MpegFlowBlogBack to home
← Topics·2 pieces

Autoscaling

Video transcoding's workload shape (queue-driven, variable-throughput, periodic-spike) maps cleanly onto KEDA queue-depth autoscaling — but only with the right pool topology. These pieces cover ScaledObject configuration for video pools, scale-to-zero semantics, drain patterns for upgrades without dropped encodes, and where queue-depth scaling stops being sufficient.

Reference architectures · 2
  • Cost-aware spot-instance encoder pool
    Production architecture for running video transcoding on AWS Spot, GCP Preemptible, and Azure Spot instances. Interruption-tolerant queue topology, fleet diversification, atomic upload semantics, and the cost math that makes self-hosted video pipelines beat per-minute pricing at scale.
    May 9, 2026
  • Kubernetes deployment with KEDA autoscaling
    Production K8s topology for MpegFlow — API tier, shared workers via Helm, dedicated workers via Operator, KEDA queue-depth autoscaling, leader election, pool pause for cost savings.
    May 5, 2026
Adjacent topics
  • Kubernetes· 3
  • Spot instances· 1
  • cost optimization· 2
  • KEDA· 1
  • Operator· 2
© 2026 MpegFlow, Inc. · Trust & complianceAll systems nominal·StatusPrivacy