refactor tests using INVM acceptors/connectors

[jmesnil]

We have a lot of tests which uses the INVM acceptors.
All these tests shares the same INVMRegistry singleton.

If a test fails and does not clean up its acceptor(s), all the subsequent tests which try to register an INVM acceptor with the same ID will fail.
1 failing test will result in hundred of failing tests making it very difficult to see regression, esp. since we have a few intermittent failing tests using INVM acceptors.

I propose that we refactor every tests using INVM acceptor so that:
* in the setUp, we clean up the INVM registry (we don't care what happened before)
* in the tearDown, we check that the INVM registry size is set to 0

I'll add a InVMTestCase and use it to extend every tests using INVM acceptor.
wdyt?

[jmesnil]

I'll also add clearing the InVMRegistry in the setUp() of UnitTestCase

[clebert.suconic]

"jmesnil" wrote:
I'll also add clearing the InVMRegistry in the setUp() of UnitTestCase

don' t we have tests that share a server start?

Also: we have lots of tests where tearDown is not calling its super method.

[jmesnil]

don' t we have tests that share a server start?

I'm not aware of any such test. Do you have an example?

Also: we have lots of tests where tearDown is not calling its super method.

Yes, I'm fixing it when I see setUp/tearDown w/o the call to the super methods.

The idea is to prevent all the cascading failures asap.
I've got a regression in the failover tests and it is impossible to check what's going on with all the noisy cascading failures