Managed change data capture with Debezium
Debezium reads your database's change log and publishes every change as an event. OSStend designs the pipeline, moves you onto it and runs the connectors, offsets and upgrades 24/7: in OSStend's cloud, inside your cloud account or co-managed in your data center.
45 minutes with an engineer: we review your workload and sketch the design.
- ✓Operated by AceMQ: 160+ customers, 10+ years
- ✓24/7 on-call engineers
Managed Debezium from OSStend is for teams that chose Debezium for change data capture and need engineers to own it. We run Debezium on Kafka Connect, on MSK Connect or as standalone Debezium Server, watch connector health and lag, back up offsets and schema history, and plan Debezium and Kafka upgrades together.
- PostgreSQL
- MySQL
- MongoDB
- Redis
- Valkey
- OpenSearch
- Kafka
- RabbitMQ
- AWS
- Google Cloud
- Azure
- Kubernetes
- OpenShift
- Prometheus
- Grafana
- Datadog
- Dynatrace
- Splunk
What managed Debezium means, and who runs what
Debezium is an open-source change data capture (CDC) platform that reads a database's transaction log and emits one event per row change. Managed Debezium means OSStend engineers run the connectors and what they depend on, and you keep the source databases and the applications that consume the events.
| Layer | What it covers | Who runs it |
|---|---|---|
| Source database settings | Log level, retention of change logs, replication slots or CDC tables, and the connector's database user. Set in the design, applied with your DBA. | Co-owned: your DBA changes the database, OSStend specifies and monitors the settings |
| Debezium connectors or Debezium Server | Connector configuration, snapshots, heartbeats, schema history, restarts and version upgrades. | OSStend |
| Kafka and Kafka Connect, or the sink | The brokers and Connect workers, or the RabbitMQ, Kinesis or Pub/Sub target a standalone Debezium Server writes to. | OSStend, or your platform team where the sink is yours |
| Consumers | The applications, warehouses and search indexes that read the events. | You |
The split is written into the design you sign off, so the on-call team knows which side of each boundary it owns.
Debezium is open source under the Apache License 2.0, checked 18 Nov 2026. There is no licence fee for the software: what you pay for is the operating of it. For the background, read what change data capture is and how the tools compare.
Debezium on Kafka Connect, MSK Connect or Debezium Server
Debezium runs three common ways: as connectors inside Kafka Connect, on Amazon MSK Connect, or as the standalone Debezium Server that needs no Kafka. The runtime sets the sink options and what has to be operated, so it is chosen in the design review.
| Runtime | What it is | Fits when |
|---|---|---|
| Kafka Connect | Debezium source connectors run in a Kafka Connect cluster next to Kafka, so events land in topics. Connect workers, offsets and the schema history topic live in Kafka. | Kafka is already the backbone and several consumers read the same changes. See managed Kafka. |
| Amazon MSK Connect | The same connectors run as an AWS-managed Kafka Connect service writing to Amazon MSK. AWS runs the connector infrastructure; Debezium configuration, behaviour and recovery stay with whoever supports Debezium. | The estate is on AWS and MSK. OSStend owns the Debezium side; see the Amazon MSK comparison. |
| Debezium Server | A standalone application that streams change events from a source database to other systems without Kafka. Its documented sinks include Apache Kafka, Amazon Kinesis, Google Cloud Pub/Sub and RabbitMQ. | You do not need Kafka and want fewer moving parts, for example on Kubernetes with a RabbitMQ or Kinesis target. |
Debezium Server's sinks and offset storage options are listed in the Debezium Server documentation, checked 18 Nov 2026. A third form, the Debezium Engine, embeds the connector in your own Java application; we do not run that on your behalf.
If you are weighing Debezium against AWS's migration service for the same job, see Debezium vs AWS DMS.
Sources and sinks OSStend runs Debezium against
Debezium has source connectors for SQL Server, Oracle, PostgreSQL, MySQL, MongoDB and Db2, among others. Each reads the database's own change mechanism, so each fails in its own way, and each gets its own checks.
| Source | What the connector reads | What we watch |
|---|---|---|
| PostgreSQL | Committed changes through logical decoding, using a replication slot. | Slot lag and WAL growth, heartbeats on quiet databases, failover and slot loss. See Postgres CDC. |
| MySQL | The binary log. | Binlog retention against the longest outage the design allows. |
| SQL Server | The change tables that SQL Server's own CDC fills. | SQL Server Agent and cleanup jobs, log size, capture instance changes after schema changes. |
| Oracle | The redo log through LogMiner. | Log retention, archived log availability and the offset position (SCN). |
| MongoDB | Changes from the replica set or cluster. | The oplog window against connector downtime. |
| Db2 | The change data Db2 captures for the connector. | Capture jobs and prerequisites, which differ by Db2 platform. |
Source list from the Debezium connector index, which also lists Cassandra, MariaDB and others, checked 18 Nov 2026. SQL Server snapshot, schema history and offset behaviour is described in the SQL Server connector documentation. Which connector versions run where is settled in the design review.
Sinks follow the runtime. Kafka Connect writes to Kafka topics; Debezium Server can write to Kafka, RabbitMQ, Amazon Kinesis and Google Cloud Pub/Sub, and to other systems its documentation lists. Downstream of the topics, sink connectors and stream processors are part of the design when you want OSStend to run them.
What we run 24/7 for Debezium
Running Debezium well is mostly operating the things around the connector: its offsets, its schema history, the source's logs and the Kafka beneath it. Engineers on call handle these, with alerts forwarded to your own monitoring.
- ✓Connector health. Connector and task state, restarts after failures, and alerts when a connector is up but not moving.
- ✓Lag alerts. Lag between the source log and the connector, and between the connector and the consumers, so a slow stage is found before consumers notice.
- ✓Offset and schema-history backup. Offsets say where the connector stopped; the schema history says what each table looked like then. Both are backed up and restore-tested, because a connector that loses either cannot resume cleanly.
- ✓Snapshot and re-snapshot runbooks. The initial snapshot, a targeted re-snapshot of a table, and recovery of a lost schema history topic, using the snapshot modes Debezium documents.
- ✓Upgrades of Debezium and Kafka together. Connector, Connect and broker versions move in one plan, rehearsed in a non-production environment with a rollback path, on your change calendar.
- ✓Disaster recovery for offsets. The design states how offsets and schema history reach the recovery site and how the connector resumes there without a manual copy.
- ✓Source database changes. Schema changes, capture instance changes and log retention settings are reviewed with your DBA before they hit the connector.
- ✓Monitoring. Metrics and alerts forwarded to your Datadog, Dynatrace, Splunk or Prometheus, with runbooks per alert.
There is no fixed menu here: the connectors, worker counts and alert thresholds your workload needs are written into the design you sign off before anything is built, and that design is what OSStend runs.
The failures managed Debezium is built to handle
Debezium pipelines fail in a handful of repeatable ways: lag that grows for hours, log positions the connector never read, offsets that cannot be restored, upgrades nobody dares to run and a source that deletes its logs. Each has a runbook.
| Failure | Why it happens | What we do |
|---|---|---|
| Lag grows to hours | One stage is the bottleneck: a single connector or worker, a slow sink, a large transaction or a snapshot running alongside streaming. | Measure each stage, then split the load across tasks or workers, tune the sink, or schedule the snapshot apart from streaming. |
| Skipped log positions, missing updates | The source purged the log before the connector read it, or a connector was restarted from the wrong position. | Size source log retention for the longest planned outage, alert on the gap between the stored offset and the oldest log available, and re-snapshot the affected tables when a gap is real. |
| Offsets copied by hand for DR | The offset store was never designed as part of disaster recovery. | Replicate or back up offsets and schema history as part of the design, and test the failover resume. |
| Nobody wants to upgrade | Connector, Connect and Kafka versions are coupled and the last upgrade hurt. | Upgrade review, non-production rehearsal, a rollback plan and one change window for the whole stack. |
| The source's logs are removed | Maintenance or cleanup jobs on the source delete change data a reader still needs. | Pause or tune source-side cleanup while a connector is down, so it can catch up; we write that into the runbook. |
| WAL fills the disk (PostgreSQL) | A replication slot holds WAL until its consumer confirms it, so a stopped connector keeps the disk growing. | Alert on slot lag, use heartbeats, and cap retention with max_slot_wal_keep_size where the design accepts a re-snapshot. |
Slot behaviour is in the PostgreSQL logical decoding documentation and the replication settings; the connector side is in the Debezium PostgreSQL connector documentation, checked 18 Nov 2026.
From our on-call engineers: what we test before CDC goes to production
Most CDC trouble shows up the first time something breaks, so we break things on purpose before go-live. These are the checks we have used on Debezium pipelines for customers, and they are the first things we run on yours.
- ✓We designed and hardened a SQL Server CDC pipeline that runs standalone Debezium Server, publishing to RabbitMQ on Kubernetes, and reviewed it with hands-on scenarios before it went near production.
- ✓Failure and recovery tests are part of that review: node failure and reconnect, and a broker outage, where we confirm the connector stops, restarts by itself and resumes without losing messages.
- ✓We size the source's log retention for hours of downtime, because while Debezium is down the source has to keep its change logs until the connector catches up.
- ✓We check that delete events are captured. A proof of concept can miss them until the configuration is corrected, so it is a named test.
- ✓Retry and dead-letter handling, tracing and monitoring are designed in, and every alert has a runbook an on-call engineer can follow at 3 a.m.
- ✓For an Oracle, Kafka and Debezium estate we have reviewed a Debezium and Kafka upgrade before production and designed disaster recovery, where offsets had been copied by hand until then.
Security and network paths
A CDC pipeline copies the contents of your database, so it is run like the database: private network paths, encrypted connections and secrets in your own store. Where data may not leave a site, the data center option runs it on your hosts, air-gapped sites included.
The connector's database user gets only the privileges the connector needs. Connections between the source, Debezium, Kafka and the consumers use TLS, and credentials come from your secrets manager rather than connector files. The security page lists the controls that apply across OSStend, and deployment shows where each option runs.
Design. Move. Run. for change data capture
For change data capture, an engineer designs the runtime, connectors and recovery plan and you sign it off, OSStend moves you across with both paths live, and named engineers then run it 24/7.
Design
An engineer reviews the source databases, the change volume, the consumers and the recovery targets, then writes the design: runtime, connectors, sinks, log retention, offset and schema-history storage and where it runs. You sign it off before anything is built.
Move
OSStend deploys the pipeline next to your current one, snapshots the sources and compares both sides until the events match, then moves consumers across in stages. If you are leaving DMS, a managed connector or a self-run Debezium, the old path stays live until the new one is verified.
Run
Named engineers run it 24/7: connector health and lag, offsets and schema history, snapshots, upgrades and incident response, with the response commitment written into your contract.
Deployment and service levels
Debezium runs in OSStend's cloud, inside your own cloud account, or co-managed in your data center. The first two are fully managed; in a data center OSStend engineers operate it on infrastructure you own.
Dedicated in OSStend's cloud
Connectors and Kafka or Debezium Server on single-tenant infrastructure on Google Cloud or AWS, with a private path to your source databases.
Fully managedIn your cloud account
Debezium, Kafka Connect or MSK Connect inside your cloud account and VPC, so the change data and the bill stay with you. AWS first, then Azure and Google Cloud.
Fully managedIn your data center
Debezium on your hosts, Kubernetes or OpenShift, next to the source databases, air-gapped sites included, delivered through an operator with supported charts and images.
Co-managed
Fully managed deployments carry an uptime commitment, and a data center deployment carries a 24/7 response-time commitment because the hosts and disks are yours. The service levels page sets out the tiers.
Uptime and response figures are set per plan and written into your contract. AceMQ, which operates OSStend, sets its support response targets from 15 minutes, by plan.
Moving to managed Debezium from DMS, a managed connector or a self-run setup
Teams usually arrive from one of three places: AWS DMS, a vendor's managed CDC connector, or a Debezium they run themselves. The move is the same: build the new pipeline beside the old, prove the events match, then switch consumers.
- From AWS DMS
- compare the two in Debezium vs AWS DMS, then run Debezium beside the DMS task until the targets agree.
- From Confluent Cloud or another managed Kafka
- see Confluent alternatives and the Confluent to Kafka migration guide for the Kafka side.
- From Amazon MSK or MSK Connect
- see the Amazon MSK comparison; the connectors can stay on MSK Connect while OSStend owns the Debezium operations.
- From a Debezium you run yourself
- the review starts with your offsets, schema history, log retention and upgrade state, and ends with a plan you can read.
Debezium also feeds AI pipelines: keeping embeddings fresh with Kafka and CDC describes the pattern. The databases underneath are on the managed PostgreSQL, MySQL and MongoDB pages.
Proof from AceMQ's operating record
The CDC design and recovery work is done by AceMQ's engineers. AceMQ operates OSStend with 160+ customers, 10+ years of running open-source infrastructure and engineers on call 24/7, as acemq.com states.
IBM MQ to RabbitMQ, without replacing the core
American National Insurance moved its Windows-based IBM MQ queues to RabbitMQ on Kubernetes first, with both brokers live until behaviour was validated, while IBM MQ licensing costs were progressively removed. Use case on acemq.com.
AceMQ engagementA Kafka CDC pipeline it did not need
American National replaced a Kafka-based change data capture pipeline for SQL Server with standalone Debezium publishing to RabbitMQ, removing Kafka clusters and Kafka Connect the use case did not need. Use case on acemq.com.
AceMQ engagementPricing for managed Debezium
Managed Debezium is priced from four drivers: the sources and connectors, the change volume, the SLA tier and where it runs. An engineer prices the design after a workload review. There is no free tier, card sign-up or per-connector-hour meter.
- Sources and connectors
- the number of databases, tables and connectors, and whether a snapshot or re-snapshot is in scope.
- Cluster size
- the number of source databases and connectors, change volume, retention and the Kafka or sink that carries the events.
- SLA tier
- fully managed or co-managed, and the uptime or response-time commitment attached.
- Where it runs
- OSStend's cloud, your cloud account or your data center.
Give the source databases and engines, the rough change volume, the sink and where it must run, and the engineer prices that design.
The pricing page shows the model that applies across every engine.
When to contact OSStend, and when not to
Contact OSStend when a Debezium pipeline needs an owner, or has to run in your own account or data center. If a plan or a one-time migration tool does the job, use that.
Contact us if
- ✓You chose Debezium and want engineers on call for the connectors, offsets, snapshots and upgrades.
- ✓Lag, skipped log positions or failed recoveries are happening, or the last upgrade was put off.
- ✓The pipeline must run in your own cloud account or your data center.
- ✓You want Debezium, Kafka and the source database under one contract and one on-call team.
Not a fit if
- ✕You only need a one-time database migration. A migration service or a dump and restore is simpler.
- ✕A managed connector on a plan already holds your pipeline and you are happy with it.
- ✕You want a free tier, a self-serve console or card sign-up. For CDC, OSStend begins with a conversation with an engineer.
- ✕You run Debezium yourself and only want advice: speak to AceMQ.
Questions about managed Debezium
These are the questions engineers and platform teams ask before handing change data capture to a managed service.
What is managed Debezium?
Debezium is open-source software that reads a database's change log and publishes each change as an event. Managed Debezium is that software run for you: connector health, lag alerts, offset and schema-history backup, snapshots, upgrades and recovery, by engineers on call 24/7.
Is OSStend affiliated with the Debezium project?
No. OSStend is not affiliated with or endorsed by the Debezium project or its owner. Debezium is open-source software under the Apache License 2.0, and OSStend runs it for customers. Trademarks belong to their owners.
Do I need Kafka to use Debezium?
No. Debezium Server is a standalone application that streams changes to systems such as RabbitMQ, Amazon Kinesis and Google Cloud Pub/Sub without Kafka (Debezium Server documentation, checked 18 Nov 2026). Kafka Connect is the other common runtime, and the better fit when many consumers read the same changes.
Which databases can OSStend run Debezium against?
SQL Server, Oracle, PostgreSQL, MySQL, MongoDB and Db2, among the sources Debezium lists. The source, its version and its log settings are confirmed in the design review.
Can it run in my AWS account, or on MSK Connect?
Yes. OSStend runs Debezium inside your cloud account, on Kafka Connect or on Amazon MSK Connect, and owns the Debezium operations. AWS runs the MSK Connect infrastructure.
What happens if the connector is down longer than the source keeps its logs?
Changes made in the gap can no longer be read from the log, so the affected tables need a re-snapshot. We size log retention for the longest outage the design allows, alert on the gap, and keep a re-snapshot runbook.
How are Debezium and Kafka upgrades handled?
As one plan: a review of versions and settings, a rehearsal in a non-production environment, a rollback path and a change window on your calendar.
How is this different from AWS DMS?
Both read database logs. DMS is an AWS service you configure as tasks; Debezium is open source, has more source and sink options through Kafka Connect and Debezium Server, and leaves the operations to you or to a provider such as OSStend. See Debezium vs AWS DMS.
What does managed Debezium cost?
The price follows the sources and connectors, the change volume, the SLA tier and where it runs. The pricing page sets out the model, and an engineer prices the design after a workload review.
Trademarks and affiliation
Debezium is a trademark of its respective owner, and Apache Kafka is a trademark of the Apache Software Foundation. OSStend is not affiliated with or endorsed by these projects, and the project names are used only to say which software we run.
Related pages
- GuideChange data capture (CDC): methods and toolsChange data capture explained: log-based vs query-based CDC, the main CDC tools, Postgres CDC and replication slots, and when to pay someone to run it.
- KafkaManaged Kafka in your cloud or data centerManaged Kafka with the broker count, partition plan, KRaft, Connect and schema registry designed for your estate, run 24/7 in your cloud or your data center.
- AlternativeAmazon MSK pricing and alternatives in 2026Amazon MSK pricing is only part of the picture. See the documented broker and partition limits, and an alternative: Apache Kafka sized to your workload.
- ComparisonDebezium vs AWS DMS: change data capture comparedDebezium vs AWS DMS for change data capture: sources, targets, operations, log handling and when each fits. Read from both vendors' documentation, 18 Nov 2026.
- AI data · Fresh contextEmbeddings freshness with Kafka and Postgres CDCPostgres CDC with Debezium and Kafka Connect keeps embeddings current: re-embed on change, handle deletes, and stop agents answering from stale data.
- SecuritySecurity and compliance at OSStendOSStend security: private networking, your identity provider, your key management and audit trail, plus air-gapped delivery and FIPS hosts you control.
Talk to an engineer about your CDC pipeline
45 minutes with an engineer: bring your source databases, your change volume and where the events go, and the engineer sketches the design and the runbooks.
- ✓A design sketch for your workload, not a plan menu
- ✓A zero-downtime cutover path for what you run today
- ✓A price for that design, after the review