thrift: TNonblockingServer: clean up state in the
---------------------------------------------------
Key: THRIFT-1290
URL: https://issues.apache.org/jira/browse/THRIFT-1290
Project: Thrift
Issue Type: Improvement
Components: C++ - Library
Reporter: Dave Watson
Priority: Minor
Attachments:
0017-thrift-TNonblockingServer-clean-up-state-in-the-dest.patch
>From cd9c1a10cb4df058fbdbed1b98a21a7a7470a28c Mon Sep 17 00:00:00 2001
From: Adam Simpkins <[email protected]>
Date: Tue, 6 Apr 2010 20:43:23 +0000
Subject: [PATCH 17/33] thrift: TNonblockingServer: clean up state in the
destructor
Summary:
Implement the TNonblockingServer destructor, so that it closes the
listen socket, destroys the event_base, and deletes any TConnections
left in the connectionStack_. However, TNonblockingServer doesn't keep
track of active TConnection objects, so those objects are still leaked.
As part of this, I also changed the code to use event_init() rather than
event_base_new(). This way we won't set the global event_base inside
libevent, and we can be sure that no one else will be using it after the
TNonblockingServer is destroyed.
I grepped through all of [fb code base] to check for any other direct uses of
event_set(), and didn't see any places that weren't also using
event_base_set(). Therefore it seems like it should be safe to stop
initializing the global event_base pointer.
Test Plan:
Tested with the test code in [a fb unittest], which creates, stops, and then
deletes several TNonblockingServers. Ran it under valgrind, and now it
only complains about any active connections being leaked.
Revert Plan:
OK
---
lib/cpp/src/server/TNonblockingServer.cpp | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
