[
https://issues.apache.org/jira/browse/HDFS-14646?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17766559#comment-17766559
]
Matthew Sharp commented on HDFS-14646:
--------------------------------------
We hit this running Hadoop 3.3.0 as well with 3 NNs. It stayed quiet for years
until we saw enough write activity. The txns since last checkpoint metrics
would show a really bad lag and eventually got far enough out that one of the
SNNs would crash since it couldn't find the older txn id. Shutting down one SNN
over the last few weeks seems to have resolved this so it does appear to be an
active issue running 3 NNs. It may be worth noting that the fsimage size did
not seem to play too much of a factor. We see this logging exceptions in a dev
cluster which is only hundreds of MBs.
> 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
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
