Agent Skills: Oracle Cloud Incident Runbook

Oracle Cloud Incident Runbook

Overview

Self-service runbook for when OCI instances go down and the status page stays green. OCI's status page has a history of not acknowledging outages in real time (London Jan 2026 — 502s and instances disappearing for 10 minutes with no status update). OCI Support response times average 4+ hours for Sev-1 tickets. This runbook gives you health probes, automated instance recovery, cross-AD failover, and cross-region failover — all executable without waiting on Oracle.

Purpose: Detect OCI service degradation independently, recover instances automatically, and fail over to alternate availability domains or regions when the primary is impacted.

Prerequisites

• OCI CLI installed and configured — ~/.oci/config validated (see oraclecloud-install-auth)
• Python 3.8+ with the OCI SDK — pip install oci
• Pre-configured resources: at least one compute instance, a VCN with subnets in multiple ADs
• IAM policies: manage instances, manage volumes, inspect work-requests in the target compartment
• Boot volume backups enabled (recovery depends on having a recent backup)

Instructions

Step 1: Independent Health Probes

Do not trust the OCI status page alone. Run your own health checks against the OCI API:

import oci
import time

config = oci.config.from_file("~/.oci/config")

def probe_oci_health(config):
    """Probe OCI API endpoints independently of the status page."""
    results = {}

    # Probe 1: Identity service (lightest call)
    try:
        start = time.time()
        identity = oci.identity.IdentityClient(config)
        identity.list_regions()
        results["identity"] = {"status": "healthy", "latency_ms": int((time.time() - start) * 1000)}
    except oci.exceptions.ServiceError as e:
        results["identity"] = {"status": "degraded", "error": str(e.status)}

    # Probe 2: Compute service
    try:
        start = time.time()
        compute = oci.core.ComputeClient(config)
        compute.list_instances(compartment_id=config["tenancy"], limit=1)
        results["compute"] = {"status": "healthy", "latency_ms": int((time.time() - start) * 1000)}
    except oci.exceptions.ServiceError as e:
        results["compute"] = {"status": "degraded", "error": str(e.status)}

    # Probe 3: Networking service
    try:
        start = time.time()
        network = oci.core.VirtualNetworkClient(config)
        network.list_vcns(compartment_id=config["tenancy"], limit=1)
        results["networking"] = {"status": "healthy", "latency_ms": int((time.time() - start) * 1000)}
    except oci.exceptions.ServiceError as e:
        results["networking"] = {"status": "degraded", "error": str(e.status)}

    return results

health = probe_oci_health(config)
for service, status in health.items():
    print(f"  {service}: {status['status']} ({status.get('latency_ms', 'N/A')}ms)")

Step 2: Severity Classification

| Level | Condition | Detection | Response Time | |-------|-----------|-----------|---------------| | P1 — Total Outage | All API probes fail, instances unreachable | All 3 probes return degraded | Immediate — trigger cross-region failover | | P2 — Partial Degradation | Some APIs slow (>2s latency), instance agent unresponsive | Latency >2000ms on any probe | 15 min — attempt instance recovery, prepare failover | | P3 — Intermittent | Sporadic 429/500 errors, some requests succeed | Occasional probe failures | 1 hour — enable retries, monitor trend |

Step 3: Instance Recovery Actions

# Check current instance state
oci compute instance get --instance-id "$INSTANCE_OCID" \
  --query 'data."lifecycle-state"' --raw-output

# Action: Reset (hard reboot) — fastest recovery for hung instances
oci compute instance action --instance-id "$INSTANCE_OCID" \
  --action RESET

# Action: Reboot (graceful) — for instances still responding to ACPI
oci compute instance action --instance-id "$INSTANCE_OCID" \
  --action SOFTRESET

# Action: Stop then Start — forces reallocation to new host hardware
oci compute instance action --instance-id "$INSTANCE_OCID" \
  --action STOP
# Wait for STOPPED state
oci compute instance action --instance-id "$INSTANCE_OCID" \
  --action START

Step 4: Cross-AD Failover

When an entire availability domain is impacted, launch a replacement instance in a different AD using the latest boot volume backup:

import oci

config = oci.config.from_file("~/.oci/config")
compute = oci.core.ComputeClient(config)
blockstorage = oci.core.BlockstorageClient(config)

# Find latest boot volume backup
backups = blockstorage.list_boot_volume_backups(
    compartment_id="COMPARTMENT_OCID",
    boot_volume_id="BOOT_VOLUME_OCID",
    sort_by="TIMECREATED",
    sort_order="DESC",
    limit=1,
).data

if not backups:
    raise RuntimeError("No boot volume backups found — cannot failover")

latest_backup = backups[0]
print(f"Using backup: {latest_backup.id} from {latest_backup.time_created}")

# Launch replacement instance in alternate AD
launch_details = oci.core.models.LaunchInstanceDetails(
    availability_domain="AD-2",  # Different from the impacted AD
    compartment_id="COMPARTMENT_OCID",
    shape="VM.Standard.E4.Flex",
    shape_config=oci.core.models.LaunchInstanceShapeConfigDetails(ocpus=2, memory_in_gbs=16),
    display_name="failover-instance",
    source_details=oci.core.models.InstanceSourceViaBootVolumeDetails(
        source_type="bootVolume",
        boot_volume_id=latest_backup.id,
    ),
    create_vnic_details=oci.core.models.CreateVnicDetails(
        subnet_id="SUBNET_OCID_IN_AD2",
    ),
)

response = compute.launch_instance(launch_details)
print(f"Failover instance launching: {response.data.id}")

Step 5: Cross-Region Failover

For region-wide outages, use a pre-replicated boot volume in the DR region:

# List available boot volume replicas in DR region
oci bv boot-volume-backup list \
  --compartment-id "$COMPARTMENT_OCID" \
  --query 'data[0].id' --raw-output \
  --region us-phoenix-1

# Launch instance in DR region
oci compute instance launch \
  --compartment-id "$COMPARTMENT_OCID" \
  --availability-domain "AD-1" \
  --shape "VM.Standard.E4.Flex" \
  --shape-config '{"ocpus":2,"memoryInGBs":16}' \
  --display-name "dr-failover-instance" \
  --source-boot-volume-id "$DR_BACKUP_OCID" \
  --subnet-id "$DR_SUBNET_OCID" \
  --region us-phoenix-1

Step 6: Status Monitoring Script

Run this in a loop during an incident to detect recovery:

#!/bin/bash
while true; do
  STATUS=$(oci compute instance get --instance-id "$INSTANCE_OCID" \
    --query 'data."lifecycle-state"' --raw-output 2>&1)
  TIMESTAMP=$(date -u +%H:%M:%S)
  echo "$TIMESTAMP | Instance: $STATUS"
  if [ "$STATUS" = "RUNNING" ]; then
    echo "$TIMESTAMP | Instance recovered."
    break
  fi
  sleep 30
done

Output

Successful execution produces:

• Independent health probe results for Identity, Compute, and Networking services
• Severity classification (P1/P2/P3) based on probe results
• Instance recovery action (reset, reboot, or stop/start) applied to the impacted instance
• Cross-AD failover instance launched from the latest boot volume backup (if AD-level failure)
• Cross-region failover instance launched in the DR region (if region-level failure)
• Continuous monitoring log tracking instance lifecycle state until recovery

Error Handling

| Error | Code | Cause | Solution | |-------|------|-------|----------| | NotAuthenticated | 401 | API key expired during incident | Use oci session authenticate for token-based short-term auth | | NotAuthorizedOrNotFound | 404 | IAM policy missing for recovery actions | Pre-create policy: allow group sre to manage instances in compartment prod | | TooManyRequests | 429 | Rate limited during mass recovery | OCI has no Retry-After header — back off 30 seconds between calls | | InternalError | 500 | OCI service itself is degraded | This confirms the outage — proceed to cross-region failover | | No boot volume backups | — | Backups not configured | Cannot failover without backups — enable in Block Storage > Boot Volumes > Backup Policies | | ServiceError status -1 | — | API timeout (region unreachable) | Switch to DR region immediately: --region us-phoenix-1 |

Examples

Quick health check (CLI one-liner):

# Test if OCI API is responsive
oci iam region list --output table && echo "OCI API: OK" || echo "OCI API: UNREACHABLE"

Automated recovery wrapper:

#!/bin/bash
set -euo pipefail
INSTANCE_OCID="${1:?Usage: recover.sh <instance-ocid>}"
STATE=$(oci compute instance get --instance-id "$INSTANCE_OCID" \
  --query 'data."lifecycle-state"' --raw-output)
case "$STATE" in
  RUNNING) echo "Instance is healthy" ;;
  STOPPED) oci compute instance action --instance-id "$INSTANCE_OCID" --action START ;;
  *)       oci compute instance action --instance-id "$INSTANCE_OCID" --action RESET ;;
esac

Resources

• OCI Instance Recovery — stop, start, reboot, and reset actions
• OCI Status Page — official service health (may lag actual incidents)
• OCI CLI Reference — command-line interface documentation
• OCI Python SDK Reference — ComputeClient, BlockstorageClient APIs
• OCI Known Issues — platform-known issues and workarounds
• SDK Troubleshooting — connectivity and timeout debugging

Next Steps

After stabilizing the incident, run oraclecloud-debug-bundle to collect forensic data for the postmortem, or review oraclecloud-prod-checklist to harden your environment against future outages.