Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
0.8, 0.9
-
Debian, Redhat, Windows
-
Patch Available
Description
During deployment, we noticed a bug inside the new (optional) boost thread replacement of pthread.
It moves the line:
state_ = starting;
from after the creation of the boost::thread to immediately before it. The bug was that one of the first pieces of threadMain (the thread routine of the thread being created) is:
if (thread->state_ != starting)
{ return (void*)0; }So there was a race condition between the line that set state_ to "starting", and the line that checked to make sure that it was "starting". That ended meaning that sometimes calling "start()" would not result in the thread's runnable being called.
Testing: this was tested in production, but NOT with thrift test suite (however I believe ./configure --enable_boostthreads and testing will cover this)