[
https://issues.apache.org/jira/browse/DRILL-5659?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16084984#comment-16084984
]
Parth Chandra commented on DRILL-5659:
--------------------------------------
So, it looks like this:
The original code for reading (synchronously) from a socket looked like this:
{code}
while(1){
size_t dataBytesRead = m_socket.read_some(boost::asio::buffer(inBuf,
bytesToRead),
errorCode);
if(errorCode){
break;
}
...
}
{code}
and now it looks like
{code}
while(1){
size_t dataBytesRead = m_socket.read_some(boost::asio::buffer(inBuf,
bytesToRead),
errorCode);
if(EINTR != errorCode.value()){
break;
}
...
}
{code}
The original works correctly and the new code doesn't.
If {code}errorCode.value() == 0{code} we break out of the while loop even if we
have not managed to read all the data off the socket. This typically will
happen when there is a large data set and the server is returning data really,
really fast so there is a lot of data in the socket stream.
Modifying the code to check for errorCode (which is equivalent to checking
errorCode.value() != 0) first, makes everyone happy again.
Will submit a patch after a little more testing.
> C++ Client (master) behavior is unstable resulting incorrect result or
> exception in API calls
> ---------------------------------------------------------------------------------------------
>
> Key: DRILL-5659
> URL: https://issues.apache.org/jira/browse/DRILL-5659
> Project: Apache Drill
> Issue Type: Bug
> Reporter: Rob Wu
> Fix For: 1.11.0
>
> Attachments: 1-10cppClient-1.10.0Drillbit-hive.log,
> 1-10cppClient-1.10.0Drillbit-metadata and catalog test.log,
> 1-10cppClient-1.9.0Drillbit-dfs.log, 1-10cppClient-1.9.0Drillbit-hive.log,
> 1-10cppClient-1.9.0Drillbit-metadata and catalog test.log,
> 1-11cppClient-1.10.0Drillbit-hive.log, 1-11cppClient-1.10.0Drillbit-metadata
> and catalog test.log, 1-11cppClient-1.9.0Drillbit-dfs.log,
> 1-11cppClient-1.9.0Drillbit-hive.log, 1-11cppClient-1.9.0Drillbit-metadata
> and catalog test.log
>
>
> I recently compiled the C++ client (on windows) from master and tested
> against a 1.9.0 drillbit. The client's behavior does not meet the stable
> release requirement.
> Some API functionalities are broken and should be investigated.
> Most noticeable is the getColumns(...) call. It will throw an exception with
> "Cannot decode getcolumns results" when the number of rows (column records)
> exceeds a certain number.
> I also noticed that: during query execution + data retrieval, if the table is
> large enough, the result coming back will start to become NULL or empty.
> I will see if I can generate some drillclient logs to put in the attachment.
> I will also compile and test on other platforms.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
