Re: [PATCH v3 9/9] tests/qtest: massively speed up migration-test

[Peter Xu]

On Thu, Jun 01, 2023 at 05:36:42PM +0100, Daniel P. Berrangé wrote:
> > The problem is non-live is already not real world usage in most cases.  It
> > seems we all agreed that it's the code paths to cover not real world usages
> > in the tests, or maybe not?
> 
> The cases that i made run non-live are exercising code paths that
> are only relevant during migration setup, so the non-live vs live
> status is irrelevant to the code coverage attained for those tests.
> Other remaining live tests give sufficient coverage for the live
> scenario

IMHO actually I think it's good we discuss about the goal of target of what
qtest plays a role here, hitting the issue of over-used CI and test running
too long.

For migration tests I'm not sure how Juan sees this I'd say it's good
enough to make it "just to cover code paths as much as possible".

We're never covering real use cases anyway, IMHO, if we start with 128M VM
running sequential dirty workload only.  We still need to rely on real
migration tests carried out elsewhere I believe.

So I'd say we're fine to leverage a new parameter especially if the
parameter is simple enough (for logical change I only expect a few LOCs
with switchover-hold) to help us achieve it.

Let's keep that on top anyway, no matter what.  Thanks.

-- 
Peter Xu