Definition
Auto-remediation (automated remediation) is the execution of a corrective action on a production system by software rather than by a person, triggered by a detected fault. It ranges from a fixed script that runs on an alert to an agent that chooses the action from the system’s current state and verifies the result.
Explanation
There are three levels, and they are often confused. Scripted auto-remediation maps one alert to one fixed action, such as restarting a service when a health check fails. Runbook-triggered auto-remediation executes a multi-step procedure written in advance, with branching on known conditions. Agentic auto-remediation reads the live state of the application and the infrastructure, forms a diagnosis, selects an action from an allowed set, executes it and checks that the fault cleared. Each level handles more variety and needs more safety: an allowed-action list, a bounded blast radius, a dry run, a rollback path, an approval step for anything not yet trusted, and an audit record. In security tooling the same phrase means automatically patching or quarantining. In operations it means restoring service.
How DataAgent approaches it
DataAgent publishes a fault catalog of the Kubernetes and AWS faults it is built to detect, diagnose and remediate. Autonomy is granted per fault class, not as one global switch: a class that has not yet earned it through verified fixes goes to a person for approval, and every plan can be dry-run first. The agent’s own permissions bound what any automated step can reach.
Example
A pod enters CrashLoopBackOff two minutes after a config edit. Scripted level: restart the pod, which crashes again. Runbook level: if a config changed in the last 10 minutes, revert it and restart. Agentic level: read the pod’s last log lines, see a parse error on the new key, revert that single key, restart, confirm the container stays running for five minutes, and record the diff.
Related terms
- Incident remediation
- Runbook automation
- Self-healing infrastructure
- Approval gates
- Blast radius
- Rollback
FAQ
- What is the difference between auto-remediation and runbook automation?
- Runbook automation executes a procedure someone wrote in advance. Auto-remediation is the broader goal of fixing faults without a human. Runbook automation is one way to get there, and agentic remediation is another.
- Is auto-remediation safe in production?
- It is safe when the action is reversible, limited in scope, dry-run before execution, verified afterwards, and only enabled for fault classes where it has a track record. It is not safe as a blanket switch.
- What can be auto-remediated in Kubernetes and AWS today?
- Common candidates are rolling back a deployment, scaling replicas or an Auto Scaling group, reverting a config change, cordoning a bad node and clearing a stuck rollout. Data migrations and anything that deletes state are usually excluded.