On Fri, Dec 29, 2017 at 01:01:25PM +0800, Jeegn Chen wrote:
> Hi Robin
>
> > Rough implementation:
> > - For writes, the RGW zone data describes which pool maps to each
> > storage class.
> > (corner case: multipart uploads might need each part in a consistent pool)
> [Jeegn]: Do you mean mean different placement polices are
> corresponding to STORAGE CLASSes?
Yes, placement_pools are already most of the way to storage class behavior, but
they seemingly were modeled after Swift rather than S3.
> > - The bucket-index data already describes the RADOS POOL to _read_ from
> > (corner case: old buckets/objects don't have this set)
> [Jeegn]: Are you referring to the tail_bucket in the RGWObjManifest,
> which is somewhat a copy of placement rule? Currently, it seems to
> support the copy across buckets with the same data pool and I have not
> found the logic to deal with different pools. But yes, maybe we can
> reuse it instead of add additional xattr to track the tail pools.
I haven't confirmed exactly where in the codebase it is, but visible in
the 'radosgw-admin bi get/list' output is the pool ID.
In the following snippet, look at entry.ver.pool. That's the ID for my
.rgw.data.1 pool.
$ sudo radosgw-admin --bucket CENSORED bi list --object .bash_profile
[
{
"type": "plain",
"idx": ".bash_profile",
"entry": {
"name": ".bash_profile",
"instance": "",
"ver": {
"pool": 5,
"epoch": 100414
},
"locator": ".bash_profile",
"exists": "true",
"meta": {
"category": 1,
"size": 81,
"mtime": "2015-07-02 15:31:59.000000Z",
"etag": "8040ef233daa49c60b93cb3a45617e19",
"owner": "CENSORED",
"owner_display_name": "CENSORED",
"content_type": "application\/octet-stream",
"accounted_size": 81,
"user_data": ""
},
"tag": "default.136713729.262057859",
"flags": 0,
"pending_map": [],
"versioned_epoch": 0
}
}
]
> > - radosgw-admin already contains bucket/object rewrite functionality,
> > that would effectively copy from an old pool into the new pool.
> > (note: I don't think is well-documented at all)
> [Jeegn]: Are you talking about "radosgw-admin bucket rewrite"? But per
> check through RGWRados::rewrite_obj() and check_min_obj_stripe_size()
> in the master branch, the functionality is used to migrate from
> explicit-obj implementation to manifest implementation. Or you mean
> some command else or some implementation wip?
I think this functionality is close enough for slight modifications to
get what we want, of a manifest-to-manifest rewrite.
--
Robin Hugh Johnson
Gentoo Linux: Dev, Infra Lead, Foundation Treasurer
E-Mail : robbat2@xxxxxxxxxx
GnuPG FP : 11ACBA4F 4778E3F6 E4EDF38E B27B944E 34884E85
GnuPG FP : 7D0B3CEB E9B85B1F 825BCECF EE05E6F6 A48F6136
Attachment:
signature.asc
Description: Digital signature
