‘TCP Provider: An Existing Connection Was Forcibly Closed by the Remote Host’ in MS Fabric SQL DB

Microsoft Fabric continues to unify data engineering, analytics, and storage across the OneLake platform, including fully managed SQL Database in Fabric. While the platform is designed to provide reliable replication and analytics integration, customers may occasionally encounter connectivity‑related errors—particularly during replication workflows.

One commonly reported issue appears in the Replication Monitor:

TCP Provider: An existing connection was forcibly closed by the remote host.
(provider: TCP Provider, error: 0)

This blog explains why this occurs, what scenarios trigger it, and how to resolve cases where a table does not appear in the SQL Analytics Endpoint, even though replication is enabled.

Scenario Overview

Working on support cases with customers reported:

• Replication Monitor shows
  “TCP Provider: An existing connection was forcibly closed by the remote host."
• A table (e.g., TestTable) is not visible in the SQL Analytics Endpoint.
• Dropping and recreating the table does not resolve the issue.
• Running diagnostic queries such as EXEC sp_help_change_feed shows metadata, but replication state does not align with the monitor.
• The issue has been present since the table was created, not something that started later.

What This Error Actually Means

According to Microsoft documentation on connection termination errors (Azure SQL & Fabric SQL):

• The TCP provider reports this error when the remote host forcibly terminates the connection.
• It is a client‑side error, not a SQL engine error.
• It may occur during:
  • Pre‑login handshake
  • Connection initialization
  • Login processing
  • Data‑stream operations

In Fabric SQL Database, this error can surface when replicators or mirroring agents reset or fail a connection unexpectedly.

Why This Happens in Fabric SQL Database Replication

• The issue is often tied to internal replication pipeline health.
• Replication may fail for a specific table even though other tables replicate successfully.
• The Fabric SQL replicator may hit an internal exception or unexpected disconnect.
• Because replication is asynchronous, intermittent platform‑side resets may trigger the TCP provider error on the client (replication monitor) even when the database itself is healthy.

One key detail:
Customers confirmed the issue existed since the table was first created, suggesting the replicator failed to initialize its metadata or state for that table.

Table Not Visible in SQL Analytics Endpoint

A second symptom is that the affected table does not appear in the SQL Analytics Endpoint, even after:

• Dropping and recreating the table
• Rechecking change feed metadata
• Confirming workspace and database permissions
• Executing EXEC sp_help_change_feed 

This indicates the replicator never successfully onboarded the table.

Backend Investigation

• The issue was isolated to publisher‑side replication components.
• Telemetry showed no systemic availability incidents.
• Restarting the replicator mitigated the problem in multiple cases.
• After restart, the table became visible in the SQL Analytics Endpoint, and replication turned green.

Thus, the root cause is typically replicator state corruption or partial initialization, not table schema or user configuration.

Available Workarounds Today

Fabric currently provides public REST APIs for customers to control mirroring/replication using the Mirroring – Start Mirroring API:

POST https://api.fabric.microsoft.com/v1/workspaces/{workspaceId}/sqlDatabases/{sqlDatabaseId}/startMirroring

This API allows:

• Restarting mirroring/replication
• Recovering replicator state without waiting for backend intervention

Important notes:

• UI controls for “Start / Stop Replicator” are not yet available but planned for future releases.
• Proper permissions are required (SQLDatabase.ReadWrite.All).

Recommended Troubleshooting Steps

1. Check whether the issue existed since table creation

If yes → strongly indicates replicator never initialized the table.

2. Validate replication metadata

Run:

EXEC sp_help_change_feed;

Compare table info with Replication Monitor output.

3. Try restarting the replicator

Use:

• REST API (available now) or
• Open a support ticket to request replicator restart.

4. Confirm table visibility in SQL Analytics Endpoint

Once the replicator is healthy, the table should appear automatically.

5. If issue persists

• Collect replicator traces (support can request internal telemetry)
• Provide workspace ID, database ID, and full error text
• Open a support ticket with Microsoft to get replicator restart by engineering team

Key Takeaways

• The error “TCP Provider: An existing connection was forcibly closed by the remote host” usually indicates a replicator‑side interruption, not a customer misconfiguration.
• Fabric SQL replication depends on internal mirroring pipelines; failures or state corruption for a specific table can prevent visibility in SQL Analytics Endpoint.
• Restarting the replicator often resolves the issue.
• REST APIs are available to automate this until UI controls ship.
• Dropping/recreating the table does not fix underlying replicator state issues.

Conclusion

Microsoft Fabric SQL Database provides powerful integrated replication for analytics, but occasional replication initialization issues can surface as TCP provider disconnection errors. Understanding the behavior of the replicator, combined with using available REST APIs and diagnostic tools, allows rapid mitigation and ensures tables appear in the SQL Analytics Endpoint as expected.

If you encounter this error consistently for a specific table, follow the steps above—or open a Microsoft support incident for deeper platform‑side investigation.