[ofa-general] Re: OpenSM detection of duplicated GUIDs on loopback

[Sasha Khapyorsky]

On 12:00 Sun 29 Jul     , Eitan Zahavi wrote:
> > On 14:27 Fri 27 Jul     , Eitan Zahavi wrote:
> > > The problem I have with back-to-back plug is that it is a 
> > fatal case 
> > > if found in a case where there was no use of this plug.
> > > So we will  need some sort of user input if it is OK or not.
> > 
> > Ok, and let's add cl_qmap_count() check there.
> Not following you.

With back-to-back network cl_qmap_count(&sw_guid_tbl) should be 0.

> > > The case of moving a port in the middle of a sweep can be easily 
> > > detected if instead of reporting an error a second check of the 
> > > original DR where the same GUID was found is performed...
> > 
> > Do you mean to resend NodeInfo request to the original location?
> > Assuming so, I guess it should be instead of second heavy 
> > sweep, and it is a good idea. The only small downside of this 
> > I can see is potential timeouts (and discovery slowdown). But 
> > anyway it is much better then fatal error. Thanks!
> 
> So we are inline with this one .
> Instead of changing the order of things we could generate list of DR's
> that are to be re-scanned
>  during drop-mgr and then abort if really dulicates.

I will need to look at code...

Sasha