Re: explicitly specifying pgnum on pool creation

[Josh Durgin <josh.durgin@inktank.com>]

On 11/04/2012 03:42 AM, Sage Weil wrote:
> The wip-explicit-pgnum changes the 'ceph osd pool create <name> <pgnum>'
> command to require the pg_num value instead of defaulting to 8.  This
> would make it harder for users to get this wrong.

I like this. It'd be great if we could add a pgnum argument to
'rados mkpool' and the corresponding librados pool_create* functions
too.

> On the other hand, it probably also breaks some scripts for deploying
> OpenStack that create volume and image pools.  :/

If those scripts are creating pools with 8 pgs, they're going to
create problems anyway. I don't mind breaking them as long as it's well
documented (i.e. release notes and docs on 'ceph osd pool create'
should mention the change).

> Ideas?
>
> The original idea was that the monitor would automagically notice when a
> small pgnum pool gets lots of objects and trigger a split.  Even if we
> don't do that, pretty soon now you'll be able to explicitly increase
> pg_num.  I still think it might be a good idea to require it up-front,
> though.

Until split is implemented and well-tested, I think it makes sense to
require it up-front. It's easy to relax that requirement later if we
want to make it more automatic.

> An alternative would be to default to a larger number (say, num_osds <<
> 2).  My concern there is that it makes it easy to create lots of pgs for
> pools that may be small, and it's still not large enough to get good
> performance for the "create pool, run rados bench" crowd.

It would also be easy to create too few pgs, and then add a bunch of
osds later.

Josh