From mboxrd@z Thu Jan  1 00:00:00 1970
Content-Type: multipart/mixed; boundary="===============4272381162792312830=="
MIME-Version: 1.0
From: Walker, Benjamin <benjamin.walker at intel.com>
Subject: Re: [SPDK] Github issue #407 and libibverbs events
Date: Mon, 10 Sep 2018 16:46:34 +0000
Message-ID: <19155b4fea7de44e231c7b168d972b1f66e2b9ca.camel@intel.com>
In-Reply-To: 122B872DCF83AB4DB816E25A2C1AD08D8BB6CF5C@IRSMSX102.ger.corp.intel.com
List-ID: <spdk@lists.01.org>
To: spdk@lists.01.org

--===============4272381162792312830==
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable

On Mon, 2018-09-10 at 08:54 +0100, Szwed, Maciej wrote:
> Hi guys,
> I was investigating github issue #407 and found out that we don=E2=80=99t=
 manage
> events in NVMf host side (perf in this case). We do that for NVMf target =
->
> spdk_nvmf_process_cm/ib_event(). We should probably implement new poller =
that
> will handle incoming events in host side and change code accordingly. Wha=
t do
> you think?
>  =

> Here is documentation for reference:
> https://pdfs.semanticscholar.org/9584/17ad2d476c0a4821925cf8474fabcb1229c=
e.pdf
> Chapter 3.7
>  =


Yes, we should be doing this. It is theoretically possible to transparently
recover from connection errors if we were to handle them on both the target=
 and
the initiator side. Now that the target logic is in place, it's a great tim=
e to
go improve the initiator behavior.

> Regards,
> Maciek


--===============4272381162792312830==--