Skip to content

Troubleshooting: Control 2.17 - Multi-Agent Orchestration Limits

Last Updated: January 2026

Common Issues

Issue Cause Resolution
Depth limit not enforcing Tracking variable not implemented Verify depth tracking in agent topics
Circuit breaker stuck open Reset timeout too long or failures continue Check downstream agent health; adjust reset timeout
Cascade failures occurring Circuit breaker not configured Implement circuit breaker pattern
HITL timeouts causing abandonment Timeout too short or approvers unavailable Adjust timeout; ensure approver coverage
Orchestration too slow Multiple sequential calls Consider parallel calls where safe

Detailed Troubleshooting

Issue: Delegation Depth Limit Not Enforcing

Symptoms: Agents can chain beyond configured depth limits

Diagnostic Steps:

  1. Verify depth tracking variable exists: 

    Copilot Studio > Agent > Topics > Check for orchestration_depth variable

  2. Check depth increment logic:

  3. Variable should increment before each delegation

  4. Should be passed to delegated agent

  5. Verify depth check condition:

  6. Condition should compare against max_depth
  7. Should block or error if exceeded

Resolution:

  • Implement depth tracking if missing
  • Fix increment logic (ensure it increments BEFORE delegation)
  • Add proper condition check before delegation calls
  • Consider using Power Automate for complex orchestration with better control

Issue: Circuit Breaker Stuck Open

Symptoms: Delegated agent calls permanently blocked even after issue resolved

Diagnostic Steps:

  1. Check circuit breaker state in your monitoring system

  2. Verify the downstream agent is actually healthy:

  3. Test direct calls to the agent

  4. Check agent health metrics

  5. Check reset timeout configuration:

  6. Timeout may be longer than expected
  7. Half-open test may be failing

Resolution:

  • Manually reset circuit breaker if available
  • Fix downstream agent issues
  • Adjust reset timeout to appropriate duration
  • Verify half-open test is configured correctly

Issue: Cascade Failures in Orchestration

Symptoms: One agent failure causes entire chain to fail

Diagnostic Steps:

  1. Check if circuit breakers are implemented

  2. Review error handling in orchestrating agents:

  3. Are failures being caught?

  4. Is there fallback behavior?

  5. Check timeout configuration:

  6. Timeouts should be shorter than total allowed time
  7. Cascading timeouts should not exceed total

Resolution:

  • Implement circuit breakers on all agent-to-agent calls
  • Add proper error handling with fallbacks
  • Configure appropriate timeouts at each level
  • Consider bulkhead pattern for isolation

Issue: HITL Causing User Abandonment

Symptoms: Users leave before HITL approval completes

Diagnostic Steps:

  1. Check HITL timeout configuration:
  2. Is timeout appropriate for the approval process?

  3. Are approvers available during business hours?

  4. Review approval routing:

  5. Are requests going to available approvers?

  6. Is there a backup approver chain?

  7. Check user communication:

  8. Are users informed of the wait?
  9. Is there a way to check status?

Resolution:

  • Adjust timeout based on actual approval times
  • Implement backup approver chain
  • Provide status updates to waiting users
  • Consider async patterns where appropriate

How to Confirm Configuration is Active

Depth Limiting

  1. Create a test scenario that would exceed depth
  2. Verify the request is blocked
  3. Check logs for depth violation event

Circuit Breaker

  1. Simulate failures in a test environment
  2. Verify circuit opens after threshold
  3. Wait for reset and verify it closes

HITL Checkpoints

  1. Trigger a HITL-required operation
  2. Verify approval request is generated
  3. Test approval and denial paths

Escalation Path

If issues persist after troubleshooting:

  1. Copilot Studio Admin - Agent configuration issues
  2. Power Automate Admin - Flow-based orchestration issues
  3. AI Governance Lead - Policy and limit questions
  4. Microsoft Support - Platform limitations

Known Limitations

Limitation Impact Workaround
No native depth tracking Must implement manually Use conversation variables; document pattern
Circuit breaker not built-in Requires custom implementation Use Power Automate with error handling
Limited visibility into chains Hard to trace multi-agent flows Implement correlation IDs; use Application Insights
HITL requires polling No native webhook support Use Power Automate adaptive cards
Cross-environment orchestration complex Multi-env chains hard to manage Keep orchestration within single environment

Back to Control 2.17 | Portal Walkthrough | PowerShell Setup | Verification Testing