On Aug 16, 2023 / 16:14, Daniel Wagner wrote:
> On Wed, Aug 16, 2023 at 12:04:24PM +0000, Shinichiro Kawasaki wrote:
[...]
> > 2) Rework _find_nvme_dev into two new functions _find_nvme_ctrl_dev and
> > _find_nvme_ns_dev, and do the readiness check in _find_nvme_ns_dev.
> > IMO, this confusion comes from the fact that _find_nvme_dev returns control
> > device, but some test cases use it to operate namespaces by adding "n1" to
> > the control device name. If a test case uses namespace device, it's the
> > better to call _find_nvme_ns_dev. But I worry this approach may be too much.
>
> As we already have an argument parser in _nvme_connect_subsys, we could
> also introduce a new option which allows to select the wait type. With
> this _nvmet_passtrhu_target_connect could be something like
>
> _nvmet_passthru_target_connect() {
> [...]
>
> _nvme_connect_subsys "${trtype}" "${subsys_name}" \
> --wait-for=device || return
>
> [...]
> }
>
> and for the rest of the test cases we just set the default for
> --wait-for to ns.
Thanks. This idea sounds good. It will fix the failure and avoid the 1 second
wait. Will implement this and send out as v2.
