CoreWeave Incident Runbook

Community-contributed. Not affiliated with, endorsed by, or sponsored by CoreWeave, Inc. CoreWeave is a registered trademark of CoreWeave, Inc.

Overview

Respond to GPU workload incidents by stabilizing customer impact, preserving redacted evidence, and restoring a known-good state. The incident commander owns communications and escalation; responders use only the access required for triage.

Prerequisites

• An incident ID, named commander, affected namespace/service, and on-call route.
• Authorized read-only cluster access plus a documented production rollback revision.
• A redaction policy for logs, prompts, model artifacts, and credentials.

Instructions

1. Declare severity and scope, then capture pod, event, node, and service status.
2. Stabilize impact with the documented scale, failover, or rollback action before root-cause work.
3. Collect only redacted, bounded diagnostics and escalate hardware or capacity issues through CoreWeave support.
4. Verify recovery against the service SLO, update stakeholders, and create follow-up work for root cause and prevention.

Triage Steps

# 1. Check pod status
kubectl get pods -l app=inference -o wide

# 2. Check recent events
kubectl get events --sort-by=.lastTimestamp | tail -20

# 3. Check node status
kubectl get nodes -l gpu.nvidia.com/class -o wide

# 4. Check GPU health
kubectl exec -it $(kubectl get pod -l app=inference -o name | head -1) -- nvidia-smi

Common Incidents

Inference Service Down

1. Check pod status and events
2. If OOMKilled: reduce batch size or upgrade GPU
3. If ImagePullBackOff: check registry credentials
4. If Pending: check GPU quota and availability

GPU Node Failure

1. Pods will be rescheduled automatically
2. If no capacity: scale down non-critical workloads
3. Contact CoreWeave support for extended outages

Model Loading Failure

1. Check HuggingFace token secret exists
2. Verify model name spelling
3. Check PVC has sufficient storage
4. Review container logs for download errors

Rollback

kubectl rollout undo deployment/inference

Output

• A time-stamped incident record with scope, owner, stabilization action, and redacted evidence.
• A verified recovery or an explicit escalation with a safe customer-impact mitigation.

Error Handling

Incident complication	Required response
Rollback fails	Stop repeated deploy attempts, escalate to the platform owner, and preserve events.
Diagnostics include a secret	Restrict distribution, rotate the secret, and recollect redacted evidence.
No GPU capacity is available	Prioritize critical services under approved policy; do not remove tenant quotas.
Recovery cannot meet SLO	Declare the continuing impact and use the approved fallback or communication path.

Examples

For a failing production rollout, stabilize first and capture only the relevant evidence:

kubectl -n inference-prod rollout undo deployment/inference
kubectl -n inference-prod rollout status deployment/inference --timeout=10m
kubectl -n inference-prod get events --sort-by=.lastTimestamp | tail -30

Record the revision, outcome, and redacted events in the incident. Do not retry a broken image or expose a troubleshooting endpoint publicly.

Resources

• CoreWeave Support
• CoreWeave Status

Next Steps

For data handling, see coreweave-data-handling.