You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@thrift.apache.org by "Ioannis Efthymiou (Jira)" <ji...@apache.org> on 2021/11/13 15:10:00 UTC

[jira] [Commented] (THRIFT-5480) TThreadPoolAsyncServer using TFramedTransport mistakenly drops client

    [ https://issues.apache.org/jira/browse/THRIFT-5480?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17443120#comment-17443120 ] 

Ioannis Efthymiou commented on THRIFT-5480:
-------------------------------------------

I added a minimal example server, and a python client. You could try different versions of clients as well. You can see that you can't call the first method after you've called the second method. As a matter of fact, the transport will be closed on the server side, from TEndpointTransport.ResetConsumedMessageSize throwing a TTransportException.

I was under the impression that the check disabled, it's just reset for the next time it's needed. I suppose I missed something

> TThreadPoolAsyncServer using TFramedTransport mistakenly drops client
> ---------------------------------------------------------------------
>
>                 Key: THRIFT-5480
>                 URL: https://issues.apache.org/jira/browse/THRIFT-5480
>             Project: Thrift
>          Issue Type: Bug
>          Components: netstd - Library
>    Affects Versions: 0.15.0
>            Reporter: Ioannis Efthymiou
>            Assignee: Jens Geyer
>            Priority: Major
>              Labels: Oneway, TFramedTransport, TThreadPoolAsyncServer
>         Attachments: 0001-Reset-message-size-in-TFramedTransport.ReadFrameAsyn-1.patch, ThriftBug.zip
>
>          Time Spent: 0.5h
>  Remaining Estimate: 0h
>
> TThreadPoolAsyncServer using TFramedTransport silently drops a client service using oneway methods when the second message is bigger than the first one.
> I have an application implementing a one way service, depending on it's state. I created a TThreadpoolAsync server using the netstd 0.15 version, and I noticed that after a point, the notifications stopped coming, but there was no error on either side. The methods all worked individually, but some did not work after others.
> Eliminating all possible causes on my side, I debugged the source code, and discovered that a transport exception was thrown from TEndpointTransport when a message size was bigger than the proceeding one, because TFramedTransport.ReadFrameAsync was not resetting the message size when it was done reading it, thus causing TEndpointTransport.UpdateKnownMessageSize to throw the exception. The server interpreted the exception as if the client had disconnected, so it silently ignored it.
>  
> I've attached a patch that corrects this behaviour



--
This message was sent by Atlassian Jira
(v8.20.1#820001)