PGBackend: Multiple PG implementations.

PGBackend: Multiple PG implementations.

[Lluís Pàmies i Juárez]

Hello,

I was looking at the work done to re-architecture the OSD backend.
Since it seems that it allows now to have more than one PG
implementation I decided to subclass PG and PGBackend to create an
alternative implementation to ReplicatedPG and ReplicatedBackend.
However, I noticed a couple of problems:

* All methods in watch.h and watch.cc expect ReplicatedPG objects
instead of generic PG objects.
* New pools created in the OSDMonitor are created as replicated pools
(g_pool_t::TYPE_REP). This is a hardcoded value.

I made a few changes to solve these two issues that allowed me to have
more than one PG implementation running without problems. You can find
the changes here:

https://github.com/ceph/ceph/pull/744

Are these changes useful for the re-architecture of the OSD ?

PS: Implementing and running an alternative PG requires other changes
in the CrushCompiler and other header files.

Best,
Lluis

Re: PGBackend: Multiple PG implementations.

[Loic Dachary]

[-- Attachment #1: Type: text/plain, Size: 2052 bytes --]

Hi Lluís,

David Zafman and Samuel Just are working to re-architecture the core of Ceph so that and ErasureCodedBackend can be implemented. You can browse their commits to see what they are up to. The transcript of the last CDS summit will also give you a general view of the direction in which it is heading : 

http://wiki.ceph.com/01Planning/CDS/Emperor/Transcript_:_Erasure_coded_storage_backend_%28step_2%29

and a more recent interview of Samuel Just will give you a short update :

http://dachary.org/?p=2320

I'm not the best qualified to comment on the generic part of your patches, but I think it overlaps with the work being done.

Cheers

On 19/10/2013 01:05, Lluís Pàmies i Juárez wrote:
> Hello,
> 
> I was looking at the work done to re-architecture the OSD backend.
> Since it seems that it allows now to have more than one PG
> implementation I decided to subclass PG and PGBackend to create an
> alternative implementation to ReplicatedPG and ReplicatedBackend.
> However, I noticed a couple of problems:
> 
> * All methods in watch.h and watch.cc expect ReplicatedPG objects
> instead of generic PG objects.
> * New pools created in the OSDMonitor are created as replicated pools
> (g_pool_t::TYPE_REP). This is a hardcoded value.
> 
> I made a few changes to solve these two issues that allowed me to have
> more than one PG implementation running without problems. You can find
> the changes here:
> 
> https://github.com/ceph/ceph/pull/744
> 
> Are these changes useful for the re-architecture of the OSD ?
> 
> PS: Implementing and running an alternative PG requires other changes
> in the CrushCompiler and other header files.
> 
> Best,
> Lluis
> --
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

-- 
Loïc Dachary, Artisan Logiciel Libre
All that is necessary for the triumph of evil is that good people do nothing.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 263 bytes --]