# Leaving Aurora without losing automatic failover

DevFeed: [Leaving Aurora without losing automatic failover](<https://devfeed.tech/articles/leaving-aurora-without-losing-automatic-failover-141007.md>)

Original publisher: [Read original article](<https://dev.37signals.com/leaving-aurora-without-losing-automatic-failover/>)

Author: Paul S.

Published: 2026-10-20T17:00:00Z

Content type: article

Language: en

Sources: [37signals Dev](<https://devfeed.tech/sources/37signals-dev.md>)

Topics: [Amazon Aurora](<https://devfeed.tech/topics/amazon-aurora.md>), [reliability-engineering](<https://devfeed.tech/topics/reliability-engineering.md>)

Tags: [2022](<https://devfeed.tech/tags/2022.md>), [2023](<https://devfeed.tech/tags/2023.md>), [2025](<https://devfeed.tech/tags/2025.md>), [amazon-aurora](<https://devfeed.tech/tags/amazon-aurora.md>), [amsterdam](<https://devfeed.tech/tags/amsterdam.md>), [apps](<https://devfeed.tech/tags/apps.md>), [asynchronous](<https://devfeed.tech/tags/asynchronous.md>), [automated](<https://devfeed.tech/tags/automated.md>), [automatic](<https://devfeed.tech/tags/automatic.md>), [aws](<https://devfeed.tech/tags/aws.md>), [backups](<https://devfeed.tech/tags/backups.md>), [bugs](<https://devfeed.tech/tags/bugs.md>), [consul](<https://devfeed.tech/tags/consul.md>), [cost](<https://devfeed.tech/tags/cost.md>), [critical](<https://devfeed.tech/tags/critical.md>), [cron](<https://devfeed.tech/tags/cron.md>), [dash](<https://devfeed.tech/tags/dash.md>), [data-centers](<https://devfeed.tech/tags/data-centers.md>), [failover](<https://devfeed.tech/tags/failover.md>), [leaving](<https://devfeed.tech/tags/leaving.md>)

## AI overview

37signals replaced Aurora while retaining automatic MySQL failover with Orchestrator for topology and promotion, Consul for the primary record, and existing F5 load balancers health-checking small nginx endpoints on each database host. Apps keep fixed read and write addresses; after promotion, Consul updates the primary, host endpoints report healthy roles, and F5 routes reconnections. The system uses asynchronous GTID replication, limits automatic failovers to the primary's data center, and requires human intervention for cross-data-center moves. The team reports that load tests resumed writes 12 seconds after killing a primary. Consul elections briefly removed databases from F5 pools, and incomplete topology visibility once required manual repointing, prompting fallback behavior, alerts and runbooks. Aurora migration cut costs substantially; the account also reports faster HEY requests and queries, though failover still carries the risk of losing transactions not yet replicated.

## Source excerpt

How we replaced Amazon Aurora in our data centers without losing automated failovers.