What can cause an object to be not_found even though R+W>N?
will at trueex.com
Mon May 7 16:58:25 EDT 2012
Hi, I am observing the following:
(Setup, 3 physical nodes, default settings of R 2, W 2, N 3, LevelDB
backend, using riak:local_client() from inside the same BEAM.)
1. Three fresh nodes are started up with empty data and ring state
directories, the second two are told to join the first, and then a bunch of
seed data is loaded, as part of a deployment script.
2. The nodes are all stopped at the end of the deployment.
3. When the system is started for real, the nodes are brought back up, and
are re-connecting from the ring state they had persisted during deployment.
4. One of our own OTP applications running inside the cluster detects that
riak_kv is started and ring ready, it begins to try to load and process
some of the data.
5. It first lists all of the keys from a bucket, then retrieve them one by
PROBLEM: Some of the objects that were in the list of keys return not_found
A) I see from some searching on the mailing list archives that as of
mid-2011 objects may be unavailable during handoffs. Is this still the case
with riak_kv 1.1?
B) What is the canonical way to make sure that all handing off of the
current ring is done after joining and that retrieval operations will be
C) In the case of physical node failure and failover, is there ever a
chance of erroneous not_founds with the default R/W/N settings?
THE INFORMATION CONTAINED IN THIS COMMUNICATION IS INTENDED FOR THE NAMED
RECEIVER ONLY. THE TRANSMISSION MAY CONTAIN PRIVILEGED AND CONFIDENTIAL
MATERIAL. IF YOU ARE NOT THE NAMED RECIPIENT, PLEASE BE ADVISED THAT ANY
USE, DISSEMINATION OR UNAUTHORIZED COPYING OF THE MATERIAL IS STRICTLY
PROHIBITED. IF YOU HAVE RECEIVED THIS TRANSMISSION IN ERROR, PLEASE NOTIFY
SUPPORT at TRUEEX.COM AND DESTROY THE RECEIVED COPY. THANK YOU.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the riak-users