3725953 - Record missing in Microsoft Fabric mirrored table after stop/start of replication flow in SAP Datasphere

When configuring mirroring between SAP S/4HANA and Microsoft Fabric through SAP Datasphere, the following behavior may be observed:

• A record created in the source system just before stopping and restarting a replication flow is visible in the ADLS file storage but does not appear in the mirrored table in Microsoft Fabric.
• This occurs after a Stop/Start of the replication flow, whereas resuming the flow does not produce this issue.

SAP Datasphere

This behavior is expected and relates to how SAP Datasphere manages CDC checkpoints for mirroring scenarios:

Resume Run:

• When a replication flow is resumed, the existing CDC subscription and checkpoint are preserved.
• All pending changes, including those that occurred while the Cloud Connector or target was temporarily unavailable, are processed and delivered to both ADLS and the mirrored table.

Stop and Restart:

• Stopping the replication flow terminates the CDC session.
• Restarting creates a new CDC session from the last acknowledged checkpoint, which may exclude in-flight records that were written to ADLS but not yet consumed by the mirroring pipeline.
• As a result, some records appear in ADLS but are missing in the Fabric mirrored table.

This behavior does not occur when the target is a Datasphere table, since replication and persistence are handled internally within the same system.

The observed behavior is expected and is related to how SAP Datasphere manages CDC checkpoints for mirroring scenarios. When the replication flow is "resumed", the existing CDC subscription and checkpoint are preserved, allowing all pending changes that occurred while the Cloud Connector was unavailable to be correctly processed and delivered to both ADLS and the mirrored table in Microsoft Fabric.

However, when the replication flow is "stopped and restarted", the CDC session is terminated and a new session is created from the last acknowledged checkpoint, which may exclude in-flight records that were written to ADLS but not yet consumed and committed by the mirroring pipeline. 

As a result, such records can appear in ADLS but be missing in the Fabric mirrored table. This behavior does not occur when the target is a Datasphere table because replication and persistence are handled internally within the same system. For mirroring scenarios, SAP therefore recommends using "Resume Run" after temporary connectivity issues rather than Stop/Start to ensure delta continuity.

DWC, Datasphere, ADLS, Microsoft Fabric, S/4HANA, file, storage

KBA , DS-DI-RF , Replication Flow Runtime , Problem