Will the write request starve?

Will the write request starve?

[Xiao Ni]

Hi all

   I meet a problem and I can't find the answer. Could you help me?
When raid5d want to handle a request, it need to get stripes by __get_priority_stripe.
In the __get_priority_stripe function, it always try to get stripe from handle_list. 

   Make an assumption that the raid of level 5 is assembled with 4 disks, and the raid 
is normal. Now the a write request just want to write a single chunk, so we choose RMW 
and the stripe will put on the delay_list, then activate it and put it to hold_list. 
But if there are many read requests after the write request, when raid5d handle stripes
it will get stripe from handle_list. So will the write request starve? 

   Certainly I know it will not, because many tests have been done. But I can't find 
the reason why it will not starve. 

   And there is a variable bypass_count, what's the function about it? If bypass_count <
bypass_threshold, it can get a chance to wait for the next write request to get a full write.
So avoid read requests. Am I right?

Best Regards
Xiao

Re: Will the write request starve?

[Xiao Ni]

----- Original Message -----
> From: "Xiao Ni" <xni@redhat.com>
> To: linux-raid@vger.kernel.org
> Cc: "jes sorensen" <jes.sorensen@redhat.com>
> Sent: Wednesday, September 3, 2014 1:59:08 PM
> Subject: Will the write request starve?
> 
> Hi all
> 
>    I meet a problem and I can't find the answer. Could you help me?
> When raid5d want to handle a request, it need to get stripes by
> __get_priority_stripe.
> In the __get_priority_stripe function, it always try to get stripe from
> handle_list.
> 
>    Make an assumption that the raid of level 5 is assembled with 4 disks, and
>    the raid
> is normal. Now the a write request just want to write a single chunk, so we
> choose RMW

Sorry, it'll choose RCW

> and the stripe will put on the delay_list, then activate it and put it to
> hold_list.
> But if there are many read requests after the write request, when raid5d
> handle stripes
> it will get stripe from handle_list. So will the write request starve?
> 
>    Certainly I know it will not, because many tests have been done. But I
>    can't find
> the reason why it will not starve.
> 
>    And there is a variable bypass_count, what's the function about it? If
>    bypass_count <
> bypass_threshold, it can get a chance to wait for the next write request to
> get a full write.
> So avoid read requests. Am I right?
> 
> Best Regards
> Xiao
> --
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>