Re: [HPDD-discuss] Using multiple NIDs for the same failnode

Aurélien Thanks for providing this detailed advise. One slight correction: although 2.5.x will have scheduled quarterly maintenance releases, we do still support many sites on 2.4.x releases and so it is quite possible that we will issue a 2.4.3 release at some point. If/when we do I would certainly expect to include a fix for LU-4460. Regards Peter On 3/4/14, 5:34 AM, "DEGREMONT Aurelien" <aurelien.degremont@cea.fr<mailto:aurelien.degremont@cea.fr>> wrote: Le 04/03/2014 04:47, Adesanya, Adeyemi a écrit : This looks a lot like the bug described in LU-4460. Is there a fix/workaround for this in a current maintenance release? This is exactly what is described in ticket LU-4460 I reported and was just recently fixed. Fix versions: Lustre 2.6.0<https://jira.hpdd.intel.com/browse/LU/fixforversion/10595>;, Lustre 2.5.1<https://jira.hpdd.intel.com/browse/LU/fixforversion/10608> AFAIK 2.4 is no more supported, 2.4.2 was the last one. You should either rebuild 2.4.2 with this patch, or upgrade to 2.5.1 when it will out (end of march). 2.5 is the current maintenance release and it is fixed in the new one. As a workaround, as your are updating from 2.1.4, you just need to change the mgsnid of your target to be ok. failover NID are only used when building MGS config file. But, as you are upgrading, those files are already present on MGS. So, I advice you to use tunefs.lustre to only keep one NID for your MGS and wait for a Lustre version with this fix to re-enable it. Aurélien