Merge pull request #220307 from abdullahzmaili/patch-1

Clarify restart provisioning operation

Author: prmerger-automator[bot]

articles/active-directory/app-provisioning/how-provisioning-works.md

@@ -136,7 +136,23 @@ After the initial cycle, all other cycles will:
 The provisioning service continues running back-to-back incremental cycles indefinitely, at intervals defined in the [tutorial specific to each application](../saas-apps/tutorial-list.md). Incremental cycles continue until one of the following events occurs:
 
 - The service is manually stopped using the Azure portal, or using the appropriate Microsoft Graph API command.
-- A new initial cycle is triggered using the **Restart provisioning** option in the Azure portal, or using the appropriate Microsoft Graph API command. This action clears any stored watermark and causes all source objects to be evaluated again.
+- A new initial cycle is triggered using the **Restart provisioning** option in the Azure portal, or using the appropriate Microsoft Graph API command. This action clears any stored watermark and causes all source objects to be evaluated again. This will not break the links between source and target objects. To break the links use [Restart synchronizationJob](https://learn.microsoft.com/graph/api/synchronization-synchronizationjob-restart?view=graph-rest-beta&tabs=http) with the following request: 
+
+# [HTTP](#tab/http)
+<!-- {
+  "blockType": "request",
+  "name": "synchronizationjob_restart"
+}-->
+```http
+POST https://graph.microsoft.com/beta/servicePrincipals/{id}/synchronization/jobs/{jobId}/restart
+Authorization: Bearer <token>
+Content-type: application/json
+{
+   "criteria": {
+       "resetScope": "Full"
+   }
+}
+```
 - A new initial cycle is triggered because of a change in attribute mappings or scoping filters. This action also clears any stored watermark and causes all source objects to be evaluated again.
 - The provisioning process goes into quarantine (see below) because of a high error rate, and stays in quarantine for more than four weeks. In this event, the service will be automatically disabled.