as it starts
uploading, I issue a "dd" command on 10.251.43.82 to fill up the small volume
(causing the disk to fill up while the local DN is writing to that volume)
- If this patch works correctly, it should figure out that .82 is the problem
node, and recover the pipeline to the other two nodes in the cluster. It should
never eject a different node from the pipeline.
Although the write did succeed, it incorrectly decided .148 was the bad
datanode even though .82 experienced the failure. I've attached the logs from
the perspective of the writer as well as the 3 DNs in the pipeline.
Although it didn't show up in this particular capture, I also saw the following
exception in a few tests:
2009-12-17 18:36:31,622 WARN org.apache.hadoop.hdfs.DFSClient: DFSOutputStream
ResponseProcessor exception for block
blk_6564765548548790058_1354java.util.NoSuchElementException
at java.util.LinkedList.remove(LinkedList.java:788)
at java.util.LinkedList.removeFirst(LinkedList.java:134)
at
org.apache.hadoop.hdfs.DFSClient$DFSOutputStream$ResponseProcessor.run(DFSClient.java:2435)
I'll make a local edit to fix the log message for "Expecting seqno" to actually
print the expected seqno, and try to rerun the tests, as that might help figure
out the issue.
> DFS write pipeline : DFSClient sometimes does not detect second datanode
> failure
> ---------------------------------------------------------------------------------
>
> Key: HDFS-101
> URL: https://issues.apache.org/jira/browse/HDFS-101
> Project: Hadoop HDFS
> Issue Type: Bug
> Affects Versions: 0.20.1
> Reporter: Raghu Angadi
> Assignee: Hairong Kuang
> Priority: Blocker
> Fix For: 0.21.0
>
> Attachments: detectDownDN-0.20.patch, detectDownDN.patch,
> detectDownDN1.patch, hdfs-101.tar.gz
>
>
> When the first datanode's write to second datanode fails or times out
> DFSClient ends up marking first datanode as the bad one and removes it from
> the pipeline. Similar problem exists on DataNode as well and it is fixed in
> HADOOP-3339. From HADOOP-3339 :
> "The main issue is that BlockReceiver thread (and DataStreamer in the case of
> DFSClient) interrupt() the 'responder' thread. But interrupting is a pretty
> coarse control. We don't know what state the responder is in and interrupting
> has different effects depending on responder state. To fix this properly we
> need to redesign how we handle these interactions."
> When the first datanode closes its socket from DFSClient, DFSClient should
> properly read all the data left in the socket.. Also, DataNode's closing of
> the socket should not result in a TCP reset, otherwise I think DFSClient will
> not be able to read from the socket.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
