[
https://issues.apache.org/jira/browse/HDFS-14646?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17222993#comment-17222993
]
Adam Binford commented on HDFS-14646:
-------------------------------------
We just encountered this on our 3.1.4 cluster, and it drove me crazy trying to
figure out what the heck was going on. We currently have 4 namenodes (we're
trying to transition from our original namenode to 3 new namenodes, hence 4 in
the mean time). The checkpointing to the ANN seems to work once per hour (the
default checkpoint interval), but the SNN also tries to upload to all the other
SNN which fail with this exception. This seems to make even the successful
upload to the ANN "appear" to fail (since uploads to the other SSNs failed), so
all the SSNs are just constantly trying to checkpoint and fail over and over.
Our current image is ~3GB compressed.
Any plans to try to get this in soon? would be great if this could get fixed in
the next release. It seems like the only option is to not use more than one SNN
until then.
> Standby NameNode should not upload fsimage to an inappropriate NameNode.
> ------------------------------------------------------------------------
>
> Key: HDFS-14646
> URL: https://issues.apache.org/jira/browse/HDFS-14646
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: hdfs
> Affects Versions: 3.1.2
> Reporter: Xudong Cao
> Assignee: Xudong Cao
> Priority: Major
> Labels: multi-sbnn
> Attachments: HDFS-14646.000.patch, HDFS-14646.001.patch,
> HDFS-14646.002.patch
>
>
> *Problem Description:*
> In the multi-NameNode scenario, when a SNN uploads a FsImage, it will put
> the image to all other NNs (whether the peer NN is an ANN or not), and even
> if the peer NN immediately replies an error (such as
> TransferResult.NOT_ACTIVE_NAMENODE_FAILURE, TransferResult
> .OLD_TRANSACTION_ID_FAILURE, etc.), the local SNN will not terminate the put
> process immediately, but will put the FsImage completely to the peer NN, and
> will not read the peer NN's reply until the put is completed.
> Depending on the version of Jetty, this behavior can lead to different
> consequences :
> *1.Under Hadoop 2.7.2 (with Jetty 6.1.26)*
> After peer NN called HttpServletResponse.sendError(), the underlying TCP
> connection will still be established, and the data SNN sent will be read by
> Jetty framework itself in the peer NN side, so the SNN will insignificantly
> send the FsImage to the peer NN continuously, causing a waste of time and
> bandwidth. In a relatively large HDFS cluster, the size of FsImage can often
> reach about 30GB, This is indeed a big waste.
> *2.Under newest release-3.2.0-RC1 (with Jetty 9.3.24) and trunk (with Jetty
> 9.3.27)*
> After peer NN called HttpServletResponse.sendError(), the underlying TCP
> connection will be auto closed, and then SNN will directly get an "Error
> writing request body to server" exception, as below, note this test needs a
> relatively big FSImage (e.g. 10MB level):
> {code:java}
> 2019-08-17 03:59:25,413 INFO namenode.TransferFsImage: Sending fileName:
> /tmp/hadoop-root/dfs/name/current/fsimage_0000000000003364240, fileSize:
> 9864721. Sent total: 524288 bytes. Size of last segment intended to send:
> 4096 bytes.
> java.io.IOException: Error writing request body to server
> at
> sun.net.www.protocol.http.HttpURLConnection$StreamingOutputStream.checkError(HttpURLConnection.java:3587)
> at
> sun.net.www.protocol.http.HttpURLConnection$StreamingOutputStream.write(HttpURLConnection.java:3570)
> at
> org.apache.hadoop.hdfs.server.namenode.TransferFsImage.copyFileToStream(TransferFsImage.java:396)
> at
> org.apache.hadoop.hdfs.server.namenode.TransferFsImage.writeFileToPutRequest(TransferFsImage.java:340)
> at
> org.apache.hadoop.hdfs.server.namenode.TransferFsImage.uploadImage(TransferFsImage.java:314)
> at
> org.apache.hadoop.hdfs.server.namenode.TransferFsImage.uploadImageFromStorage(TransferFsImage.java:249)
> at
> org.apache.hadoop.hdfs.server.namenode.ha.StandbyCheckpointer$1.call(StandbyCheckpointer.java:277)
> at
> org.apache.hadoop.hdfs.server.namenode.ha.StandbyCheckpointer$1.call(StandbyCheckpointer.java:272)
> at java.util.concurrent.FutureTask.run(FutureTask.java:266)
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
> at java.lang.Thread.run(Thread.java:748)
> 2019-08-17 03:59:25,422 INFO namenode.TransferFsImage: Sending fileName:
> /tmp/hadoop-root/dfs/name/current/fsimage_0000000000003364240, fileSize:
> 9864721. Sent total: 851968 bytes. Size of last segment intended to send:
> 4096 bytes.
> java.io.IOException: Error writing request body to server
> at
> sun.net.www.protocol.http.HttpURLConnection$StreamingOutputStream.checkError(HttpURLConnection.java:3587)
> at
> sun.net.www.protocol.http.HttpURLConnection$StreamingOutputStream.write(HttpURLConnection.java:3570)
> at
> org.apache.hadoop.hdfs.server.namenode.TransferFsImage.copyFileToStream(TransferFsImage.java:396)
> at
> org.apache.hadoop.hdfs.server.namenode.TransferFsImage.writeFileToPutRequest(TransferFsImage.java:340)
> {code}
>
> *Solution:*
> A standby NameNode should not upload fsimage to an inappropriate NameNode,
> when he plans to put a FsImage to the peer NN, he need to check whether he
> really need to put it at this time.
> In detail, local SNN should establish an HTTP connection with the peer NN,
> send the put request, and then immediately read the response (this is the key
> point). If the peer NN does not reply an HTTP_OK, it means the local SNN
> should not put image at this time.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
