1510026 - Make crashtests flush the compositor after every test

Status: RESOLVED FIXED

(Testing :: Reftest, enhancement)

Description

[Kartikaya Gupta (email:[e-mail])]

The crashtests are implemented as "load only" reftests, which means they run with the reftest harness and don't do any snapshotting. With e10s/OMTC this means that the reftest harness is mostly completely asynchronous from the actual compositor work, and so if the compositor gets wedged or crashes or whatever, the test harness doesn't know about it until later (possibly much later). It seems that often the next test that does some sync IPC (via AdvanceTimeAndRefresh or similar) ends up the one manifesting the failure.

We should make the reftest harness more robust to this sort of thing, by forcing a round-trip to the compositor that flushes everything.

Comment 1

[Kartikaya Gupta (email:[e-mail])]

https://treeherder.mozilla.org/#/jobs?repo=try&group_state=expanded&revision=e202de36903c8e5055e62b7f41c499798acc5920

Comment 2

[Kartikaya Gupta (email:[e-mail])]

Also https://treeherder.mozilla.org/#/jobs?repo=try&group_state=expanded&revision=545cdeba0a1329526030282cb20a6bacfce70a89 to make sure there's no lurking failures on win/mac for qr

Comment 3

[Kartikaya Gupta (email:[e-mail])]

Attachment: Bug 1510026 - Send a sync IPC round-trip to the compositor after every crashtest. r?mattwoodrow

This sends a sync IPC message to the compositor after each crashtest, to
ensure that any stuff inflight in the compositor settles down. In
particular this should round-trip through all of the relevant compositor
threads (i.e. for WebRender it should ensure the scene builder, render
backend, and renderer threads all get flushed).

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/2d25f33db080
Send a sync IPC round-trip to the compositor after every crashtest. r=mattwoodrow

Comment 5

[Cosmin Sabou [:CosminS]]

https://hg.mozilla.org/mozilla-central/rev/2d25f33db080