Hi Yehuda,

Thanks for you help.

No, listing does not work, if I remove the old index objects.

I guessed, I could use the resharding for my purpose. I just tried
* copy the index object
* rewrite bucket metadata
* reshard
=> I get new index objects at the old place. Metadata gets turned back 

Maybe this is more complicated as expected?

Best regards,


-----Urspr√ľngliche Nachricht-----
Von: Yehuda Sadeh-Weinraub [mailto:yeh...@redhat.com]
Gesendet: Donnerstag, 15. Februar 2018 00:21
An: Ingo Reimann
Cc: ceph-users
Betreff: Re: [ceph-users] rgw: Moving index objects to the right index_pool

On Tue, Feb 13, 2018 at 11:27 PM, Ingo Reimann <ireim...@dunkel.de> wrote:
> Hi List,
> we want to brush up our cluster and correct things, that have been
> changed over time. When we started with bobtail, we put all index
> objects together with data into the pool rgw.buckets:
> root@cephadmin:~# radosgw-admin metadata get bucket:some-bucket {
>     "key": "bucket:some-bucket",
>     "ver": {
>         "tag": "_zgv1FXm604BQtdiZnLkaiXN",
>         "ver": 1
>     },
>     "mtime": "2016-01-08 04:53:38.000000Z",
>     "data": {
>         "bucket": {
>             "name": "some-bucket",
>             "pool": "rgw.buckets",
>             "data_extra_pool": "",
>             "index_pool": "rgw.buckets",
>             "marker": "default.101387371.6",
>             "bucket_id": "default.101387371.6",
>             "tenant": ""
>   [...]
> With Jewel, we introduced a default-placement and put the index of new
> buckets into the pool rgw.buckets.index. Now, we`d like to correct the
> old buckets and shift all the indices where they belong to.
> I can
> * copy the .dir.MARKER.SHARD# objects to the index pool
> * modify the buckets metadata
> But:
> * when I try to modify the metadata of the bucket instance, the
> index_pool does not get changed
> * radosgw-admin bucket stats still shows the old index pool
> Did I miss something?

You didn't miss much. There is a guard in the code that prevents you from 
modifying the placement pools. I have this commit that changes that (but 
that probably doesn't help you much):


The way to go forward for you I think would be by reshrding the buckets, 
which will put the bucket indexes in the correct place. Can you actually 
list the bucket indexes right now?


> NB: Just now, I perform all the operations with a jewel 10.2.2 rgw.
> Luminous is available but not active yet.
> Best regards,
> Ingo Reimann
> Dunkel GmbH
> _______________________________________________
> ceph-users mailing list
> ceph-users@lists.ceph.com
> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
ceph-users mailing list

Reply via email to