Re: Just a thought - linux raid wiki - raid 5 grows getting stuck at 0%

[NeilBrown <neilb@suse.com>]

[-- Attachment #1: Type: text/plain, Size: 2323 bytes --]

On Thu, Nov 17 2016, Wols Lists wrote:

> I've just been doing a bit of work on the wiki when I put 2 and 2
> together, and hope I haven't made 5 ...
>
> Bitmaps interfere with grow operations, I believe ...

Used to, yes.  Don't any more.

>
> And mdadm has recently been modified so that bitmaps are switched on by
> default, I also believe ...
>
> Could this be silently blocking the grow, so you don't get any error and
> it just sits there doing nothing?

Nope.  If it was a problem it would explicitly fail.

>
> But that brings up two points. Should mdadm now explicitly look for a
> bitmap when growing an array, and warn that the bitmap needs to be
> disabled? Apparently it currently comes up with some errors that hint at
> the cause rather than explicitly say so.

If you try to reshape an array which has a bitmap, on a kernel that
doesn't support reshaping arrays with bitmaps, mdadm hits this line of
code:

			if (err == EBUSY &&
			    (array.state & (1<<MD_SB_BITMAP_PRESENT)))
				cont_err("Bitmap must be removed before size can be changed\n");

or maybe this one

		pr_err("Cannot set array shape for %s\n",
		       devname);
		if (err == EBUSY &&
		    (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
			cont_err("       Bitmap must be removed before shape can be changed\n");

or

			if (err == EBUSY &&
			    (array.state & (1<<MD_SB_BITMAP_PRESENT)))
				cont_err("Bitmap must be removed before level can be changed\n");



>
> And secondly, I'm sure someone sent an email to the list that explained
> when a bitmap was created. I can't find it in my archives. The man page
> says it's created if the array is over 100G, but I'm sure the email gave
> rather more detail than that and that the figure actually varied.

	if (!s->bitmap_file &&
	    s->level >= 1 &&
	    st->ss->add_internal_bitmap &&
	    (s->write_behind || s->size > 100*1024*1024ULL)) {
		if (c->verbose > 0)
			pr_err("automatically enabling write-intent bitmap on large array\n");
		s->bitmap_file = "internal";
	}

So it looks like "over 100G" is the only test.

>
> But I'm sure you can see from this that I'm updating this bit of the
> wiki. Is there anything else I ought to know about bitmaps, so I can
> document it? :-)

If only I could run "diff" between my brain and the wiki....

Thanks for doing this!

NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 800 bytes --]