Backfill 6 months data: orchestrate without affecting production

Ситуация: New computed metric needs backfill 6 months historic. Cluster shared с production. How orchestrate без breaking SLAs?

ETL Airflow. Backfill = re-process historic dates. Cluster: 200 cores shared. Production SLA-bound.

Доступные данные

• airflow_runs: task_id, ts, status, runtime
• cluster_metrics: ts, cpu_used, memory_used
• sla_tracking: workflow, target_time, hit

Задачи

• Backfill strategy: parallel vs serial.
• Resource budget: production reservation.
• Monitoring: SLA impact.
• Rollback plan.

Все кейсы для подготовки → ← Все кейсы