# Google Cloud Datastream

> Serverless CDC from operational databases into BigQuery

Datastream is Google Cloud's serverless change data capture and replication service, streaming changes from Oracle, MySQL, PostgreSQL, SQL Server, MongoDB, and Spanner into BigQuery and Cloud Storage with minimal latency.

Website: https://cloud.google.com/datastream

## Use it when

- Operational data should land in BigQuery continuously with nothing in between; the direct path with a tunable staleness limit is why the service exists.
- You want zero replication infrastructure: serverless, no instance to size or patch, backfill handled by the same service.
- Sources span MySQL, PostgreSQL (including AlloyDB), Oracle, SQL Server, and Spanner, plus SaaS applications like Salesforce and ServiceNow.
- Billing should follow volume: per GiB processed, flat-rate backfill, 500 GiB free monthly.

## Think twice when

- The destination is not Google: targets are BigQuery, Cloud Storage, and Iceberg, with Dataflow templates bridging elsewhere; outside Google Cloud it is rarely the answer.
- Changes must be transformed in flight; Datastream replicates and nothing more, transformation happens in BigQuery after landing.
- Schema drift includes type changes; new columns and tables replicate, type changes do not.

## How it runs

Fully managed and serverless: configure a source (with redo logs, row binlog, or a publication and slot enabled per engine) and a destination, and the service streams changes with at-least-once delivery and event UUIDs for deduplication.

## Details

### General

| Attribute | Value |
| --- | --- |
| License | Commercial |
| Deployment | Managed, Serverless |
| Workload | Streaming |
| Operational complexity | Low |
| Pricing | Subscription |
| Documentation | https://docs.cloud.google.com/datastream/docs/overview |

### Change Data Capture

| Attribute | Value |
| --- | --- |
| Sources | MySQL, PostgreSQL, Oracle, SQL Server, Spanner; MongoDB in preview |
| Source requirements | Oracle redo logs, MySQL row binlog, Postgres publication and slot |
| What you run | Nothing: serverless; no instance to size or patch |
| Initial snapshot | Backfill handled by the same service |
| End-to-end lag | Seconds, tuned by the BigQuery staleness limit you set |
| Delivery guarantee | At-least-once: event UUIDs for deduplication |
| Schema drift | Handled: new columns and tables replicate; type changes don't |
| Transformations | None: replication only; transform in BigQuery after landing |
| Destinations | BigQuery, Cloud Storage; Iceberg tables |
| Billing unit | Per GiB processed, with backfill flat-rate and 500 GiB free monthly |

## Capabilities

- [Change Data Capture](https://matca.io/capabilities/change-data-capture)

---

Source: https://matca.io/google-cloud-datastream
Last updated: 2026-08-31T11:40:49.024Z