Meetup will be in hybrid mode and the panel discussion will be in English. The physical part takes place in Keboola Prague office. Online stream will be available on Youtube for registered attendees. Records will be published to community's Youtube channel
[EN] Running 10,000 ephemeral stateful jobs in managed Kubernetes daily - Ondřej Hlaváček
Keboola, as a data processing platform, executes thousands of unique customer jobs daily. These range from data extraction and heavy lifting in Python to training machine learning models, with durations varying from minutes to hours. These jobs are stateful and cannot be suspended or relocated; furthermore, some are not even idempotent, as they interact with external services. What are the challenges of running such workloads on a managed Kubernetes service?
Coffee Break
[EN] Understanding storage requirements and options for running your stateful workloads - Ondřej Smola
Managing storage is hard, especially in dynamic environments such as Kubernetes. By understanding storage options and their limits paired with a mature open source ecosystem and built-in knowledge of Kubernetes operators we can make it manageable and even fun at times. With projects such as Rook, Longhorn, or TopoLVM you can provide your storage layer when managed options are not available or you want to keep your data inside your cluster. In the end, we will go deeper into how we are running multiple self-managed stateful workloads such as PostgreSQL, ClickHouse, Etcd, and machine learning model distribution.