---
title: "Failover"
description: "2 posts about Failover, summarised, each linking to the original."
---

# Failover
> 2 posts about Failover, summarised, each linking to the original.

## Articles

### [Self-service regional failover with Amazon Application Recovery Controller Region Switch and centralized ingress controls](https://yomu.fyi/post/self-service-regional-failover-with-amazon-application-recovery-contro.md)
- Company: [AWS](https://yomu.fyi/company/aws.md)
- Author: Greg Willson
- Published: Nov 10, 2026
- Type: Problem & solution

Centralized ingress proxy fleets enforce security in regulated industries but create coordination overhead that conflicts with application team autonomy for disaster recovery failover. Amazon Application Recovery Controller (ARC) Region Switch, launched in August 2025, provides a fully managed failover orchestrator that addresses configuration drift, slow detection, and recovery logic drift through automatic plan evaluation and codified workflows. The solution uses centrally managed AWS Transit Gateways with application-team-owned Route 53 Private Hosted Zones, establishing clear boundaries where teams control DNS records without needing to coordinate with infrastructure teams. Application teams create Region Switch plans with execution blocks for database promotion, compute scaling, and traffic redirection, and can trigger them automatically from CloudWatch alarms or manually with approval steps. A central parent plan can run each team’s child plan in dependency order, enabling coordinated portfolio-level failover while preserving team ownership of recovery logic and maintaining security posture.


### [Best Practices for Conversation Relay Provider Failover](https://yomu.fyi/post/best-practices-for-conversation-relay-provider-failover.md)
- Company: [Twilio](https://yomu.fyi/company/twilio.md)
- Author: Daniel Jansson
- Published: Nov 4, 2026
- Type: Tutorial

Twilio Conversation Relay uses separate third-party providers for transcription (ASR) and text-to-speech (TTS), each of which can fail under high load or during outages. To maintain uninterrupted Voice AI calls, developers must implement an application-level failover strategy that switches providers when errors occur. Failures manifest differently depending on timing: startup failures end the session with an ErrorCode, while mid-call TTS failures keep the call alive and require the WebSocket server to detect the error and send an end message with handoffData. The action URL handler must branch on specific error codes (64114 for ASR, 64111/64112 for TTS) and return new TwiML with alternative providers. To prevent infinite loops, the strategy limits retry attempts, changes providers on each attempt, and includes a final fallback that does not use Conversation Relay, such as or .
