[jira] [Commented] (THRIFT-2682) TThreadedServer leaks per-thread memory

["James E. King, III (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/THRIFT-2682?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14486441#comment-14486441 ] 

James E. King, III commented on THRIFT-2682:
--------------------------------------------

I recently ran valgrind on the C++ unit test suite and found no leaks in this area.  Is this still an issue?  If so, adding a unit test that exposes it would be a great way to get started on the road to a fix.

> TThreadedServer leaks per-thread memory
> ---------------------------------------
>
>                 Key: THRIFT-2682
>                 URL: https://issues.apache.org/jira/browse/THRIFT-2682
>             Project: Thrift
>          Issue Type: Bug
>          Components: C++ - Library
>    Affects Versions: 0.9.1
>         Environment: RHEL 6.5
>            Reporter: Artem Koval
>            Priority: Critical
>         Attachments: thrift_vg.txt
>
>
> We are using Thrift 0.9.1 in Linux, both client and server are written in C++. TThreadedServer is used to process requests involving large (up to 1MB) binary parameters.
> Usage pattern is like this: the client establishes a UNIX socket connection, makes some interface calls and then disconnects.
> We noticed that the memory footprint of the server process increases over time. Valgrind inspection showed that despite the server threads have long been shutdown, their data allocated in TThreadedServer::serve() hasn't been freed and shows up as "definitely lost".
> This leak, although significant, is not critical by itself. But per-thread data includes a TBinaryProtocol object, which in turn holds a "string buffer" used to serialize binary parameters. These string buffers are quite large in our case, leading to overall memory leakage being unacceptably high.
> As a workaround, we will for now turn to TNonblockingServer, which uses persistent threads. However, a memory leak is still a very serious problem, and it should be fixed.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)