Dangling keys/objects after a batch of sequential inserts (for going on 3 days)
siraaj at khandkar.net
Thu Jul 25 11:10:02 EDT 2013
On 07/22/2013 07:52 PM, Siraaj Khandkar wrote:
> On 07/22/2013 10:29 AM, Andrew Thompson wrote:
>> You should be doing a GET before the PUT with the deletedvclock option.
>> Here is an example I used to originally test (and fix) this bug:
>> Sadly, because of backwards compatability issues, it was not possible to
>> default deletedvclock to be 'on'. Perhaps enough time has passed to
>> revisit that now. All the client libraries should understand the
>> message now.
> Ah! This makes total sense now - how can there be a resolution without a
> vclock? There cannot.
So, why the objects got deleted makes sense, but it still bothers me
that it took ~3 days for the indices to reach consistency.
There are no other buckets on that cluster and the network load was
pretty minuscule during that time - 1-2 requests per second, if that.
Can you guys shed some light there for me?
More information about the riak-users