---
title: "DynamoDB connector support in Forward"
excerpt: "DynamoDB connector support in Forward, Kafka pause visibility, and other improvements and bug fixes"
date: "2026-06-19"
---

# DynamoDB connector support in Forward

Forward now supports the [DynamoDB Connector](/forward/ingest-data/connectors/dynamodb). You can use it to mirror an Amazon DynamoDB table into a Tinybird Data Source with an initial Point-in-Time Recovery export to S3, followed by Change Data Capture from DynamoDB Streams.

DynamoDB connectors in Forward need to be defined in a `.connection` file containing the IAM access policy required to access the underlying resources. To simplify setup, the Forward CLI includes a guided wizard that walks the user through the configuration process: `tb connection create dynamodb`. DynamoDB connections are also fully supported in our TypeScript and Python SDKs.

If you are already using DynamoDB connections and want to migrate from Classic to Forward, see the updated [Migrate from Classic](/forward/guides/migrate-from-classic) guide. The migration can be performed using the `tb —cloud pull` and `tb migrate-to-forward` commands.

## Kafka ingestion pause visibility

Kafka Data Sources now show an ingestion-paused state in the UI when the Connector Circuit Breaker is open. The Data Source page includes a paused ingestion banner, links to logs, an inspect-error action, and a `tb deploy` hint for manual retry.

When ingestion is paused by the Circuit Breaker, Tinybird now retries automatically every 12 hours. If a retry succeeds, ingestion resumes without manual intervention. You can still run `tb deploy` to retry sooner after fixing the underlying connection issue. See [Kafka logs](/forward/ingest-data/connectors/kafka#kafka-logs) for details.

## Improvements and bug fixes

- [Forward]: It is now possible to only change the connection used by a Kafka Datasource. Previously, we incorrectly detected "no changes".
- [Forward]: Replacing all columns in a Data Source now forces the schema-change and backfill path instead of trying an invalid `ALTER` that drops every existing column.
- [All]: Playgrounds now debounce table descriptions, reuse known Data Source metadata, and cache successful `DESCRIBE` results to avoid repeated network calls while typing.
- [All]: Deleting the Workspace Admin token is no longer possible.
- [All]: Cloud now keeps the header and sidebar visible when Workspace access is blocked because an Organization has unpaid invoices. Organization admins see a billing link, and other Workspace members see how to contact their Organization admins.
- [All]: Dedicated Organization observability no longer shows CPU and memory charts on the Observability page. Those metrics now live on the Cluster page, and the Observability page links there.
- [All]: NDJSON URL imports can now be tuned per Workspace for `max_threads`, `max_insert_threads`, and `max_memory_usage`.
- [All]: HFI NDJSON ingestion now skips the sampling path when the sampling budget is exhausted, avoiding unnecessary JSON parsing on hot Data Sources.