Re: [PATCH v2 2/2] nvme/host/core: Allow overriding of wait_ready timeout

From: "Singh, Balbir" <sblbir@amazon.com>
To: "kbusch@kernel.org" <kbusch@kernel.org>,
	"bvanassche@acm.org" <bvanassche@acm.org>
Cc: "axboe@fb.com" <axboe@fb.com>,
	"sblbir@amzn.com" <sblbir@amzn.com>, "hch@lst.de" <hch@lst.de>,
	"linux-nvme@lists.infradead.org" <linux-nvme@lists.infradead.org>,
	"sagi@grimberg.me" <sagi@grimberg.me>
Subject: Re: [PATCH v2 2/2] nvme/host/core: Allow overriding of wait_ready timeout
Date: Tue, 17 Sep 2019 05:02:10 +0000	[thread overview]
Message-ID: <34b3e7e974346816f14eb30dc35fdcddde6e6867.camel@amazon.com> (raw)
In-Reply-To: <ee9a3b5b-2572-5022-0322-e1e2ae38904b@acm.org>

On Mon, 2019-09-16 at 20:17 -0700, Bart Van Assche wrote:
> On 9/16/19 7:56 PM, Singh, Balbir wrote:
> > On Mon, 2019-09-16 at 19:14 -0600, Keith Busch wrote:
> > [ ... ]
> > In my case I was doing a simple mirror (using madadm across two nvme
> > devices) and when I get timeouts on one, I need to wait up to 128 seconds
> > before switching over. Ideally I want this to be really fast and drop
> > the slow broken controller.
> > 
> > > If you really need the driver to do spec non-compliant behavior, we have
> > > quirks for that.
> > 
> > I like the quirks approach, but it assumes the timeout value is not
> > variable, but rather fixed by the quirk. I was attempting to really have
> > mirrored IO timeout quickly on a bad device.
> 
> Other Linux kernel storage transports (FC, SRP) decouple the failover 
> timeout from the I/O timeout. See also the output of git grep -nH 
> fast_io_fail for the kernel source tree. See also the documentation of 
> fast_io_fail_tmo in https://linux.die.net/man/5/multipath.conf. Maybe we 
> need something similar for NVMe?
> 

Great suggestion!

I tried something similar with wait_ready, isolating it from io_timeout,
since it is sort of the bottleneck w.r.t. recovery in a mirrored
configuration (timeouts on one of the volumes)

In this case it needs to be system-wide to remove the non-responsive
controller within the speculated timeout, not everyone needs it, but
it sort of gives similar behaviour to (FC, SRP) failure in a multipath
setup. I am not too familiar with FC, SRP setup, but it seems like
we should provide an aggregated timeout for full reset work on failure/
timeouts. Would that make sense to try and build or did you have some
other design in mind?

> Bart.

Cheers,
Balbir Singh.
_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme