How to Fix Airbyte CDC Syncs That Stop or Duplicate Rows
Postgres logical replication setup, WAL retention, slot leaks and idempotency traps — the failures that quietly stop a CDC pipeline and how to detect them…
Open-source data pipeline platform with 600+ connectors you can self-host.
Airbyte is an open-source ELT platform that moves data from 600+ sources into your warehouse, and unlike most competitors you can self-host the whole thing for free. Airbyte Cloud starts around $10/mo for small volumes with usage-based credits on top.
Every company eventually needs to copy data from twenty SaaS tools into one warehouse. Custom Python scripts are the default answer, and they break silently when an API changes. Airbyte’s bet is that connectors should be a commodity written by a community.
Connectors exist for Postgres, MySQL, Salesforce, HubSpot, Stripe, Google Ads, Shopify, MongoDB and hundreds of long-tail APIs. Most are maintained by the community, which is both the strength and the weakness.
The Connector Builder is genuinely useful, not a checkbox feature. It has a low-code UI where you declare the request path, pagination and response parser, then test against a live API. Building a working connector took me under an hour on a REST API with cursor pagination.
The software is. The cost is your time. Self-hosted Airbyte needs upgrades, connector image builds, and someone to debug a sync that silently stopped. Airbyte Cloud is consumption-priced with per-row credits, so a high-volume sync can cost more than you expect — check the pricing calculator before moving a 50-million-row table.
Fivetran’s connectors are more polished and it handles schema drift better out of the box — you pay for that, roughly $500+/mo at small volumes. Airbyte is the pick when engineering time is cheaper than the license, or when you need a connector nobody sells. Don’t self-host unless someone actually owns it; abandoned Airbyte installs are a common failure pattern.