Re: [PATCH] bcache: make stripe_size configurable and persistent for hardware raid5/6

[Eric Wheeler]

On Fri, 7 Jan 2022, Eric Wheeler wrote:
> On Fri, 7 Jan 2022, Martin K. Petersen wrote:
> > Eric,
> > 
> > > Even new new RAID controlers that _do_ provide `io_opt` still do _not_ 
> > > indicate partial_stripes_expensive (which is an mdraid feature, but Martin 
> > > please correct me if I'm wrong here).
> > 
> > partial_stripes_expensive is a bcache thing, I am not sure why it needs
> > a separate flag. It is implied, although I guess one could argue that
> > RAID0 is a special case since partial writes are not as painful as with
> > parity RAID.
> 
> I'm guessing bcache used did some optimization for 
> queue->limits.raid_partial_stripes_expensive because md raid5 code sets 
> this flag.  At least when using Linux md as the RAID5 implementation it 
> gets configured automatically:
>    raid5.c:       mddev->queue->limits.raid_partial_stripes_expensive = 1;
> 
> https://elixir.bootlin.com/linux/latest/source/drivers/md/raid5.c#L7729
> 
> Interestingly only bcache uses it, but md does set it.

Ok so `git blame` shows that Kent added this to md/raid5.c in 
c78afc6261b (Kent Overstreet 2013-07-11 22:39:53 -0700 7526)
	mddev->queue->limits.raid_partial_stripes_expensive = 1;

    bcache/md: Use raid stripe size
    
    Now that we've got code for raid5/6 stripe awareness, bcache just needs
    to know about the stripes and when writing partial stripes is expensive
    - we probably don't want to enable this optimization for raid1 or 10,
    even though they have stripes. So add a flag to queue_limits.

Kent, Martin:

Do you think we should leave the md-specific 
raid_partial_stripes_expensive setting and require users of RAID 
controllers to set the bit themselves in bcache---or---remove all 
raid_partial_stripes_expensive code and always treat writes as "expensive" 
when `opt_io` is defined?

--
Eric Wheeler


> 
> > The SCSI spec states that submitting an I/O that is smaller than io_min
> > "may incur delays in processing the command". And similarly, submitting
> > a command larger than io_opt "may incur delays in processing the
> > command".
> > 
> > IOW, the spec says "don't write less than an aligned multiple of the
> > stripe chunk size" and "don't write more than an aligned full
> > stripe". That leaves "aligned multiples of the stripe chunk size but
> > less than the full stripe width" unaccounted for. And I guess that's
> > what the bcache flag is trying to capture.
> 
> Maybe any time io_opt is provided then partial_stripes_expensive should be 
> flagged too and any code to the contrary should be removed?
> 
> Question: Does anyone have a reason to keep partial_stripes_expensive in 
> the kernel at all?
> 
> > SCSI doesn't go into details about RAID levels and other implementation
> > details which is why the wording is deliberately vague. But obviously
> > the expectation is that partial stripe writes are slower than full.
> > 
> > In my book any component in the stack that sees either io_min or io_opt
> > should try very hard to send I/Os that are aligned multiples of those
> > values. I am not opposed to letting users manually twiddle the
> > settings. But I do think that we should aim for the stack doing the
> > right thing when it sees io_opt reported on a device.
> 
> Agreed, thanks for the feedback!
> 
> -Eric
> 
> 
> > 
> > -- 
> > Martin K. Petersen	Oracle Linux Engineering
> > 
>