[
https://issues.apache.org/jira/browse/HDFS-15390?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17126638#comment-17126638
]
Sean Chow edited comment on HDFS-15390 at 6/5/20, 2:57 PM:
-----------------------------------------------------------
It's easy to reproduce. You have setup HA namenodes, and a new machine with the
same hostname with nn2(standby), and copied name-data directory.
# Use {{hdfs dfs put}} to write a bigfile (to make it a long running client)
# Stop old nn2, and start new nn2
# Update the nn2 hostname to resolve as the new ipaddr on all hosts
# Failover from nn1 to nn2
# Now you found the client occurs error continuously. (In yarn nodemanager
scenario, this nodemanager is totally sick until restarted)
There 's two way to fix this:
# When updateAddress is true, do not handle ConnectionFailure this round
# When address change detected, update namenode proxies (only with
{{ConfiguredFailoverProxyProvider}})
Method one is easy, and in this connection lifecycle the client will use the
right {{server}} to connect. But when the client connection closed and create a
new one. It will always try to getConnection from the retired ipaddr, because
the namenode proxies is still the old one.
Method two solve the root cause. Everytime the client failover namenodes, check
ipaddr changed or not. If changed, re-initialize the namenode failover proxies.
was (Author: seanlook):
There 's two way to fix this:
# When updateAddress is true, do not handle ConnectionFailure this round
# When address change detected, update namenode proxies (only with
{{ConfiguredFailoverProxyProvider}})
Method one is easy, and in this connection lifecycle the client will use the
right {{server}} to connect. But when the client connection closed and create a
new one. It will always try to getConnection from the retired ipaddr, because
the namenode proxies is still the old one.
Method two solve the root cause. Everytime the client failover namenodes, check
ipaddr changed or not. If changed, re-initialize the namenode failover proxies.
> client fails forever when namenode ipaddr changed
> -------------------------------------------------
>
> Key: HDFS-15390
> URL: https://issues.apache.org/jira/browse/HDFS-15390
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: dfsclient
> Affects Versions: 2.10.0, 2.9.2, 3.2.1
> Reporter: Sean Chow
> Priority: Major
> Attachments: HDFS-15390.01.patch
>
>
> For machine replacement, I replace my standby namenode with a new ipaddr and
> keep the same hostname. Also update the client's hosts to make it resolve
> correctly
> When I try to run failover to transite the new namenode(let's say nn2), the
> client will fail to read or write forever until it's restarted.
> That make yarn nodemanager in sick state. Even the new tasks will encounter
> this exception too. Until all nodemanager restart.
>
> {code:java}
> 20/06/02 15:12:25 WARN ipc.Client: Address change detected. Old:
> nn2-192-168-1-100/192.168.1.100:9000 New: nn2-192-168-1-100/192.168.1.200:9000
> 20/06/02 15:12:25 DEBUG ipc.Client: closing ipc connection to
> nn2-192-168-1-100/192.168.1.200:9000: Connection refused
> java.net.ConnectException: Connection refused
> at sun.nio.ch.SocketChannelImpl.checkConnect(Native Method)
> at
> sun.nio.ch.SocketChannelImpl.finishConnect(SocketChannelImpl.java:717)
> at
> org.apache.hadoop.net.SocketIOWithTimeout.connect(SocketIOWithTimeout.java:206)
> at org.apache.hadoop.net.NetUtils.connect(NetUtils.java:530)
> at org.apache.hadoop.net.NetUtils.connect(NetUtils.java:494)
> at
> org.apache.hadoop.ipc.Client$Connection.setupConnection(Client.java:608)
> at
> org.apache.hadoop.ipc.Client$Connection.setupIOstreams(Client.java:707)
> at
> org.apache.hadoop.ipc.Client$Connection.access$2800(Client.java:368)
> at org.apache.hadoop.ipc.Client.getConnection(Client.java:1517)
> at org.apache.hadoop.ipc.Client.call(Client.java:1440)
> at org.apache.hadoop.ipc.Client.call(Client.java:1401)
> at
> org.apache.hadoop.ipc.ProtobufRpcEngine$Invoker.invoke(ProtobufRpcEngine.java:232)
> at com.sun.proxy.$Proxy9.addBlock(Unknown Source)
> at
> org.apache.hadoop.hdfs.protocolPB.ClientNamenodeProtocolTranslatorPB.addBlock(ClientNamenodeProtocolTranslatorPB.java:399)
> at sun.reflect.GeneratedMethodAccessor8.invoke(Unknown Source)
> at
> sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
> at java.lang.reflect.Method.invoke(Method.java:498)
> at
> org.apache.hadoop.io.retry.RetryInvocationHandler.invokeMethod(RetryInvocationHandler.java:193)
> at
> org.apache.hadoop.io.retry.RetryInvocationHandler.invoke(RetryInvocationHandler.java:102)
> {code}
>
> We can see the client has {{Address change detected}}, but it still fails. I
> find out that's because when method {{updateAddress()}} return true, the
> {{handleConnectionFailure()}} thow an exception that break the next retry
> with the right ipaddr.
> Client.java: setupConnection()
> {code:java}
> } catch (ConnectTimeoutException toe) {
> /* Check for an address change and update the local reference.
> * Reset the failure counter if the address was changed
> */
> if (updateAddress()) {
> timeoutFailures = ioFailures = 0;
> }
> handleConnectionTimeout(timeoutFailures++,
> maxRetriesOnSocketTimeouts, toe);
> } catch (IOException ie) {
> if (updateAddress()) {
> timeoutFailures = ioFailures = 0;
> }
> // because the namenode ip changed in updateAddress(), the old namenode
> ipaddress cannot be accessed now
> // handleConnectionFailure will thow an exception, the next retry never have
> a chance to use the right server updated in updateAddress()
> handleConnectionFailure(ioFailures++, ie);
> }
> {code}
>
--
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
