Automatic rollbacks are a last resort

Automatic rollbacks in software deployment, often seen as a tool for enhancing reliability, are critiqued by Steve Fenton for their limitations and the missed opportunities they present for learning and system improvement. Many deployment failures, such as expired credentials, would also thwart rollback attempts, and automatic rollbacks can obscure the understanding of failure conditions, preventing valuable insights. Fenton highlights the power of human-driven resilience, arguing that modern approaches like Continuous Delivery and progressive delivery strategies, including blue/green deployments, canary deployments, and feature flags, provide more robust and adaptive solutions. These practices emphasize frequent, small changes with continuous validation, fostering a resilient and adaptable deployment pipeline. Additionally, rollbacks might inadvertently introduce untested states or compatibility issues, making them a last resort that should involve human oversight. By leveraging these modern techniques, organizations can improve their deployment processes, ultimately reducing the perceived need for automatic rollbacks and building more robust systems.