Re: [PATCH v3] nvme-fabrics: reject I/O to offline device

[James Smart <james.smart@broadcom.com>]

On 2/20/2020 12:34 AM, Sagi Grimberg wrote:
>
>>> +static void nvme_failfast_work(struct work_struct *work)
>>> +{
>>> +    struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
>>> +            struct nvme_ctrl, failfast_work);
>>> +
>>> +    spin_lock_irq(&ctrl->lock);
>>> +    if (ctrl->state == NVME_CTRL_CONNECTING) {
>>> +        set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
>>> +        dev_info(ctrl->device, "failfast expired set for controller 
>>> %s\n", ctrl->opts->subsysnqn);
>>
>> Please break up the line.
>>
>> But looking at the use of NVME_CTRL_FAILFAST_EXPIRED, it almost seems
>> like this is another controller state?
>
> It actually is a controller state. this is just adding another state
> without really fitting it into the state machine. I'd personally
> would want that to happen, but I know James had some rejects on
> this.

I don't believe its a controller state - rather an attribute, relative 
to within the Connecting state (which may be in the process of 
retrying), where instead of busying any io request that may be received, 
we start to fail them.    I don't see this time window as something the 
controller has to actually transition through - meaning we aren't 
reconnecting while in this state.


>
> Also, I still say that its default changes the existing behavior which
> is something we want to avoid.

I thought the last patch was the same as existing behavior (e.g. default 
tmo is 0, and if 0, the timer, which sets the now-fail flag, never gets 
scheduled).

in the last review, I was exploring the option if we did want to change 
the default, as the default is a rather long wait (I did vacilate).  But 
I'm fine with keeping things the same.

-- james


_______________________________________________
linux-nvme mailing list
linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme