fix(api): report which health probe failed and keep error names in logs#192

Generated by Polylane Autofix from a fix run.

A failing health probe returns an empty negative answer, so a deploy that stops at the gate or an uptime alert leaves nobody able to say which dependency broke. This change records that reason, and gives the same treatment to two failures the service swallows: a subscription that cannot reconcile, and a stalled batch of delivery emails.

Why this fix

The health route is the deploy gate and the uptime probe, and it was the one place a 503 carried no reason at all: the probe now names the dependency and the error's tag on the request span, so a database outage is told from a storage one without a redeploy to find out. A subscription that fails to reconcile logs which account and the error's name, which separates one broken row from a vendor-wide failure after a lost webhook. A delivery-email batch that stops settling logs the error's name, so a code defect is told from an infrastructure fault. Every field is an id, a tag or a count, never a payload, an address or SQL.

View thread View autofix

Generated by Polylane. You can ask follow-ups by mentioning @polylane in a comment.