AWS MediaConvert を使用したサーバーレス動画処理パイプライン

MicrocosmWorks designed a segmented processing architecture where Step Functions orchestrate the pipeline: Lambda functions split source videos into segments, AWS MediaConvert handles the actual transcoding without Lambda timeout constraints, and a final Lambda stitches the output. This hybrid approach keeps the serverless cost model while supporting videos of any duration.

MicrocosmWorks measured a 70-85% cost reduction for bursty video processing workloads compared to running dedicated EC2 encoding instances. The serverless pipeline incurs zero cost when idle and scales to hundreds of concurrent jobs during peak periods, with AWS MediaConvert's per-minute pricing eliminating the need to provision for peak capacity.

MicrocosmWorks configured AWS Step Functions with per-step retry policies and exponential backoff, ensuring that a failed transcode step retries automatically without restarting the entire pipeline. Each stage writes intermediate outputs to S3, so recovery resumes from the last successful checkpoint rather than reprocessing from the source file.

MicrocosmWorks optimized the pipeline for near-real-time use cases with cold start mitigation using provisioned concurrency on critical Lambda functions and MediaConvert reserved transcoding slots. For live workflows, the pipeline achieves 2-5 minute end-to-end latency from upload to delivery, which is suitable for clip extraction and highlights distribution.

MicrocosmWorks builds serverless video infrastructure at rates of $25-$45/hr, with a complete MediaConvert-based pipeline including Step Functions orchestration, S3 lifecycle management, and monitoring typically requiring 250-400 development hours. The architecture's pay-per-use model means clients only pay AWS costs proportional to their actual processing volume.