Riak CS: Undeletable broken buckets

Shunichi Shinohara shino at basho.com
Mon Jul 7 05:38:48 EDT 2014


Hi Toby,

There is a rarely used option "disable_local_bucket_check" in Riak CS.
I don't know it solves your case, please let me mention it.

To use it, first set it in app.config of riak-cs (or
application:set_env/3 in shell),
{riak_cs, [...
               {disable_local_bucket_check, true},
               ...]},
Then create bucket as usual (e.g. s3cmd mb ...).

These steps solves one of partial update patterns that Andrew mentioned.

Thanks,
Shino

On Mon, Jul 7, 2014 at 2:12 PM, Toby Corkindale <toby at dryft.net> wrote:
> Hi Andrew,
> Thanks for the details.
> The Puppet config should never have let it be setup with
> allow_mult=false, but as this is a test cluster, it's possible
> something went awry there at some point.
>
> If it's not really a bug that needs reporting then I can let it go.
> Thanks,,
> Toby
>
> On 7 July 2014 12:21, Andrew Stone <astone at basho.com> wrote:
>> Hi Toby,
>>
>> We've seen this scenario before. It occurs because riak-cs stores bucket
>> information in 2 places on disk:
>>   1) Inside the user record (for bucket permissions)
>>   2) Inside a global list of buckets, since each bucket must be unique
>>
>> What has happened most likely is that the bucket is no longer stored for the
>> given user, but still in the global list of bucket. It shows up in bucket
>> lists, but the current user doesn't have permission to actually do anything
>> with it. Essentially you have partially written (or partially deleted) data.
>> I believe the only time we saw this was when Riak was configured with
>> {allow_mult, false} which is an invalid setting when used with riak-cs.
>> Riak-cs uses siblings intelligently to merge conflicting data, and without
>> that it's possible to end up in these types of scenarios. Later versions of
>> riak-cs should refuse to run with {allow_mult, false}. I'd check your riak
>> config to see if that is the case here.
>>
>> We actually have scripts to detect and remove the bad buckets that we've
>> used in support. We can probably get you a copy if you want. Just let me
>> know. And make sure when running in production that allow_mult = true.
>>
>> -Andrew
>>
>>
>>
>> On Sun, Jul 6, 2014 at 9:59 PM, Toby Corkindale <toby at dryft.net> wrote:
>>>
>>> Hi,
>>> At some point we've managed to create a couple of buckets that don't
>>> work and can't be deleted (in a development/testing cluster, not
>>> production).
>>> They show up with both 's3cmd ls' or by querying the HTTP API for a
>>> user's buckets.
>>> However attempting to list files in the bucket, or removing the
>>> bucket, or recreating the bucket, fails.
>>>
>>> It's not in a production cluster so it's not a huge concern to me, but
>>> thought I'd report the bug here in case it's of interest to you.
>>> Riak 1.4.9-1 and Riak-CS 1.4.5-1 on Ubuntu 12.04 LTS.
>>>
>>> $ s3cmd ls
>>> 2014-02-07 00:07  s3://test5403
>>> 2013-12-13 07:25  s3://test9857
>>>
>>> $ s3cmd ls s3://test5403
>>> ERROR: Bucket 'test5403' does not exist
>>> tobyc at adonai:~$ s3cmd ls s3://test9857
>>> ERROR: Bucket 'test9857' does not exist
>>>
>>> $ s3cmd rb s3://test5403
>>> ERROR: Bucket 'test5403' does not exist
>>> Bucket 's3://test5403/' removed
>>>
>>> $ s3cmd ls
>>> 2014-02-07 00:07  s3://test5403
>>> 2013-12-13 07:25  s3://test9857
>>>
>>> $ s3cmd mb s3://test5403
>>> Bucket 's3://test5403/' created
>>>
>>> $ s3cmd ls s3://test5403
>>> ERROR: Bucket 'test5403' does not exist
>>>
>>> _______________________________________________
>>> riak-users mailing list
>>> riak-users at lists.basho.com
>>> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>>
>>
>
>
>
> --
> Turning and turning in the widening gyre
> The falcon cannot hear the falconer
> Things fall apart; the center cannot hold
> Mere anarchy is loosed upon the world
>
> _______________________________________________
> riak-users mailing list
> riak-users at lists.basho.com
> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com




More information about the riak-users mailing list