Compatibility sure.

The updates to the different versions are mainly security updates.
For the hadoop 3.3.3 to 3.3.4 is security related.
The spark 3.2.0 to 3.3.1 is both security and a minor update.

I think both are fair updates to make, that does not require us to do a major 
release.
We also risk that someone points out the security vulnerabilities in an 
upcoming release if we do not update.

best regards
Sebastian



On 7 December 2022 at 13:43:50 +01:00, Matthias Boehm <[email protected]> wrote:

> yes, backwards compatibility is important to keep in mind. In the past, we 
> mentioned the *minimum* Spark/Hadoop versions SystemML/SystemDS required, 
> while still being able to run with more recent versions.
> 
> So let's separate two different aspects here: (1) the minimum version 
> compatible with our source code, and (2) the dependencies we reference for 
> tests and build. Although Spark and Hadoop are generally very good in terms 
> of backwards compatibility inside major versions, it would be good to keep 
> (1) and (2) in sync to avoid hidden incompatibilities.
> 
> Are there specific reasons for this update - API changes, security warnings, 
> etc?
> 
> Regards,
> Matthias
> 
> On 12/7/2022 1:06 PM, arnab phani wrote:
> 
> > Can these upgrades harm the backward compatibility of the next SystemDS
> > release?
> > If so, then we either need to make a major release or delay the upgrades
> > till the next major release.
> > 
> > Regards,
> > Arnab..
> > 
> > On Wed, Dec 7, 2022 at 1:01 PM <<[email protected]>> wrote:
> > 
> > 
> > > Hi all,
> > > 
> > > Just to let all know,
> > > 
> > > I am considering updating the Spark, and Hadoop version of our system,
> > > would this interfere with any ongoing work currently?
> > > 
> > > Spark 3.2.0 -> 3.3.1
> > > Hadoop 3.3.3 -> 3.3.4
> > > 
> > > best regards
> > > Sebastian
> > >

Reply via email to