Best practices
Best practices are organized into Architecture & Design, Security, and Operations & Governance categories. You will also find best practices to employ before, during, and after deployment in this section, followed by helpful tips.
Architecture and Design
Deploy Option 1 (Cross-Account IAM Role) as your baseline — it is the only option that provides connection configuration, CloudWatch metrics, and resource tag metadata. No other option substitutes for it.
Layer complementary options based on operational need:
Add EventBridge (Option 2) + Health Events (Option 4) for real-time alerting
Add CloudWatch Network Synthetic Monitor (Option 3) for fault isolation and NHI metrics
In isolated partitions with limited service availability, Option 1 alone provides substantial value
Use the two-tier opt-in model:
Tier 1 — Organizational governance team approves and authorizes the IAM role template
Tier 2 — Individual account owners tag DX resources with
DXInfoSharing=enabledbefore metadata is accessible
Maintain partition independence — Use
AWS::Partitionin all CloudFormation ARN references; never hardcode partition strings.
Security
Enforce ExternalId by default (Deployment Model A) — Remove only when the ESP account is fully trusted and console role switching is required. Document the architectural decision.
Generate ExternalId as UUID v4 — unique per customer relationship, stored consistently, transmitted via encrypted channel, and rotated aligned with organizational security policy.
Grant least privilege rigorously:
Read-only actions only (Describe, List, Get)
No
iam:PassRole, noiam:Create*, noiam:Update*, noiam:Delete*IAM self-audit scoped exclusively to the operator role and policy ARNs
Enforce RoleSessionName conventions — Require descriptive session names (e.g.,
wsp-monitoring-system,noc-operator-jsmith) via SCPs for meaningful CloudTrail analysis.Scope CloudWatch queries operationally to
AWS/DX(andAWS/NetworkMonitorif applicable) — IAM cannot enforce namespace-level conditions on CloudWatch read APIs; discipline is enforced at the monitoring workflow level.
Operations and Governance
Tag all DX resources in your resiliency topology — Include connections across all locations so the WSP can assess redundancy during a failure.
Maintain resource tags as living metadata:
Tag on creation; update on contact change; remove on decommission
Include:
AccountPOC,TechnicalPOC,NOCContact,Criticality, circuit IDs, maintenance windows
Conduct quarterly access reviews — Verify business need, ExternalId currency, and authorized personnel. Document outcomes.
Implement graduated revocation — Know the four levels (scope reduction → session revocation → role disablement → role deletion) and escalate only as needed.
Execute a formal offboarding workflow when a WSP/ESP contract ends — revoke sessions, disable role, delete role, remove all tags, review CloudTrail, document actions.
Before Deployment
Identify your DX resiliency model (Maximum, High, or Dev/Test) — this determines how many connections need tagging and informs the WSP about failover behavior.
Establish a tagging standard — Define mandatory tags (DXInfoSharing, AccountPOC, TechnicalPOC, NOCContact, Criticality) and assign ownership for maintenance.
Make an architectural decision on ExternalId — Document whether you'll enforce it (Model A) or remove it for console compatibility (Model B). Get security team sign-off.
Coordinate with your WSP/ESP — Confirm they have an AWS account in the same partition and that their monitoring system supports STS AssumeRole. Exchange ExternalId via encrypted channel.
Verify service availability in your target partition/Region — Especially CloudWatch Network Synthetic Monitor and EventBridge if deploying Options 2–4.
Review your organization's cross-account access stance — Ensure your security team approves the trust policy pattern. Consider using AWS Organizations SCPs to restrict which external accounts can be trusted.
Assess CloudWatch VIF-level metrics availability — Since March 2026, AWS Direct Connect supports BGP health metrics and prefix count at the VIF level. Confirm these are available in your Region for enhanced provider visibility.
During Deployment
Deploy the CloudFormation stack using the operational runbook commands
Tag DX resources incrementally — Start with your most critical connections, validate WSP access, then expand.
Validate role assumption from the WSP account using CLI commands before relying on automated monitoring.
Confirm CloudTrail is capturing events for the
DXNetworkOperatorRole.Test with temporary credentials yourself — Before granting access to the WSP/ESP, test the role using aws sts assume-role with the External ID to validate the access scope and confirm least-privilege.
Verify tag-based scoping — Confirm that tag:GetResources with TagFilters for DXInfoSharing=enabled returns only the intended resources. Check for untagged resources that should be included.
After Deployment
Establish your operational cadence:
Weekly: governance compliance monitoring (if implemented)
Monthly: tag accuracy verification
Quarterly: formal WSP access review
Annually: ExternalId rotation
Measure MTTR improvement — Establish a baseline before deployment and compare incident timelines after.
Train NOC teams — Ensure your operations center knows the revocation procedures and escalation paths.
Plan for WSP transitions — If you change providers, execute the full offboarding workflow before onboarding the new WSP.
Integrate with AWS Security Hub — Feed Config rule findings and CloudTrail anomaly detections into Security Hub for centralized visibility and automated remediation workflows.
Enable CloudWatch Alarms for DX metrics — Set alarms on ConnectionState changes, optical light level thresholds (ConnectionLightLevelRx/Tx), and BGP status transitions. These complement the WSP's monitoring and provide customer-side awareness.
Document the relationship register — Maintain a register mapping each IAM role ARN, External ID, provider identity, tagged resources, and contract expiration date for audit readiness.
Tips for Isolated Partitions
Coordinate with your Cloud Provisioning Office (CPO) early — additional clearances may be required.
The tag-based metadata approach is especially valuable in isolated partitions where external collaboration tools are limited.
Verify DX, CloudWatch, and (if needed) Config availability in your target Region before building automation around them
In all AWS partitions, Option 1 (IAM role delegation) alone provides substantial operational value since EventBridge and Network Monitor may not be available.
Additional Tips and Recommendations
Use unique External IDs per relationship: Generate a UUID for each WSP/ESP engagement. Never reuse External IDs across providers.
Tag early, tag consistently: Apply DXInfoSharing tags at resource creation time using AWS CloudFormation, Terraform, or AWS Service Catalog. Retroactive tagging is error-prone.
Leverage CloudWatch VIF-level metrics: Since March 2026, AWS Direct Connect supports BGP health metrics and prefix count at the VIF level in CloudWatch. Share these with providers for faster BGP troubleshooting.
Automate governance with AWS Config: Deploy custom Config rules that check for: (a) DXInfoSharing tag presence on all DX resources, (b) IAM role policy compliance, (c) trust policy External ID presence.
Implement session duration limits: Set the maximum session duration on the IAM role to the shortest practical period (e.g., 1 hour). This limits the window of exposure if a session token is compromised.
Consider CloudWatch Metric Streams for real-time: For providers requiring near-real-time telemetry, consider CloudWatch Metric Streams to push DX metrics to a Kinesis Data Firehose destination the provider controls, instead of pull-based GetMetricData calls.
Adopt Infrastructure as Code: Define the IAM role, trust policy, and Config rules in AWS CloudFormation or Terraform. This ensures reproducibility, version control, and drift detection.
Review Well-Architected Framework Reliability Pillar: Ensure your overall DX architecture aligns with REL02-BP02 (Provision redundant connectivity between private networks) before optimizing telemetry sharing.