Re: CRUSH map help

Stephen Mercier <stephen.mercier@xxxxxxxxxxxx> · Thu, 12 May 2016 14:57:37 -0700

Also, when you say new enough for the overselection hack, how new are we talking?!?!
Cheers,

-Stephen Mercier | Sr. Systems Architect
Attainia Capital Planning Solutions (ACPS)
O: (650)241-0567, 727 | TF: (866)288-2464, 727
stephen.mercier@xxxxxxxxxxxx | www.attainia.com

On May 12, 2016, at 2:46 PM, Gregory Farnum wrote:

On Thu, May 12, 2016 at 2:36 PM, Stephen Mercier <stephen.mercier@xxxxxxxxxxxx> wrote:
I'm trying to setup a crush rule, and I was hoping you guys could clarify something for me.

I have 4 storage nodes across 2 cabinets. (2x2)

I have the crush hierarchy setup to reflect this layout (as follows):

rack cabinet2 {
	id -3		# do not change unnecessarily
	# weight xxxx
	alg straw
	hash 0	# rjenkins1
	item cephstore04 weight xxxx
	item cephstore02 weight xxxx
}
rack cabinet1 {
	id -3		# do not change unnecessarily
	# weight xxxx
	alg straw
	hash 0	# rjenkins1
	item cephstore03 weight xxxx
	item cephstore01 weight xxxx
}
root default {
	id -1		# do not change unnecessarily
	# weight xxxx
	alg straw
	hash 0	# rjenkins1
	item cabinet2 weight xxxx
	item cabinet1 weight xxxx
}

The default ruleset is as follows: (Big surprise!!)

rule replicated_ruleset {
	ruleset 0
	type replicated
	min_size 1
	max_size 10
	step take default
	step choose firstn 0 type osd
	step emit
}

If I want this to ensure that there is at least 1 copy of the data in each cabinet, would I just change it to:

rule replicated_ruleset {
	ruleset 0
	type replicated
	min_size 1
	max_size 10
	step take default
	step choose firstn 0 type rack
	step emit
}

Or should it be:

rule replicated_ruleset {
	ruleset 0
	type replicated
	min_size 1
	max_size 10
	step take default
	step chooseleaf firstn 0 type rack
	step emit
}

If you only wan two copies, the chooseleaf variant is correct.

Assuming you want three copies, neither of these is quite right. The use of "firstn 0" means "take the requested number of replicas in this selection step", so both of these would be asking for 3 racks; which obviously won't work.
(The "choose" variant doesn't work either, because you're telling it to select N racks and then emit those as the object locations! You'd need to add in a chooseleaf or set of choose calls underneath it.)

Or is there something more complicated I should be doing? I took a look at https://www.mail-archive.com/ceph-users@xxxxxxxxxxxxxx/msg19140.html and it sounds like this is what I want, but I've also see examples like the following:

rule replicated_ruleset {
	ruleset 0
	type replicated
	min_size 1
	max_size 10
	step take default
        step choose firstn 2 type rack
	step chooseleaf firstn 0 type osd
	step emit
}

So this rule is saying "select 2 racks, and within each selected rack, choose N leaf nodes". That's also not quite what you'd want.

If all of your components are new enough, you can do the overselection hack:

rule replicated_ruleset {
	ruleset 0
	type replicated
	min_size 1
	max_size 10
	step take default
        step choose firstn 2 type rack
	step chooseleaf firstn 2 type osd
	step emit
}

That selects two racks (ie, both) and then chooses 2 OSDs within each rack. But if you're only asking for three copies, it'll truncate the last OSD off the list; and because it's selecting racks in a different order each time you'll get a good distribution across racks.
-Greg

As you might have noticed, I'm a little confused, so any assistance is greatly appreciated. And just to clarify once more, I want to make sure that it stores at least one copy in each rack. Advice on getting more granular is welcome as well however, as I there are pools with both 2x and 3x replication setup.

Cheers,
-Stephen Mercier | Sr. Systems Architect
Attainia Capital Planning Solutions (ACPS)
O: (650)241-0567, 727 | TF: (866)288-2464, 727
stephen.mercier@xxxxxxxxxxxx | www.attainia.com

_______________________________________________

ceph-users mailing list

ceph-users@xxxxxxxxxxxxxx

http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com