[Hawkular-dev] Cassandra upgrade

[John Sanda]

> On Jan 13, 2016, at 10:20 AM, Heiko W.Rupp <hrupp at redhat.com> wrote:
> 
> On 11 Jan 2016, at 18:45, John Sanda wrote:
> 
>> 3.x. From a development standpoint, I think it absolutely makes sense 
>> to upgrade. I would like to know what others think (Heiko in 
>> particular). There are other
> 
> Can you list some of those? Do we expect improved
> performance or reduced resource consumption?

Some of the major changes include

* materialized views 
    * Secondary indexes are typically discouraged in favor of manual, custom indexes. Materialized views can replace these.
* aggregate functions
* improvements to incremental repair
* storage engine rewrite
    * original engine written for Thrift and before CQL was implemented. New engined based around CQL.
    * reduced memory usage on reads
    * reduced storage on disk
        * clustering columns no longer stored per cell
        * column names are no longer repeated
        * fixed-width cells are written without size 

> 
> I believe though that upgrading now is better than
> later in the game. And if only with respect to not
> needing any sstableupgrade runs.

Running sstableupgrade or similar maintenance is something that we won’t be able to avoid whether it is not or at some later point in the future.

> 
> It has the drawback of productization though as you write:
> 
>> considerations like productization build changes and OpenShift 
>> integration. There would be a data migration involved with OpenShift 
>> since we already shipped Cassandra 2.x. When upgrading across major 
>> versions, you are supposed to run the sstableupgrade utility on each 
>> table. We would need to do this for OpenShift. Whether or not we 
>> choose to upgrade now, we will need to deal with this migration step 
>> in OpenShift at some point.
> 
> Yes. What runtimes are expected here? If they are typically in
> some minute ranges, then they could be more or less be done
> during a normal openshift version upgrade I guess.

We will need to do some testing, but I believe that this can be done as a rolling upgrade.

> _______________________________________________
> hawkular-dev mailing list
> hawkular-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/hawkular-dev