Alert retries

Every channel retries a failed delivery, since a webhook or Discord can be down as well as a mail server. Retrying is the delivery queue's job: each run of the alerts job sends what is due, so a failure is never retried twice over.

alerts({
  channels: [mail({ transport, from }), webhook({ retry: { maxAge: "1h" } })],
  retry: { attempts: 5, backoff: "exponential", maxAge: "24h" },
});

The config only needs what differs from the defaults. A channel's own retry overrides the plugin's, key by key, so webhook({ retry: { maxAge: "1h" } }) above keeps 5 exponential attempts but gives up after an hour.

retry: { maxAge: "24 hours" } is a type error: a Duration is a number followed by s, m, h or d.

Choosing valuesCopy Anchor Link

• Chat channels such as Discord: a short maxAge, like "1h", since a late alert there is noise.
• Webhooks that feed another system: more attempts and a longer maxAge, so a receiver that is down for a while still gets every alert.
• Mail: the defaults. A wrong password or an unverified domain fails every try; POST .../targets/:name/test shows the provider's answer at once.

Watching failuresCopy Anchor Link

• A target whose last delivery failed has state: "failing" and the error in stateReason.
• GET /v2/projects/:project/alerts/deliveries?status=pending lists deliveries waiting for a retry with their attempts, lastError and nextAttemptAt; status=failed lists the ones that ran out.
• GET /v2/admin/alerts/status counts pending deliveries and names the failing targets across projects.
• The cleanup job removes sent deliveries after 30 days and failed ones after 90.