Automatic Failover

The Yuki Proxy can automatically reroute your queries straight to Snowflake or BigQuery if the proxy itself ever becomes unhealthy - so a proxy-side issue degrades optimization, not your queries.

How It Works

A lightweight watcher continuously checks the proxy's health. If it detects sustained failures, it switches your traffic to a minimal passthrough path that forwards directly to your warehouse, bypassing the proxy entirely. Once the proxy recovers, traffic switches back automatically - no manual step, no restart, no query downtime.

What to expect during a failover

Your queries keep running. You lose Yuki's optimization for their duration - no cost/performance insights, no query rewriting - but nothing fails outright. Normal operation resumes automatically once the proxy is healthy again.

This is enabled by default for Fully Hosted (SaaS) deployments - there's nothing to configure. For Helm deployments, see below.

Configuring It Yourself (Helm Deployments)

If you manage your own yuki-values.yaml, failover is available as an opt-in addition to the parameters in Helm Deployment:

passthrough:
  enabled: true
  mode: auto # auto | forceProxy | forcePassthrough
  watcher:
    failureThreshold: 3   # consecutive failed checks before switching
    successThreshold: 3   # consecutive healthy checks before switching back
    pollIntervalSeconds: 5

Parameter	Description
passthrough.enabled	Turns the failover mechanism on. Off by default.
passthrough.mode	auto (default) switches based on health. forcePassthrough routes to the warehouse directly regardless of proxy health - useful during planned proxy maintenance. forceProxy disables automatic failover - useful if you need to guarantee traffic never leaves the proxy.
passthrough.watcher.failureThreshold / successThreshold	How many consecutive checks are required before switching. Lower values react faster; higher values avoid switching on brief blips.

Mode changes take effect within one poll interval - no pod restart required.

A note on scope

Failover is a liveness-based safety net: it catches the proxy being down (crashed, restarting, unreachable), not subtler issues where the proxy is running but misbehaving. It's a floor under availability, not a substitute for monitoring.

Next Step

Proceed to Update Connection Strings to route your tools through the proxy.