# Milvus Source Connector for Debezium: Enabling Event-Driven Vector Architectures

DevFeed: [Milvus Source Connector for Debezium: Enabling Event-Driven Vector Architectures](<https://devfeed.tech/articles/milvus-source-connector-for-debezium-enabling-event-driven-vector-architectures-111256.md>)

Original publisher: [Read original article](<https://debezium.io/blog/2026/11/02/milvus-source-connector-for-debezium/>)

Author: Kartik Angiras

Published: 2026-11-02T00:00:00Z

Content type: article

Language: en

Sources: [Debezium](<https://devfeed.tech/sources/debezium-io.md>)

Topics: [Debezium Connector](<https://devfeed.tech/topics/debezium-connector.md>), [Apache Kafka](<https://devfeed.tech/topics/apache-kafka.md>), [Publish-subscribe pattern](<https://devfeed.tech/topics/pubsub.md>), [event-sourcing](<https://devfeed.tech/topics/event-sourcing.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [hybrid-search](<https://devfeed.tech/topics/hybrid-search.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [debezium](<https://devfeed.tech/tags/debezium.md>), [elasticsearch](<https://devfeed.tech/tags/elasticsearch.md>), [event](<https://devfeed.tech/tags/event.md>), [event-driven](<https://devfeed.tech/tags/event-driven.md>), [kafka](<https://devfeed.tech/tags/kafka.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [postgres](<https://devfeed.tech/tags/postgres.md>), [redis](<https://devfeed.tech/tags/redis.md>), [source](<https://devfeed.tech/tags/source.md>), [stream](<https://devfeed.tech/tags/stream.md>), [vector](<https://devfeed.tech/tags/vector.md>), [vector-database](<https://devfeed.tech/tags/vector-database.md>)

## AI overview

The Debezium 3.7 Milvus source connector reads Milvus 2.5's Kafka write stream, uses etcd checkpoints and Milvus gRPC for snapshots and schemas, and emits inserts and deletes as Debezium events with typed vector fields. Timetick watermarks restore commit order across virtual channels, and Kafka Connect sinks can deliver changes to systems such as Postgres, Elasticsearch, or S3. The connector requires a self-hosted deployment using Kafka as its broker; Milvus Lite, Pulsar, RocksMQ, and Milvus 2.6 are unsupported. It is incubating: its default checkpoint key can miss changes at snapshot handoff, and a crash while events await ordering can lose them. It uses at-least-once snapshot delivery, so idempotent sinks are important; deletes have only primary-key before images, and upserts appear as delete/insert pairs.

## Source excerpt

Google Summer of Code 2026 with the Debezium community was the best learning experience I have had as an engineer. Over one summer I built a Kafka Connect source connector from scratch for a database with no documented way to read its changes. Most of the work was taking the database apart to find out where those changes live, writing the design down before building any of it, and carrying each piece through review with the Debezium maintainers. The connector shipped in Debezium 3.7. What came out of that summer is a Debezium source connector that turns Milvus, the open source vector database, into a first-class source of change events. I built it because Milvus has no change stream of its own. If you need its inserts and deletes somewhere else (Elasticsearch for keyword search, Redis as a cache, Postgres for reporting), you either write to both systems from the application and drift apart on the first partial failure, or you run a batch export that cannot see a delete. The connector reads the write stream Milvus already publishes to its message queue, puts the events back in commit order, and emits each insert and delete as a normal Debezium change event with the vector as a typed field. From there any Kafka Connect sink is a Milvus integration, and the service that writes to Milvus never has to know. This post covers how it works and what you can build once Milvus is one. Read more...