How we detect downtime on your clients’ sites

Glow monitors your clients’ websites itself, from its own infrastructure. This article explains what a check actually does, and how we decide that a site is genuinely down before alerting you.

How often we check

You choose the interval per website, from 3 minutes upwards, under Uptime, Settings. You can also set an interval once and apply it to every site in your account and to any site you add later.

What a check does

We request the site the way a visitor’s browser would, follow redirects, and read the final response.

• A 2xx or 3xx response is up.
• A 4xx or 5xx response is down. The page may still render for a human, so we tell you the exact status we received, for example 500 Internal Server Error, rather than a generic outage.
• No response at all, a DNS failure or a timeout is down, and we say which.
• An expired certificate, or one that does not match the domain, is also down: browsers refuse to load the site, so visitors cannot reach it even though the server answered.

We never alert on a single failed check

A single failure has more than one possible cause, so a second opinion comes from somewhere genuinely different before you hear anything.

When a check fails, we re-check from a different region on a different network. That distinguishes a real outage from a routing problem or a block that only affects one datacentre. Only if that confirms the failure do we record the site as down and alert you.

When a firewall blocks us instead

Some failures are not outages at all. If a site answers with a challenge, typically 401, 403 or 429, the server is up and talking, it is simply refusing us. Re-checking from another random cloud address proves nothing there, so we re-check from our own fixed address instead, which is the one you can allow through a firewall.

If it still refuses, we mark the site as challenged rather than down. You will see a notice in the uptime section explaining that our monitor is being blocked, and you will not be alerted for an outage that is not happening. See our monitor access guide for how to clear it.

Certificates

We check the certificate on every request, and warn you in the uptime section when one is within 30 days of expiring. That is a warning rather than an alert: the site still works, it just will not for much longer. A certificate that has already expired or does not match the domain is treated as downtime, because that is what a visitor experiences.

Certificate monitoring can be switched off per website under Uptime, Settings if a site legitimately runs without a valid certificate.

What you receive

When a site goes down we email the address on the account and, if you have connected it, post to Slack. The message names the reason we recorded, so you can tell a server error apart from a DNS failure without opening anything.

You will get a matching message when the site recovers, with how long the outage lasted.