Hello Jonathan,
Thanks for your quick reply. Since consistency is tunable by making
tradeoffs with latency and availability, it lead me to think about how to
improve the consistency level (better than eventual consistency) while
guarantee that latency and availability won't be affected much. So now I am
working on Cassandra and trying to make some contribution on this.
Although there are some approaches that already been done, I
would just like to figure out another approach.
I understand it theoretically, yet couldn't find a way to reproduce it by a
large set of transaction, for example, 10,000 operations..
Do you have any comments? Thank you so much!
Best,
Xindi
2015-09-08 18:08 GMT-04:00 Jonathan Ellis :
> 2 node cluster. RF=2. Take one node down. Do a write. Take the other
> node down. Do a read. It will be stale since the node that is up missed
> the original write, and you took the node that did see the write down
> before it could replay to it.
>
> This is by design though so I'm curious what kind of improvement you are
> looking for.
>
> On Tue, Sep 8, 2015 at 3:57 PM, Cindy Wang
> wrote:
>
> > Hi there,
> >
> > I currently research on the improvement of consistency level of
> Cassandra.
> > Since I am new to C*, does anyone know how to reproduce the stale reads
> on
> > the cluster? Is there a transaction lib that can be loaded into it and
> > reproduce the stale reads?
> >
> > ***TEST environment***
> >
> > AWS 3 instances(nodes) + Ubuntu 14.04 + Cassandra 2.2.1
> >
> > I plan to reproduce the stale reads first and then take a look at the
> > source code trying to find out a way for improvement.
> >
> > Thanks so much. Any of your help is highly appreciated!!!
> >
> > Best,
> > Xindi
> >
>
>
>
> --
> Jonathan Ellis
> Project Chair, Apache Cassandra
> co-founder, http://www.datastax.com
> @spyced
>
