You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@lucene.apache.org by "Cao Manh Dat (Jira)" <ji...@apache.org> on 2019/11/05 11:34:00 UTC
[jira] [Comment Edited] (SOLR-13896) Paused a non-leader node can
cause recovery on other nodes
[ https://issues.apache.org/jira/browse/SOLR-13896?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16967444#comment-16967444 ]
Cao Manh Dat edited comment on SOLR-13896 at 11/5/19 11:33 AM:
---------------------------------------------------------------
There are two ways to fix this problem
* change pollQueueTimeout to resonable amount, It just need to be larger than the time between parsing 2 consecutive documents in a batch of a request. Let's say 10 seconds.
* interrupt ConcurrentUpdateSolrClients belong to different replicas in parallel.
* interrupt ConcurrentUpdateSolrClient first then wait for it to complete later.
Attached a patch corresponds the first solution since it easy.
was (Author: caomanhdat):
There are two ways to fix this problem
* change pollQueueTimeout to resonable amount, It just need to be larger than the time between parsing 2 consecutive documents in a batch of a request. Let's say 10 seconds.
* interrupt ConcurrentUpdateSolrClients belong to different replicas in parallel.
* interrupt ConcurrentUpdateSolrClient first then wait for it to complete later.
Attached a patch for the first solution since it easy.
> Paused a non-leader node can cause recovery on other nodes
> ----------------------------------------------------------
>
> Key: SOLR-13896
> URL: https://issues.apache.org/jira/browse/SOLR-13896
> Project: Solr
> Issue Type: Bug
> Security Level: Public(Default Security Level. Issues are Public)
> Reporter: Cao Manh Dat
> Assignee: Cao Manh Dat
> Priority: Major
> Attachments: SOLR-13896.patch
>
>
> All stacktraces below based on 7.5 branch. This problem still exist at 8.x branches. Here is the scenario, we have 3 replicas
> * L: the leader replica
> * R: the normal replica
> * P: the poor one which was paused then resumed
> L is trying to send data to R, P during that P get paused, here is what happen at L's threads.
> * Thread 1 is stucking at this line of StreamingSolrClients
> {code:java}
> public synchronized void blockUntilFinished() {
> for (ConcurrentUpdateSolrClient client : solrClients.values()) {
> client.blockUntilFinished();
> }
> } {code}
> basically this thread is trying to wait for other sender threads to finish. Let's assume that this is the content of *solrClients.values : [clientToP, clientToR]*
> * Thread 2 coressponds to *clientToP* since P is paused, it doesn't close the connection. it just keep the connection and never return any data backs to L. So this thread stuck with this stack trace, waiting for response data from *P* (with timeout=600000ms)*.* Therefore it cause the thread1 stuck at *clientToP.blockUntilFinished()*
> {code:java}
> java.lang.Thread.State: RUNNABLE java.lang.Thread.State: RUNNABLE at java.net.SocketInputStream.socketRead0(Native Method) at java.net.SocketInputStream.socketRead(SocketInputStream.java:116) at java.net.SocketInputStream.read(SocketInputStream.java:171) at java.net.SocketInputStream.read(SocketInputStream.java:141) at org.apache.http.impl.io.SessionInputBufferImpl.streamRead(SessionInputBufferImpl.java:137) at org.apache.http.impl.io.SessionInputBufferImpl.fillBuffer(SessionInputBufferImpl.java:153) at org.apache.http.impl.io.SessionInputBufferImpl.readLine(SessionInputBufferImpl.java:282) at org.apache.http.impl.conn.DefaultHttpResponseParser.parseHead(DefaultHttpResponseParser.java:138) at org.apache.http.impl.conn.DefaultHttpResponseParser.parseHead(DefaultHttpResponseParser.java:56) at org.apache.http.impl.io.AbstractMessageParser.parse(AbstractMessageParser.java:259) at org.apache.http.impl.DefaultBHttpClientConnection.receiveResponseHeader(DefaultBHttpClientConnection.java:163) at org.apache.http.impl.conn.CPoolProxy.receiveResponseHeader(CPoolProxy.java:165) at org.apache.http.protocol.HttpRequestExecutor.doReceiveResponse(HttpRequestExecutor.java:273) at org.apache.http.protocol.HttpRequestExecutor.execute(HttpRequestExecutor.java:125) at org.apache.solr.util.stats.InstrumentedHttpRequestExecutor.execute(InstrumentedHttpRequestExecutor.java:120) at org.apache.http.impl.execchain.MainClientExec.execute(MainClientExec.java:272) at org.apache.http.impl.execchain.ProtocolExec.execute(ProtocolExec.java:185) at org.apache.http.impl.execchain.RetryExec.execute(RetryExec.java:89) at org.apache.http.impl.execchain.RedirectExec.execute(RedirectExec.java:111) at org.apache.http.impl.client.InternalHttpClient.doExecute(InternalHttpClient.java:185) at org.apache.http.impl.client.CloseableHttpClient.execute(CloseableHttpClient.java:83) at org.apache.http.impl.client.CloseableHttpClient.execute(CloseableHttpClient.java:56) at org.apache.solr.client.solrj.impl.ConcurrentUpdateSolrClient$Runner.sendUpdateStream(ConcurrentUpdateSolrClient.java:347){code}
> * Since *clientToR* is the second element of the array, is never get called (or at least after the timeout). This problem cause Thread 3, to stuck at this line
> {code:java}
> upd = queue.poll(pollQueueTime, TimeUnit.MILLISECONDS); {code}
> note that pollQueueTime == Integer.MAX_VALUE (this set by StreamingSolrClients). Therefore unless clientToR.blockUntilFinished() is called (which interrupt Thread 3) this Thread 3 will stuck at above line forever
> * because *clientToR* is sending data to R but never close the outputstream, so basically R just waiting forever (until timeout at 120000ms later). Which then lead to this exception
> {code:java}
> o.a.s.h.RequestHandlerBase java.io.IOException: java.util.concurrent.TimeoutException: Idle timeout expired: 120003/120000 mso.a.s.h.RequestHandlerBase java.io.IOException: java.util.concurrent.TimeoutException: Idle timeout expired: 120003/120000 ms at org.eclipse.jetty.server.HttpInput$ErrorState.noContent(HttpInput.java:1080) at org.eclipse.jetty.server.HttpInput.read(HttpInput.java:313) at org.apache.solr.servlet.ServletInputStreamWrapper.read(ServletInputStreamWrapper.java:74) at org.apache.commons.io.input.ProxyInputStream.read(ProxyInputStream.java:100) at org.apache.solr.common.util.FastInputStream.readWrappedStream(FastInputStream.java:79) at org.apache.solr.common.util.FastInputStream.refill(FastInputStream.java:88) at org.apache.solr.common.util.FastInputStream.peek(FastInputStream.java:60) at org.apache.solr.handler.loader.JavabinLoader.parseAndLoadDocs(JavabinLoader.java:107) at org.apache.solr.handler.loader.JavabinLoader.load(JavabinLoader.java:55) {code}
> * After that the leader put all replicas including none-paused one to recovery
>
> It is a very bad outcome and, this is not just theoretical problem since some cloud platforms can freeze a node when doing maintenance.
> Thanks [~ab] and [~shalin] on helping me debugging this problem.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org