I'm about to use strimzi kraft mode for production. Is it stable?

[SteveArseneLee]

I'm a novice data engineer trying to use apache kafka on kubernetes for commercial use. I'm trying to use Kraft mode itself because I heard it's stable, but I'm wondering if it's also stable when using it through strimzi. I'm not sure because I've seen an article about limitation in other discussions. I need advice from a lot of people. Thank you.

[scholzj]

KRaft still has some limitations both in Apache Kafka itself or in Strimzi - for example JBOD storage support, scaling of controller nodes, etc. You should probably read the docs and decide for yourself if it is ready for you or not.

[scholzj]

Well, as I said, these limitations are in Kafka in the first place. So if they matter to you, then a KRaft cluster with Bitnami does not help you in any way.

[SteveArseneLee]

I apologize for asking the obvious over and over again, and I really appreciate your response.

[scholzj]

For the JBOD part, I think the decision might be very easy ... if you for example plan to run in production as a small cluster with 100Gi disks per broker and small throughput, JBOD is not likely to add much value. If you plan to run with 10Ti of disks and very big througput, JBOD might be super important. We have some videos on YouTube covering the JBOD (dis)advantages.

The scaling of the controllers is probably trickier to decide on as it relates to infrastructure you use etc.

[ppatierno]

+1 on all Jakub said. Think that Strimzi deploys your cluster which is anyway using Kafka (from the upstream community). So if limits are in Kafka, whatever the solution you use to deploy the cluster, they will stay there. Unless any other form of commercial version of Kafka that could have some stuff already fixed.

[SteveArseneLee]

My current k8s cluster is 16 cores & 16GB or less for both master and worker, and I have allocated 500GB or less, so I don't think I'm going to allocate that many resources for kafka as well. So I think I can pass on the JBOD issue as Jakub advised, but I'm still not sure about the controller scaling. I'm not sure how I'm going to orient my production environment with kubernetes at work, as I've just installed and used it monolithically on premise for learning purposes. But I want to make it as stable as possible, so I'm looking into different things.