SQL Server to Kafka
real-time replication
TwinCommit replicates Microsoft SQL Server to Apache Kafka in real time by reading the SQL Server transaction log and applying every committed change exactly once and in commit order. Use it to publish every committed SQL Server change as an event stream, with no triggers or capture tables on the source.
SQL Server
log reader
commit-ordered · exactly-once
Kafka
native writer
How capture works on SQL Server
TwinCommit reads the active and backed-up transaction log directly, so you do not need SQL Server Agent CDC jobs or change tables.
Prerequisites
- ✓FULL recovery model
- ✓Primary key on each replicated table
- ✓VIEW SERVER STATE and db_owner on the source database
How apply works on Kafka
Each source transaction is published with Kafka transactions, so consumers using read_committed never see a partial transaction. Messages are keyed by primary key in JSON, Avro or Protobuf.
Requirements
- ✓A topic naming template
- ✓Producer ACLs including transactional.id
- ✓Optional Schema Registry for Avro and Protobuf
Setup
Set up SQL Server → Kafka in four steps
- 1
Prepare SQL Server
Run twincommit check. It verifies full recovery model and the other prerequisites, and prints the exact fix for anything missing.
- 2
Point at Kafka
Add a target block for Kafka. TwinCommit can generate target DDL with twincommit ddl -apply.
- 3
Snapshot
twincommit snapshot copies each table under a brief share lock and records the exact log position.
- 4
Stream
twincommit run streams every committed change from that position, exactly once and in commit order.
source:
type: sqlserver
dsn: ${SOURCE_DSN}
target:
type: kafka
bootstrap_servers: broker-1:9092,broker-2:9092
topic: "{schema}.{table}"
format: avro
schema_registry: http://registry:8081
tables:
- source: SALES.ORDERS
- source: SALES.ORDER_LINES
exclude_columns: [INTERNAL_NOTES]SQL Server to Kafka replication FAQ
How does TwinCommit replicate SQL Server to Kafka?
+
TwinCommit reads changes from the SQL Server transaction log, buffers them per transaction, and applies each committed transaction to Kafka in commit order. Each source transaction is published with Kafka transactions, so consumers using read_committed never see a partial transaction. Messages are keyed by primary key in JSON, Avro or Protobuf.
Does SQL Server to Kafka replication need triggers or capture tables?
+
No. TwinCommit reads the SQL Server log directly and creates no objects in the source database.
Is SQL Server to Kafka replication exactly-once?
+
Yes. TwinCommit commits its checkpoint in the same Kafka transaction as the data, so a crash or restart never loses or duplicates a transaction.
What SQL Server versions are supported?
+
SQL Server 2016–2025 · Azure SQL Managed Instance · Azure SQL Database (target).
Stream SQL Server to Kafka today.
Download the free Developer edition, or have one of our replication engineers walk you through a migration plan.