On Mon, Feb 08, 2021 at 05:43:20PM +0300, Alexey Dobriyan wrote:
> On Fri, Feb 05, 2021 at 12:53:32AM +0000, Dmitry Fomichev wrote:
> > I like the general idea of this patch. Adding zone units looks useful,
> > especially for ZNS workloads.
>
> OK -- tests, wording.
>
> > > --- a/filesetup.c
> > > +++ b/filesetup.c
> > > @@ -1029,6 +1029,35 @@ int setup_files(struct thread_data *td)
> > > if (o->read_iolog_file)
> > > goto done;
> > >
> > > + if (td->o.zone_mode == ZONE_MODE_ZBD) {
> >
> > What about ZONE_MODE_STRIDED? Shouldn't 'z' suffix be enabled for it too?
>
> I converted zoneskip= as well, but the code asserts somewhere else.
> I didn't figure why exactly. zoneskip= and strided mode can be enabled
> separatedly.
>
> > > + struct fio_file *f;
> > > + int i;
> > > +
> > > + err = zbd_init_files(td);
> > > + if (err)
> > > + goto err_out;
> > > +
> >
> > zbd_init_files() is only called once. Why not fold the loop below into zbd_init_files()
> > since the code is ZBD-specific?
>
> OK.
I take that back!
The reason for a split is that size= and offset= and io_size= must be
available before those checks:
/*
* now file sizes are known, so we can set ->io_size. if size= is
* not given, ->io_size is just equal to ->real_file_size. if size
* is given, ->io_size is size / nr_files.
*/
so the init is done in 3 parts:
optionally create/alloc f->zbd_info
recalculate if size=/io_size=/offset=/... is supplied in zones
...
calculate ->min_zone/->max_zone, etc
but size recalculation is necessary for zonemode=strided as well.
