Re: [Qemu-devel] [PATCH] rbd: hook up cache options

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

On 05/22/2012 02:18 AM, Paolo Bonzini wrote:
> Il 17/05/2012 22:42, Josh Durgin ha scritto:
>> +     * Fallback to more conservative semantics if setting cache
>> +     * options fails. Ignore errors from setting rbd_cache because the
>> +     * only possible error is that the option does not exist, and
>> +     * librbd defaults to no caching. If write through caching cannot
>> +     * be set up, fall back to no caching.
>> +     */
>> +    if (flags&  BDRV_O_NOCACHE) {
>> +        rados_conf_set(s->cluster, "rbd_cache", "false");
>> +    } else {
>> +        rados_conf_set(s->cluster, "rbd_cache", "true");
>> +        if (!(flags&  BDRV_O_CACHE_WB)) {
>> +            r = rados_conf_set(s->cluster, "rbd_cache_max_dirty", "0");
>> +            if (r<  0) {
>> +                rados_conf_set(s->cluster, "rbd_cache", "false");
>> +            }
>> +        }
>> +    }
>
> Last time I looked at ceph, rbd_flush was not a full flush of the cache;
> it only ensured that the pending requests were sent.  So my questions are:

I'm not sure which version you were looking at, but this hasn't been
the case since caching was implemented. I don't think it was ever the
case, actually. rbd_flush has always waited for pending I/Os to complete
(be on disk on all replicas), not just be in flight.

If you're interested in the current implementation, you can see:

src/librbd.cc: librbd::flush()

which goes into:

src/osdc/ObjectCacher.cc: ObjectCacher::commit_set()

or

src/librados/IoCtxImpl.cc: IoCtxImpl::flush_aio_writes()

> 1) has this changed?  does rbd_flush now flush dirty items when
> rbd_cache_max_dirty>  0?

The rbd_cache_* options did not exist before 0.46.

> 2) should the usage of a cache be conditional on LIBRBD_VERSION_CODE>=
> LIBRBD_VERSION(0, 1, 1)?

It doesn't matter if you use an older version because the non-existent
options don't have any effect.

Thanks,
Josh