Commit 30d8a3d6 authored by Louis Brandy's avatar Louis Brandy Committed by Alecs King

fix race in ~ThreadWheelTimekeeper

Summary:
During destruction, we enqueue a callback to `eventBase_` that references `wheelTimer_`.

So long as that callback exists, we have a circular dependency as both reference the other and we -cannot- safely destroy this object (and its members). The fix is to break this dependency someway. In this case, I've chosen to simply `AndWait` until the eventBase has consumed the message referencing the timer. From here normal destruction of members can proceed safely (destorying HHWheelTimer first, EventBase second).

The problem in the current code is that the Eventbase will attempt to consume all messages during it's own destruction. If `wheelTimer_->cancelAll()` is still enqueued, it will attempt to use the now destroyed `wheelTimer_`.

Test Plan:
Running this test repeatedly was able to repro:

./folly/futures/futures-test --gtest_filter=Timekeeper.futureWithinVoidSpecialization

Reviewed By: yfeldblum@fb.com

Subscribers: enis, folly-diffs@, jsedgwick, yfeldblum, darshan, chalfant

FB internal diff: D1985967

Tasks: 6332729, 6741095

Signature: t1:1985967:1428726270:997ec277c6a73554e54b8cf673acd36ff62976e6
parent f22baf0e
......@@ -70,7 +70,7 @@ ThreadWheelTimekeeper::ThreadWheelTimekeeper() :
}
ThreadWheelTimekeeper::~ThreadWheelTimekeeper() {
eventBase_.runInEventBaseThread([this]{
eventBase_.runInEventBaseThreadAndWait([this]{
wheelTimer_->cancelAll();
});
eventBase_.terminateLoopSoon();
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment