Background and Speakers

Chaos Conf 2020 brought together practitioners across the chaos engineering and site reliability engineering (SRE) communities for three days of talks and discussions. InfoQ previewed the event with two of its featured speakers:

• Adrian Cockcroft – VP of Cloud Architecture Strategy at AWS and a long-standing voice in the chaos engineering and microservices communities.
• Yury Niño Roa – Senior Site Reliability Engineer at ADL Digital Lab, with hands-on experience applying chaos principles in enterprise environments.

The conversation touched on the state of chaos engineering adoption, the value of game days, observability, and the DevOps feedback loop — themes that remain highly relevant to engineering organisations today.

Uneven Adoption Across Organisations

One of the clearest takeaways from the discussion is that chaos engineering adoption remains inconsistent. While cloud-native organisations and large technology companies have integrated chaos practices into their standard operating procedures, many traditional enterprises are still at an exploratory stage — running occasional experiments rather than embedding failure testing into their continuous delivery pipelines.

This gap is not simply a tooling problem. Culture, organisational structure, and leadership buy-in all influence how quickly teams move from reading about chaos engineering to actually injecting failures into production or staging environments.

Game Days and Psychological Safety

Both Cockcroft and Niño Roa highlighted the enduring value of game days — structured exercises in which teams deliberately simulate incidents or failure scenarios in a controlled environment. Although game days are not a new concept, the interviewees emphasised how they contribute to psychological safety: the shared belief among team members that it is safe to take interpersonal risks, speak up, and make mistakes without fear of blame.

Google’s research into high-performing teams — notably through Project Aristotle — identified psychological safety as the single most important factor distinguishing effective teams from less effective ones. This principle applies with particular force to on-call and incident response scenarios, where engineers must act under pressure and time constraints. Running realistic incident drills through game days helps teams build the muscle memory and interpersonal trust they need before a real outage occurs.

Key Benefits of Game Days

• Practice incident detection, triage, and remediation in a low-stakes environment.
• Identify gaps in runbooks, alerting, and escalation paths before they surface in production.
• Build cross-team familiarity so that developers and operations engineers know each other’s tools and assumptions.
• Reinforce a blameless culture by separating the learning exercise from performance review contexts.

Observability: Closing the DevOps Feedback Loop

A recurring theme at Chaos Conf 2020 was the idea of completing the DevOps loop. The DevOps movement set out to break down silos between development and operations, but in practice the flow of information is often one-directional: developers ship code, operations run it, and feedback rarely travels back upstream in a structured way.

The Chaos Conf organisers framed this clearly: too often DevOps creates a one-way flow from developers to ops, with no corresponding feedback from ops to help developers build more operable and reliable systems. Chaos engineering, when combined with strong observability, offers a mechanism to close that loop.

Implementing observability across both platform and application layers — through structured logs, distributed traces, and meaningful metrics — gives operations teams the data they need to surface reliability concerns directly to the engineers who can fix them. Chaos experiments generate controlled failure signals; observability tools make those signals visible and actionable.

Where Chaos Engineering Is Headed

Looking forward, both speakers pointed to two directions for the discipline:

• Security chaos engineering: Incorporating security-focused failure experiments — such as testing how systems respond to credential compromise, network policy violations, or dependency tampering — alongside traditional infrastructure and application fault injection.
• Larger-scale experiments: Moving beyond individual service or node failures toward experiments that test broader failure modes, such as full availability zone outages, cascading dependency failures, or data-plane and control-plane separations at scale.

Automation is central to both directions. As chaos experiments grow in scope and frequency, manual execution becomes impractical. Integrating experiment scheduling and hypothesis validation into CI/CD pipelines allows teams to catch reliability regressions at the same point they catch functional regressions.

Practical Takeaways for DevOps Teams

• If your team has not run a game day recently, schedule one. Start small: simulate a single dependency failure and walk through your detection and response process end to end.
• Audit your observability stack before expanding chaos experiments. You cannot learn from an experiment you cannot observe.
• Work with developers and operations engineers together when designing experiments. Joint ownership of experiment results helps break down the one-way information flow that undermines DevOps effectiveness.
• Treat chaos engineering as a continuous practice, not a one-time audit. Reliability is not a property you verify once; it degrades as systems change.

The full Q&A between InfoQ, Adrian Cockcroft, and Yury Niño Roa was published on the InfoQ website. Chaos Conf 2020 was organised by Gremlin and ran as a free virtual event in October 2020.