Remediation library
The fixes that fire before on-call does.
A remediation library of playbook templates Paxis ships out of the box. Each one targets a confirmed anomaly, fires only when its signal window is met, and posts the resulting action to your Slack or PagerDuty channel the moment it closes.
Compute
Container-level CPU saturation and contamination.
CPU spike
CPU spike → pod restart
Catches sustained CPU saturation on a single container — when a pod pegs a core past a configured threshold for longer than the configured window, Paxis restarts the pod instead of waiting for an operator.
Action triggered
Pod restartSignals
container_cpu_usage > 0.9 * limitduration > 5mcooldown elapsed
Memory
Resident-set leaks and rolling deploys.
Memory leak
Memory leak → rollout restart
Detects monotonically rising resident-set memory across an entire deployment. Rather than restart a single pod and leave the leak in the others, Paxis triggers a rolling restart so the next pod starts clean.
Action triggered
Rollout restartSignals
rss_growth > 30% over 1hrolling avg climbno OOM event yet
Scheduling
Node liveness, drain, and reschedule paths.
Node not ready
Node not ready → drain & reschedule
When a node drops out of the Kubernetes ready set for longer than the safety window, Paxis cordons and drains it so the scheduler repaints its workloads onto healthy nodes — without losing replicas.
Action triggered
Drain & rescheduleGPU
Thermal and power throttling on accelerator nodes.
GPU throttle
GPU throttle → workload shed
Detects sustained GPU thermal or power throttling on a node. Paxis shifts the affected inference replicas to a healthy node and lowers the batch size on the original node so throughput stays predictable.
Action triggered
Workload shedFree tier available
Run a playbook on your own workload — 14-day trial
The 14-day free trial ships the same four playbooks end-to-end. Pair them with the detection + remediation loop, browse the integration recipes, and you have the full closed-loop story before the trial wraps.