Safe deployment of Azure Policy assignments

As your environment expands, so does the demand for a controlled continuous deployment (CD) pipeline with progressive exposure control. Accordingly, We recommend DevOps teams follow the safe deployment practices (SDP) framework. The safe deployment of Azure Policy definitions and assignments helps limiting the impact of unintended behaviors of policy resources.

The high-level approach of implementing SDP with Azure Policy is to gradually rollout policy assignments by rings to detect policy changes that affect the environment in early stages before it affects the critical cloud infrastructure.

Deployment rings can be organized in diverse ways. In this how-to tutorial, rings are divided by different Azure regions with Ring 0 representing non-critical, low traffic locations, and Ring 5 denoting the most critical, highest traffic locations.

Steps for safe deployment of Azure Policy assignments with deny or append effects

Use the following flowchart as a reference as we work through how to apply the SDP framework to Azure Policy assignments that use the deny or append policy effects.

Flowchart step numbers:

1. Once you've selected your policy definition, assign the policy at the highest-level scope inclusive of all deployment rings. Apply resource selectors to narrow the applicability to the least critical ring by using the "kind": "resource location" property. Configure the audit effect type by using assignment overrides. Sample selector with ChinaEast location and effect as audit:

   "resourceSelectors": [{ 
     "name": "SDPRegions", 
     "selectors": [{
         "kind": "resourceLocation",
         "in": [ "ChinaEast" ] 
     }]
   }], 
   "overrides":[{ 
     "kind": "policyEffect", 
     "value": "Audit" 
   }] 
   

2. Once the assignment is deployed and the initial compliance scan has completed, validate that the compliance result is as expected.

   You should also configure automated tests that run compliance checks. A compliance check should encompass the following logic:

   • Gather compliance results
   • If compliance results are as expected, the pipeline should continue
   • If compliance results aren't as expected, the pipeline should fail and you should start debugging

   For example, you can configure the compliance check by using other tools within your particular continuous integration/continuous deployment (CI/CD) pipeline.

   At each rollout stage, the application health checks should confirm the stability of the service and impact of the policy. If the results aren't as expected due to application configuration, refactor the application as appropriate.

3. Repeat by expanding the resource selector property values to include the next rings. locations and validating the expected compliance results and application health. Example selector with an added location value:

   "resourceSelectors": [{ 
     "name": "SDPRegions", 
     "selectors": [{
         "kind": "resourceLocation",
         "in": [ "ChinaEast", "ChinaNorth"] 
     }]
   }]
   

4. Once you have successfully assigned the policy to all rings using audit mode, the pipeline should trigger a task that changes the policy effect to deny and reset the resource selectors to the location associated with Ring 0. Example selector with one region and effect set to deny:

   "resourceSelectors": [{ 
     "name": "SDPRegions", 
     "selectors": [{
         "kind": "resourceLocation",
         "in": [ "ChinaEast" ] 
     }]
   }], 
   "overrides":[{ 
     "kind": "policyEffect", 
     "value": "Deny" 
   }] 
   

5. Once the effect is changed, automated tests should check whether enforcement is taking place as expected.

6. Repeat by including more rings in your resource selector configuration.

7. Repeat this process for all production rings.

Steps for safe deployment of Azure Policy assignments with modify or deployIfNotExists effects

The steps for policies using the modify or deployIfNotExists effects are similar to steps previously explained with the additional action of using enforcement mode and triggering a remediation task. Review the following flowchart with modified steps 5-9:

Flowchart step numbers:

1. Once you've selected your policy definition, assign the policy at the highest-level scope inclusive of all deployment rings. Apply resource selectors to narrow the applicability to the least critical ring by using the "kind": "resource location" property. Configure the enforcement mode of the assignment to DoNotEnforce. Sample selector with ChinaEast2 location and enforcementMode as DoNotEnforce:

   "resourceSelectors": [{ 
     "name": "SDPRegions", 
     "selectors": [{
         "kind": "resourceLocation",
         "in": [ "ChinaEast2" ] 
     }]
   }], 
   "enforcementMode": "DoNotEnforce"
   

2. Once the assignment is deployed and the initial compliance scan has completed, validate that the compliance result is as expected.

   You should also configure automated tests that run compliance checks. A compliance check should encompass the following logic:

   • Gather compliance results
   • If compliance results are as expected, the pipeline should continue
   • If compliance results aren't as expected, the pipeline should fail and you should start debugging

   You can configure the compliance check by using other tools within your continuous integration/continuous deployment (CI/CD) pipeline.

   At each rollout stage, the application health checks should confirm the stability of the service and impact of the policy. If the results aren't as expected due to application configuration, refactor the application as appropriate.

   You may also trigger remediation tasks to remediate existing non-compliant resources. Ensure the remediation tasks are bringing resources into compliance as expected.

3. Repeat by expanding the resource selector property values to include the next ring's locations and validating the expected compliance results and application health. Example selector with an added location value:

   "resourceSelectors": [{ 
     "name": "SDPRegions", 
     "selectors": [{
         "kind": "resourceLocation",
         "in": [ "eastUS", "ChinaEast2"] 
     }]
   }]
   

4. Once you have successfully assigned the policy to all rings using DoNotEnforce mode, the pipeline should trigger a task that changes the policy enforcementMode to Default enablement and reset the resource selectors to the location associated with Ring 0. Example selector with one region and effect set to deny:

   "resourceSelectors": [{ 
     "name": "SDPRegions", 
     "selectors": [{
         "kind": "resourceLocation",
         "in": [ "ChinaEast2" ] 
     }]
   }], 
   "enforcementMode": "Default",
   

5. Once the effect is changed, automated tests should check whether enforcement is taking place as expected.

6. Repeat by including more rings in your resource selector configuration.

7. Repeat this process for all production rings.

Next steps

• Learn how to programmatically create policies.
• Review Azure Policy as code workflows.
• Study Microsoft's guidance concerning safe deployment practices.
• Review Remediate non-compliant resources with Azure Policy.