[
https://issues.apache.org/jira/browse/CASSANDRA-17116?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17454294#comment-17454294
]
Francisco Guerrero commented on CASSANDRA-17116:
------------------------------------------------
I took a look at this as well, and it seems that the follower keeps track of
the number of {{RECEIVED}} messages it expects and the number of received
messages it has seen. Only when the follower has seen all the {{RECEIVED}}
messages it expects to see, will the follower initiate the {{COMPLETE}}
message, and close the connections.
It looks like the initiator is sending back the received message before
actually receiving the stream. The code sits in
{{org.apache.cassandra.streaming.StreamSesssion.receive(IncomingStreamMessage)}}.
We can see that the control message is being sent and then the receiver
receives the stream.
A potential solution to the race is delaying the sending of the {{RECEIVED}}
message right after the stream has been consumed (maybe in the finally block of
the {{receive}} method).
Unfortunately, I have not found a good way to reproduce the issue, so I don't
know if that will be sufficient to resolve the race issue. Maybe if [~dcapwell]
can try it, or give me some guidance on how to repro the issue, I can take a
look into it further.
> When zero-copy-streaming sees a channel close this triggers the disk failure
> policy
> -----------------------------------------------------------------------------------
>
> Key: CASSANDRA-17116
> URL: https://issues.apache.org/jira/browse/CASSANDRA-17116
> Project: Cassandra
> Issue Type: Bug
> Components: Consistency/Streaming
> Reporter: David Capwell
> Assignee: David Capwell
> Priority: Normal
> Fix For: 4.x
>
>
> Found in CASSANDRA-17085.
> https://app.circleci.com/pipelines/github/dcapwell/cassandra/1069/workflows/26b7b83a-686f-4516-a56a-0709d428d4f2/jobs/7264
> https://app.circleci.com/pipelines/github/dcapwell/cassandra/1069/workflows/26b7b83a-686f-4516-a56a-0709d428d4f2/jobs/7256
> {code}
> ERROR [Stream-Deserializer-/127.0.0.1:7000-f2eb1a15] 2021-11-02 21:35:40,983
> DefaultFSErrorHandler.java:104 - Exiting forcefully due to file system
> exception on startup, disk failure policy "stop"
> org.apache.cassandra.io.FSWriteError: java.nio.channels.ClosedChannelException
> at
> org.apache.cassandra.io.sstable.format.big.BigTableZeroCopyWriter.write(BigTableZeroCopyWriter.java:227)
> at
> org.apache.cassandra.io.sstable.format.big.BigTableZeroCopyWriter.writeComponent(BigTableZeroCopyWriter.java:206)
> at
> org.apache.cassandra.db.streaming.CassandraEntireSSTableStreamReader.read(CassandraEntireSSTableStreamReader.java:125)
> at
> org.apache.cassandra.db.streaming.CassandraIncomingFile.read(CassandraIncomingFile.java:84)
> at
> org.apache.cassandra.streaming.messages.IncomingStreamMessage$1.deserialize(IncomingStreamMessage.java:51)
> at
> org.apache.cassandra.streaming.messages.IncomingStreamMessage$1.deserialize(IncomingStreamMessage.java:37)
> at
> org.apache.cassandra.streaming.messages.StreamMessage.deserialize(StreamMessage.java:50)
> at
> org.apache.cassandra.streaming.StreamDeserializingTask.run(StreamDeserializingTask.java:62)
> at
> io.netty.util.concurrent.FastThreadLocalRunnable.run(FastThreadLocalRunnable.java:30)
> at java.lang.Thread.run(Thread.java:748)
> Caused by: java.nio.channels.ClosedChannelException: null
> at
> org.apache.cassandra.net.AsyncStreamingInputPlus.reBuffer(AsyncStreamingInputPlus.java:136)
> at
> org.apache.cassandra.net.AsyncStreamingInputPlus.consume(AsyncStreamingInputPlus.java:155)
> at
> org.apache.cassandra.io.sstable.format.big.BigTableZeroCopyWriter.write(BigTableZeroCopyWriter.java:217)
> ... 9 common frames omitted
> {code}
> When bootstrap fails and streaming is closed, this triggers the disk failure
> policy which causes the JVM to halt by default (if this happens outside of
> bootstrap, then we stop transports and keep the JVM up).
> org.apache.cassandra.streaming.StreamDeserializingTask attempts to handle
> this by ignoring this exception, but the call to
> org.apache.cassandra.streaming.messages.IncomingStreamMessage$1.deserialize
> Does try/catch and inspects exception; triggering this condition.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
