event manager: Don't worry if attempt to wake dead manager fails
authorBen Gamari <bgamari.foss@gmail.com>
Tue, 17 Jan 2017 20:52:37 +0000 (15:52 -0500)
committerBen Gamari <ben@smart-cactus.org>
Tue, 17 Jan 2017 20:56:27 +0000 (15:56 -0500)
commitd5cd505bc484edee3dbd5d41fb7a27c2e18d528d
tree07b0bb680d1f1393940d703708bf8f0183549342
parente7e5f7accbb7d9a12aee5d1468371a8ba09b598d
event manager: Don't worry if attempt to wake dead manager fails

This fixes #12038, where the TimerManager would attempt to wake up a
manager that was already dead, resulting in setnumcapabilities001
occassionally failing during shutdown with unexpected output on stderr.

I'm frankly still not entirely confident in this solution but perhaps it
will help to get a few more eyes on this.

My hypothesis is that the TimerManager is racing:

  thread                   TimerManager worker
  -------                  --------------------
  requests that thread
  manager shuts down

                           begins to clean up,
                           closing eventfd

  calls wakeManager,
  which tries to write
  to closed eventfd

To prevent this `wakeManager` will need to synchronize with the
TimerManger worker to ensure that the worker doesn't clean up the
`Control` while another thread is trying to send a wakeup. However, this
would add a bit of overhead on every timer interaction, which feels
rather costly for what is really a problem only at shutdown.  Moreover,
it seems that the event manager (e.g.  `GHC.Event.Manager`) is also
afflicted by a similar race.

This patch instead simply tries to catch the write failure after it has
happened and silence it in the case that the fd has vanished. It feels
rather hacky but it seems to work.

Test Plan: Run `setnumcapabilities001` repeatedly

Reviewers: hvr, austin, simonmar

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D2957

GHC Trac Issues: #12038
libraries/base/GHC/Event/Control.hs