Version stability

[zinebammar]

Hello,

We are currently working on updating DSS from 12.3 to 12.5, could you please confirm what is the latest stable version of DSS ? is it right 12.5 or a latest one. Thank you.

Best regards,

[Turribeach]

I don’t work for Dataiku so this is just my personal opinion but you should probably stop thinking in terms of the “latest stable version”. Dataiku like many other companies has moved to an agile delivery model where they are now pretty much doing monthly releases, sometimes even faster than that. This means that there is no concept of “latest stable version” or put it in a different way the “latest stable version” is always the last patch release. 

As Dataiku releases new versions more frequently these releases have both new features and big fixes. As they are more frequent the level of change is smaller reducing the risk of new bugs or regression bugs. You should therefore aim to always update Dataiku on a monthly basis if possible to match Dataiku’s cadence. If you don’t want to be on the just released version you could leave a month behind the latest release which usually will keep one version below the current version. But it’s been long proven than updating software more frequently is a way better way to reduce breakage risk as you do less bigger updates and pay your technical debt as you go rather than in big ballon payments. On top of that delivering new features to your business users will obviously be much appreciated by them. 

[zinebammar]

Hello, thank you for your reply.

We currently want to upgrade the version from 12.3 to 12.5 so it's a major upgrade for our DSS. Your approach is very good actually and we might think of it seriously. do you know please if migrating from 12.3 DSS to 12.5 will have an impact if we want to upgrade the cluster kubernetes installed from 1.26 to 1.27 ?  thank you 

[Turribeach]

12.3 to 12.5 is not a major upgrade at all. If you look at the Dataiku Release Notes you will see that moving from 12.3 to 12.5 is just a 3 months jump. In fact there were 6 Dataiku releases between 12.3 and 12.5 which means that's 2 a month ratio. You need to stop thinking in release numbers as these aren't anymore a good representation of how much a software product has changed. This is in line with other companies that have moved to agile development. I can't tell you if your environment will be impacted moving to a new version since I don't know your environment so I couldn't possibly tell. You should do your own testing to determine that. I also suggest reading all the release notes for every release you jump so you are not surprised by any changes. Finally in terms of Kubernetes in EKS Dataiku recently added v1.29 as a tested version so I would just upgrade to 1.29 rather than 1.27 as 1.27 standard support ends in July 2024 in AWS.

https://doc.dataiku.com/dss/latest/containers/eks/managed.html

[tgb417]

@zinebammar ,

Please note we are not using a kubernetes cluster here.  Others will have to share their experiences on that part of the upgrade.  

[tgb417 · Accepted solution]

@zinebammar ,

Welcome to the dataiku community.

It appears that the latest version right now is 12.5.2.

In general I agree with @Turribeach staying up to date with releases is a good idea.  

The one caveat, I would add to this discussion, is an evaluation of your actual costs to do a version upgrade.  If an upgrade is relatively inexpensive for you, saying just costing you the time to plan, announce the upgrade, and the down time to do the upgrade. Followed by a bit of extra support and monitoring after the upgrade, Then from my point of view quick regular upgrades make a lot of sense.  In our current operations with DSS where we are generally using the out box configuration we upgrade in step with Dataiku. We are currently using 12.5.2.

However, if you are using a highly customized configuration, where you suspect that the standard regression testing that Dataiku is doing, is not going to cover your business critical use cases. In these cases you likely need to do significant additional testing on your local use. Or if you are in some sort of highly regulated environment where there are significant requirements to show due diligence. In these cases the real costs of the upgrade are going to be a lot higher. And when these actual costs to upgrade are much higher, it likely suggests a much slower version adoption cadence, as a cost control mechanism.    

Just my $0.02 for this discussion.