[ofiwg] sockets fabtests failure in testing for 1.4.0 release

Ilango, Arun arun.ilango at intel.com
Fri Oct 28 13:59:03 PDT 2016

Running fabtests 1.4 against libfabric 1.4 for verbs between two nodes is fine (except for expected failures).

1. Running fabtests on loopback address shows rdm tests failing.
2. Testing forward compatibility of fabtests 1.3 against libfabric 1.4 shows some failures:

fi_poll: -FI_ENOSYS
fi_rdm_rma -o writedata: -FI_ENOSYS.

#2 seems like a test issue rather than provider issue. It's probably something to do with handling the return code from the test. These tests are marked as NOTRUN in fabtests 1.4 against libfabric 1.4

Since it's too late to fix #1 and it is not critical, we should go ahead and do the release and fix the issue later.


-----Original Message-----
From: Jeff Squyres (jsquyres) [mailto:jsquyres at cisco.com] 
Sent: Friday, October 28, 2016 1:00 PM
To: Hefty, Sean <sean.hefty at intel.com>
Cc: Hossain, Shantonu <shantonu.hossain at intel.com>; Ilango, Arun <arun.ilango at intel.com>; OFIWG Mailing list <ofiwg at lists.openfabrics.org>
Subject: Re: sockets fabtests failure in testing for 1.4.0 release

Ok.  AFAIK, Arun's confirmation is the last item I need before posting the tarballs / etc.

> On Oct 28, 2016, at 3:40 PM, Hefty, Sean <sean.hefty at intel.com> wrote:
>> Ok, so just to be clear: this was a bug in fabtests, and we *don't* 
>> have a forward incompatibility issue with the sockets provider.  Right?
> correct
>> If so, are we waiting for this forward compatibility test with MXM, 
>> UDP, and verbs?  Or releasing without their say-so?
> UDP was verified.  I updated the issue, but I would ignore mxm.  Arun has likely already checked verbs, but please let him verify.
> - Sean

Jeff Squyres
jsquyres at cisco.com
For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

More information about the ofiwg mailing list