The point of the test is that the client returns immediately while the server is blocking. Removing server sleeps would let incorrect oneway client implementations pass the test.
The server process is killed at the end so ideally we could avoid waiting for the sleep. But in reality I think for many test servers we end up waiting for the sleep before the subsequent RPC processing because they are single-threaded. On Sun, Nov 19, 2017 at 2:16 AM James E. King, III <[email protected]> wrote: > Does anybody know why the cross tests have a built-in sleep in the one-way > test? This slows down every test, seemingly unnecessarily. When I removed > it from the cpp tests, the cpp server tests all zoom along nicely. It's > present in c_glib server, for example, and also in go, and probably > others. If we don't need all the tests to be sleeping 1 second, we should > remove all the sleeping to speed up the tests. We do 2,000+ cross tests, > so that's a lot of sleep. >
