CDC-patterns
Parent: mongodb-kafka-connector · Published reference · snapshot 2026-09-08 · skill mongodb-operations-expert/references/mongodb-kafka-connector.md
↓ Facts as markdownall context files
The MongoDB Connector for Apache Kafka is a Kafka Connect plugin that bridges MongoDB and Kafka in both directions:
These notes link each claim to its source. Some citations identify archived research reports that are no longer hosted on this site. A published reference means the content is available; it does not certify independent review or accuracy.Read the editorial policy and follow the sources before relying on a claim.
Overview
- The MongoDB Connector for Apache Kafka is a Kafka Connect plugin that bridges MongoDB and Kafka in both directions: [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#overview]
- Source connector: MongoDB change streams → Kafka topics (CDC pipeline) [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#overview]
- Sink connector: Kafka topics → MongoDB collections (event consumer) [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#overview]
- Supports Confluent Platform, Confluent Cloud, Amazon MSK, and self-managed Kafka. [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#overview]
Basic Change Stream Source
- This publishes change events to topic mongo.mydb.orders (format: <prefix>.<db>.<collection>). [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#basic-change-stream-source]
Resume Token Persistence
- The connector automatically persists the resume token in a Kafka Connect offsets topic. On restart, it resumes from the saved token. [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#resume-token-persistence]
- If the token expires (oplog window exceeded during connector downtime): [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#resume-token-persistence]
Dead Letter Queue (DLQ)
- Configure DLQ to route failed messages instead of stopping the connector: [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#dead-letter-queue-dlq]
- DLQ messages include headers with error context. Process DLQ messages with a separate consumer for alerting or manual replay. [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#dead-letter-queue-dlq]
Source Connector Errors
- ChangeStreamHistoryLost (error code 286): [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#source-connector-errors]
- The oplog has been truncated past the resume token. Resolution: [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#source-connector-errors]
- Set startup.mode=timestamp to start from a recent time [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#source-connector-errors]
- Or re-snapshot with startup.mode=copy_existing [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#source-connector-errors]
- Increase oplog size to prevent future occurrences [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#source-connector-errors]
- InvalidResumeToken: Resume token is corrupted or from an incompatible MongoDB version. Resolution: clear stored offset and restart connector. [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#source-connector-errors]
Sink Connector Errors
- DuplicateKey (11000): Configure ReplaceOneDefaultStrategy instead of InsertOneDefaultStrategy to make sink idempotent. [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#sink-connector-errors]
- DocumentValidationFailure (121): Kafka messages don't match MongoDB $jsonSchema validator. Check message schema vs collection validator. [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#sink-connector-errors]
Sink Connector
- Worker parallelism: Set tasks.max equal to the number of Kafka partitions for the topic. [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#sink-connector]
CDC Pipeline Pattern: MongoDB → Kafka → Downstream
- For near-real-time with low latency: [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#cdc-pipeline-pattern-mongodb-kafka-downstream]
- Use poll.await.time.ms: 100 (shorter poll interval) [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#cdc-pipeline-pattern-mongodb-kafka-downstream]
- Monitor consumer lag on the Kafka topic [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#cdc-pipeline-pattern-mongodb-kafka-downstream]
- Keep poll.max.batch.size small (100-500) for lower latency at cost of throughput [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#cdc-pipeline-pattern-mongodb-kafka-downstream]
Anti-Patterns
- Single-partition topics with multiple sink tasks: Multiple sink tasks on a single partition = contention; match tasks.max to partition count [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#anti-patterns]
- Not configuring DLQ: Connector stops on first unprocessable message; always configure DLQ in production [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#anti-patterns]
- InsertOneDefaultStrategy for idempotent pipelines: Insert fails on duplicate; use ReplaceOneDefaultStrategy or BulkWriteStrategy for idempotent sinks [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#anti-patterns]
- Monitoring consumer lag but not resume token age: Consumer lag tells you about Kafka backlog; resume token age tells you about oplog risk (if token becomes invalid = full resync) [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#anti-patterns]
- Not increasing oplog for high-volume CDC: Connector outage exceeding the oplog window = full resync required; size oplog to cover expected maintenance windows [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#anti-patterns]
References
- MongoDB Kafka Connector Documentation [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#references]
- Source Connector Configuration [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#references]
- Sink Connector Configuration [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#references]
- Write Model Strategies [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#references]
- Kafka Connector GitHub [archived source: https://llms-explorer.com/sources/mdb-context-hub/mongodb-kafka-connector/#references]
Where this helps
- Streaming MongoDB changes to downstream systems in near-real-time via Kafka, a CDC pipeline, rather than polling the database on an interval. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- A Kafka Connect worker restarts and resume-token persistence needs to be understood so the source connector picks back up from where it left off instead of replaying or dropping events. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- A sink connector is failing on duplicate-key errors and the right write strategy, ReplaceOneDefaultStrategy versus InsertOneDefaultStrategy, needs to be chosen to make it idempotent. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- Sizing sink connector parallelism and matching tasks.max to the number of Kafka partitions to avoid contention. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
Project ideas
- Stand up a MongoDB change-stream source connector publishing to a topic named by the prefix.db.collection convention, then trace a document update through to the Kafka topic. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- Configure a dead-letter queue on a sink connector and deliberately send a malformed message to confirm it routes to the DLQ with error-context headers instead of stopping the connector. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- Build a resume-token recovery path that detects a ChangeStreamHistoryLost condition and falls back to startup.mode=timestamp to restart from a recent point rather than failing permanently. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- Tune a low-latency source pipeline by lowering poll.await.time.ms and monitoring consumer lag on the downstream topic to confirm the change actually reduces end-to-end latency. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
Common mistakes
- Running multiple sink tasks against a single-partition topic, which creates contention instead of parallelism; tasks.max should match the partition count. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- Not configuring a dead-letter queue, so the connector stops entirely on the first unprocessable message instead of routing it aside and continuing. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- Using InsertOneDefaultStrategy for a pipeline that needs idempotent replays, which fails on duplicate keys instead of the intended replace or upsert behavior. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- Letting the oplog window get exceeded during connector downtime, which invalidates the persisted resume token and forces a restart from a timestamp rather than a clean resume. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
Known issues
- A resume token can expire if the connector is down longer than the oplog retention window, forcing a startup.mode=timestamp restart and risking missed or re-processed events around the gap. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- ChangeStreamHistoryLost, error code 286, means the oplog has already been truncated past the resume token; by the time this error surfaces, the original position is unrecoverable and the fix is inherently lossy. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- DocumentValidationFailure, error 121, on the sink side means Kafka messages don't match the target collection's $jsonSchema validator; schema drift between producer and consumer isn't caught until write time. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
- Sink connector idempotency depends entirely on choosing the right write strategy per use case; the default strategy isn't automatically safe for every workload, so this has to be a deliberate configuration choice. [source] — AI-suggested, synthesized from this pack's existing facts — not extracted from a source document.
Children
- outbox-pattern (frontier, viability 4.20 / 5)
- exactly-once-semantics (frontier, viability 4.22 / 5)
- debezium-handlers (frontier, viability 3.73 / 5)
Frontier under this node: debezium-handlers (3.73), exactly-once-semantics (4.22), outbox-pattern (4.20)