Insights into Managed Service Provision: A STACKIT Retrospective
summary
STACKIT builds a Kubernetes-based management lane to provision PostgreSQL InfluxDB MongoDB on Kubernetes using Helm, operators and a REST API front end; metadata lives in Postgres, events use a message queue and the outbox pattern to avoid dual write failures. Explored GitOps (Argo/Flux) but hit git rate limits and push contention; moved toward an operator + agent on each cluster that watches CRDs. A past incident crashes the kube API (etcd pressure, ~40 GB RAM) so the team measures scaling and chooses the approach per tradeoffs.
AI-written summary · about
chapters (12)
- 0:00Introduction and background
- 2:11What is a managed service
- 4:01Building a management lane
- 11:49Handling instance variations
- 13:32Consistent event processing
- 15:20GitOps and scaling issues
- 17:08Agent-based cluster control
- 18:36Propagating cluster state
- 21:26Extending the pattern
- 22:09Performance and scaling lessons
- 23:14Q&A: Kubernetes API limits
- 23:55Q&A: Cluster provisioning and drift
related talks
36mCyscale's Journey on Kubernetes
30mBridging the DevOps to Data Divide with a Common Cloud Native Stack
57mSecure Kubernetes data management and scaling with Kasten K10
33mKubernetes As A Service with GitOps
26mKubernetes from the Database Out
32mGitOps, an Elegant Tool for Hybrid Cloud Kubernetes
47mPostmortem: A Modern Approach to Managing Kubernetes Clusters
50mDemystifying Kubernetes for Postgres DBAs; A Guide to Operators