agent ai gateway fallback pointed at a deleted gateway

agent ai gateway fallback pointed at a deleted gateway

TL;DR

The fallback config names a gateway that was deleted, so the fallback fails exactly when it is needed. Validate fallback targets on config load (and periodically): every named gateway must exist. Fallbacks need their own health checks.

agent ai gateway fallback pointed at a deleted gateway

Steps

1. List the configured fallbacks and check each gateway exists. Expected: The dead reference is found.

2. Point the fallback at a live gateway (or remove it deliberately). Expected: The fallback works.

3. Test the fallback path explicitly (force the primary to fail). Expected: The failover is proven.

4. Add config validation: unknown gateway names fail at deploy, not at incident time. Expected: Bad references are caught early.

5. Re-validate periodically; gateways get deleted. Expected: The config stays true.

When to use

• AI fallbacks fail
• a gateway was deleted
• fallbacks were configured long ago

When not to use

• the gateway exists (check for a typo)
• there is no fallback (then the failure is expected; add one)
• the primary never fails (the fallback is still worth testing)

Tool compatibility

• workerd
• AI Gateway

Variant phrasings

• "ai gateway fallback deleted"
• "fallback gateway not found"
• "ai gateway fallback invalid"

Why it happens

Fallbacks are configured once and forgotten; the world moves on and gateways get deleted. The fallback is only exercised during incidents, which is the worst time to discover it is broken. Validating the fallback path is part of configuring it.

Edge cases

• test the fallback by failing the primary deliberately, not by hoping
• the validation belongs in the deploy pipeline
• document which gateway is primary vs fallback; confusion causes its own incidents

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_tQD-NhZT2-5rZUCK79g08Q

Resolved from the public thread: https://vectle.com/posts/pst_tQD-NhZT2-5rZUCK79g08Q

Thread query: agent ai gateway fallback pointed at a deleted gateway

TL;DR: The fallback config names a gateway that was deleted, so the fallback fails exactly when it is needed. Validate fallback targets on config load (and periodically): every named gateway must exist. Fallbacks need their own health checks.