[
https://issues.apache.org/jira/browse/THRIFT-5327?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17266155#comment-17266155
]
Mario Emmenlauer commented on THRIFT-5327:
------------------------------------------
I'm curious to look into this, but I'm currently involved in way too much other
things, so it may take some time.
I have not read through the details of the issue, so just to be sure: Are there
clear instructions how to reproduce the issue? The best thing would be
something like a pull request with a test that fails. If this is not possible
then at least as clear instructions what to look for as possible.
The reason I'm asking for this is that we run regular heavy stress tests on
thrift. I can sometimes observe random crashes, but with low probability. And I
have not seen slow response time. So it may be hard for me to reproduce, unless
we can write a test for the specific problem.
> ServerHello is sent with a delay from thrift
> --------------------------------------------
>
> Key: THRIFT-5327
> URL: https://issues.apache.org/jira/browse/THRIFT-5327
> Project: Thrift
> Issue Type: Bug
> Components: C++ - Library
> Affects Versions: 0.13.0
> Environment: C++ Application running on linux x86_64
> Reporter: A.Hari Venkata Ramana
> Priority: Major
>
> I have a C++ environment which uses thrift for server. Server is of type
> TThreadPoolServer with TBufferedTransportFactory as transport and
> TBinaryProtocolFactory as protocol. Clients get connected to this server and
> this happens via TLS. We are using thrift server with version 0.13.0 on the
> server end.
>
> In one of the environments where a client using thrift (version 0.10.0), the
> client Hello is sent as part of initial TLS negotiation. We have observed
> that the server Hello is sent after around 25-30 seconds after the client
> Hello is received by the server.
>
> In these cases where delay is observed on server end we are printing the
> client origin details, ie., the server ip of client and the port from which
> it is connecting to the server.
>
> functionXXX(const ::apache::thrift::TConnectionInfo& connInfo){
> std::string origin = connInfo.input->getInputTransport()->*getOrigin*();
> printf("CONNECTED: %s\n", origin.c_str() );
>
> .......}
> In cases where the ServerHello is delayed from thrift, we see the junk
> characters in prints from the above code as shown below:
>
> { Info = "CONNECTED: ����:52089"}
>
> { Info = "CONNECTED: ����:52465"}
> { Info = "CONNECTED: ����:55027"}
> { Info = "CONNECTED: ����:59066"}
> In cases where the IP address of the client is shown as normal characters (as
> shown below) we dont see any delay in the serverHello sent from the thrift
> server and it just takes few milli seconds to send the serverHello back to
> client.
>
> \{ Info = "CONNECTED: 172.30.1.41:46573"}
>
> Looking at the change history we see the following change done in thrift
> 0.13.0:
>
> - [THRIFT-4762](https://issues.apache.org/jira/browse/THRIFT-4762) - cpp:
> TTransport::getOrigin() is now const
>
> Could this be the cause of the delay from the thrift server in sending the
> Server Hello back to clients as part of TLS negotiation?
>
> Any pointers here would be very useful.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
