I’m updating our rollout and rollback playbooks and need training that goes beyond theory. We run KServe + Argo Rollouts with Datadog and I want to cut rollback time from 12 minutes to under 5, with clear runbooks for failed canaries and degraded latency. Any course, workshop, or cert you’ve taken that drills incident response for LLM APIs and teaches concrete troubleshooting steps?
I cut rollback from about 12m to about 4m after Akuity’s Argo Rollouts workshop (https://akuity.io) plus a Datadog incident-response lab; the win was an AnalysisTemplate hitting Datadog p95/error-rate to auto-abort the canary and pre-scaling the stable RS with warmed KServe models so rollback is instant. Caveat: KServe cold starts will tank early canary metrics unless you pre-warm or keep a tiny shadow — ask me how I know. Want my sample AnalysisTemplate?