binderRpcTest: fix OnewayStressTest flake
On slow devices, the decref from asynchronous calls wouldn't be
processed in time. So, making multiple simultaneous twoway calls to make
sure we wait for every server thread to finish processing everything.
This could still flake if a device delayed for >500ms, but for this
amount of time performance can be considered part of a correctness.
If there are further issues, we can add additional APIs which let us
control how calls are scheduled so we can make sure to process a two-way
transaction on every thread.
Fixes: 200173589
Test: binderRpcTest OnewayStressTest on repeat
Change-Id: I719499a14530f2108644a814c5df2d1718893793
1 file changed