[dm-devel] dm mpath: Add timeout mechanism for queue_if_no_path

[Gabriel Krisman Bertazi]

Mike Snitzer <snitzer at redhat.com> writes:

> On Mon, Jan 06 2020 at  4:52pm -0500,
> Khazhismel Kumykov <khazhy at google.com> wrote:
>
>> On Mon, Jan 6, 2020 at 11:28 AM Mike Snitzer <snitzer at redhat.com> wrote:
>> >
>> > On Thu, Jan 02 2020 at  5:45pm -0500,
>> > Gabriel Krisman Bertazi <krisman at collabora.com> wrote:
>> >
>> > > From: Anatol Pomazau <anatol at google.com>
>> > >
>> > > Add a configurable timeout mechanism to disable queue_if_no_path without
>> > > assistance from multipathd.  In reality, this reimplements the
>> > > no_path_retry mechanism from multipathd in kernel space, which is
>> > > interesting for cases where we cannot rely on a daemon being present all
>> > > the time, in case of failure or to reduce the guest footprint of cloud
>> > > services.
>> > >
>> > > Despite replicating the policy configuration on kernel space, it is
>> > > quite an important case to prevent IOs from hanging forever, waiting for
>> > > userspace to behave correctly.
>> > >
>> > > Co-developed-by: Frank Mayhar <fmayhar at google.com>
>> > > Signed-off-by: Frank Mayhar <fmayhar at google.com>
>> > > Co-developed-by: Bharath Ravi <rbharath at google.com>
>> > > Signed-off-by: Bharath Ravi <rbharath at google.com>
>> > > Co-developed-by: Khazhismel Kumykov <khazhy at google.com>
>> > > Signed-off-by: Khazhismel Kumykov <khazhy at google.com>
>> > > Signed-off-by: Anatol Pomazau <anatol at google.com>
>> > > Co-developed-by: Gabriel Krisman Bertazi <krisman at collabora.com>
>> > > Signed-off-by: Gabriel Krisman Bertazi <krisman at collabora.com>
>> >
>> > This seems like a slippery slope.
>> >
>> > I've heard this line of dm-multipath concern from Google before
>> > (e.g. doesn't want to rely on availability of userspace component).
>> >
>> > Thing is, to properly use dm-multipath (e.g. to reinstate failed paths)
>> > multipathd really is needed no?
>> Yeah, in order to reinstate the failed paths, or any full recovery, we
>> do need the user space component to be running, and this doesn't aim
>> to change the responsibilities here. Rather, we're aiming to avoid
>> in-kernel hangs/processes lingering indefinitely in unkillable sleep
>> due to buggy/unavailable/down userspace multipath daemon.
>
> Sorry but the above patch header clearly states:
> "or to reduce the guest footprint of cloud services"
>
> Which implies that multipathd isn't available by design in the
> referenced environment.
>
>> >
>> > If not, how is it that the proposed patch is all that is missing when
>> > multipathd isn't running?  I find that hard to appreciate.
>> >
>> > So I'm inclined to not accept this type of change.  But especially not
>> > until more comprehensive context is given for how Google is using
>> > dm-multipath without multipathd.
>> 
>> This patch isn't aimed at enabling dm-multipath without a userspace
>> multipath daemon, rather to avoid processes hanging indefinitely in
>> the event the daemon is unable to proceed (for whatever reason).
>
> Again, I don't buy it given the patch header explicitly says
> dm-multipath could be deployed in the cloud without the benefit of
> multipathd running.

Hey Mike,

I believe that was my fault, I misunderstood the google's use case, when I
modified the commit message.

> But I'll meet you half-way to start: rather than make the timeout
> configurable on a per multipath table basis, please just set a longer
> stopgap default and allow that timeout value to be configured with a
> module parameter (e.g. dm_multipath.queue_if_no_path_timeout=120,
> and setting it to 0 disables the timeout).
>
> This would follow the same pattern that was established by DM
> thin-provisioning's no_space_timeout modparm with these commits:
> 85ad643b dm thin: add timeout to stop out-of-data-space mode holding IO forever
> 80c57893 dm thin: add 'no_space_timeout' dm-thin-pool module param
>
> That'd save us from having to do a bunch of fiddley DM multpath table
> parsing for now.  But if for some crazy reason in the future it is
> determined that a longer duration stop-gap timeout cannot be
> one-size-fits-all we can layer per device configuration in at that
> time.
>
> How does that sound?

That seems reasonable to me.  Let me see what Khazhismel thinks and I
can follow up with a v2.

-- 
Gabriel Krisman Bertazi