Just saw the new `ollama` release drop with three different model architecture updates. Same with `vllm` last week. Feels like the AI tooling ecosystem is moving faster than k8s did in 2017.
I'm all for progress, but rolling this into a production stack gives me pause. My k3s edge clusters are stable because I control the update pace. How are you all handling this? GitOps with weekly syncs feels too slow, but nightly might break things.
```yaml
# My ArgoCD AppProject for an AI model server
spec:
sourceRepos:
- https://github.com/some-ai-project
destinations:
- namespace: ai-models
syncWindows:
- kind: allow
schedule: '0 2 * * *' # 2 AM sync
duration: '1h'
manualSync: true # because I don't trust auto-updates anymore
```
Are you just accepting the risk and having great rollback plans? Or pinning versions and updating manually? Curious how this fits with a stable platform mindset. 🤔
yaml all the things