I have a RR replay of this bug. I'm looking at it.
Turns out we sync decode on reftest draw. However if there's an error the image frame might not be destroyed on time during the sync decode (it happens earlier during style flush). So if this race occurs we get a failure. There's apparently a bug that tracks this issue somewhere else. ni? seth to find the proper dependency.
Bulk assigning P3 to all open intermittent bugs without a priority set in Firefox components per bug 1298978.
5 failures in 623 pushes (0.008 failures/push) were associated with this bug in the last 7 days. Repository breakdown: * autoland: 3 * mozilla-inbound: 2 Platform breakdown: * linux64: 5 For more details, see: https://brasstacks.mozilla.com/orangefactor/?display=Bug&bugid=1226748&startday=2016-11-21&endday=2016-11-27&tree=all
7 failures in 694 pushes (0.01 failures/push) were associated with this bug in the last 7 days. Repository breakdown: * mozilla-inbound: 6 * autoland: 1 Platform breakdown: * linux64: 7 For more details, see: https://brasstacks.mozilla.com/orangefactor/?display=Bug&bugid=1226748&startday=2016-11-28&endday=2016-12-04&tree=all