GitLab

crrev.com/c/2076431 revealed an underlying race condition from the
"TestDeviceServiceDisconnect" test case, where there is essentially a
race between when the IsolatedVRDeviceProvider receives the disconnect
from the underlying device provider and when the BrowserXrRuntime would
receive it's own disconnect from the underlying device.

If the IsolatedVRDeviceProvider wins the race, the BrowserXRRuntime may
be removed before it can notify the VRService that it (and the session
it is responsible for) is being ended. Because the renderer process has
its own pipe to the device process, it is notified of the disconnect
(and ended session), even if the Browser doesn't till it about this
termination. The end result is that the only state that doesn't end up
eventually getting cleaned up is that of the session metrics helper,
which thinks that a session is still active, and may hit a DCHECK when
attempting to start a new session.

Change-Id: I5a448130cab0bc245cd4f3e6ef77e44512a89652
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2081528
Commit-Queue: Alexander Cooper <alcooper@chromium.org>
Auto-Submit: Alexander Cooper <alcooper@chromium.org>
Reviewed-by: Klaus Weidner <klausw@chromium.org>
Cr-Commit-Position: refs/heads/master@{#746992}