Follow-up on this announcement.  After further work today, it looks like
the initial diagnosis of the Issue 1707 was incorrect, and that the defect
does not obviously present a significant risk of data loss.  Work will
continue on providing an updated patch to Riak 2.9.0 - but there should be
no immediate concern over use of the unpatched version.

Sorry for causing false alarm.

Regards

Martin

On Thu, 22 Aug 2019 at 01:36, Martin Sumner <[email protected]>
wrote:

> There are 2 defects in the 2.9.0 release which should be patched within
> the next 7 days, with a new version of 2.9.0 to be published before the end
> of the month.  The defects are:
>
> https://github.com/basho/riak_kv/issues/1706
> https://github.com/basho/riak_kv/issues/1707
>
> Note that the latter defect 1707, the scope of this bug is currently
> unclear, but there exists significant risk that there is potential within
> this bug for data loss, due to incorrect pruning of siblings.
>
> Previous releases of Riak also have this bug, if the metadata cache (which
> is disabled by default) is enabled.  Enabling this cache is controlled via
> https://docs.riak.com/riak/kv/latest/configuring/reference/index.html 
> (metadata_cache_size).
> However, unlike previous releases the 2.9.0 release, when combined with use
> of the leveled backend, Riak 2.9.0 will enable the path to this bug by
> default.  This is because the logic of the metadata_cache is reused in
> order to enable leveled HEAD requests within the PUT path.  Riak 2.9.0 with
> other backends, and the metadata cache left disabled, will not have this
> bug.
>
> I apologise for the ongoing churn of issues uncovered in release 2.9.0,
> and for the potential risks associated with this latest bug in particular.
> Clearly, the high rate of discovered problems has exposed that the test
> process which surrounded this release was not good enough.  For release
> 2.9.1 and release 3.0 the move has already been made to involve independent
> testing organisations as a fundamental part of the development process,
> from the start of the process.  Regardless of this, once the immediate
> issues are patched, there is a need for some further reflection on what is
> required to assure the safety of a release of Riak in the future.
>
> Regards
>
> Martin
>
_______________________________________________
riak-users mailing list
[email protected]
http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com

Reply via email to