On Mon, Sep 11, 2023 at 11:00:17AM +0000, Shinichiro Kawasaki wrote:
> > nvmf_wait_for_state "${def_subsysnqn}" "live"
> > nvmedev=$(_find_nvme_dev "${def_subsysnqn}")
> >
> > We could make this a bit more generic and move it into the connect
> > helper. What do you think?
>
> This nvme state file check looks a bit complicated, but looks more robust than
> "nvme connect" command exit status check. Do you think that "nvme connect" can
> fail even when "nvme connect" command returns good status? If so, your approach
> will be the way to choose.
Currently, 'nvme connect' is a synchronous call for tcp/rdma but not for
fc. 'nvme connect' for tcp/rdma will report an error if something is
wrong but not for fc, because it always return successfully.
The nvme/005 is exposing the behavior differences between the
transports. My long time goal is to address and make all transport
behave the same way (unification of the state machines). But as it
currently stands fc would need someting like this to make sure we are
not blindly reporting success just because the 'nvme connect' call is
successful.
This type of check would make the test suite more robust and better in
detecting errors.
