Uptime monitoring for WordPress agencies: what to check and who to wake up
Michal
Author
A client's site going down is not the worst thing that can happen to an agency. The worst thing is the client telling you about it. Uptime monitoring exists to make sure that never happens, and most agencies either do not have it, or have it set up in a way that wakes someone up for nothing until it gets muted.
This post covers what to monitor on a WordPress site, how often, how to stop false alarms, where alerts should go, and what you can honestly promise a client.
What "down" means
A site can be up for the monitor and down for everyone else. Decide what counts before choosing a tool:
- HTTP status. A 5xx is down. A timeout is down. A 403 from a WAF that is blocking your monitor is not, which is why the monitor should identify itself and be allow-listed.
- Response time. A site that answers in twelve seconds is up by most definitions and unusable by any. Track response time on every check and alert on a sustained slowdown, not just on failure.
- Content. A 200 that serves "Error establishing a database connection" or a blank page is down. Check for a known string on the page, or at least for the WordPress markup you expect.
- SSL. A certificate that expires tonight will take the site down at midnight. Read the expiry date on each check and alert at 14 days and again at 3.
- Domain. Rarer, and more total: an expired domain takes down the site, the email and the client's reputation. A WHOIS check once a week with a 30-day warning.
- Cron. WP-Cron on a low-traffic site is a monitor in its own right. Scheduled posts not publishing and backups not running are an outage the client will discover later.
How often to check
Every minute. Five-minute checks were a reasonable compromise when monitoring was expensive; now the cost of a check is nothing and the cost of five minutes of unnoticed downtime on a shop is real. One-minute checks also give you useful response-time data: a per-minute series shows the slow patch between 02:00 and 03:00 when the backup runs, where a five-minute series shows noise.
The exception is sites behind aggressive rate limiting, where a monitor hitting every minute from one address can itself get blocked. The fix is allow-listing, not slower checks.
Stopping false alarms
Every monitoring setup that gets muted got muted because of false alarms. Three things prevent them:
- Confirm from a second location. A failed check from one network is a question, not an answer. Before alerting, re-check from two or three other places in the world. Services like Globalping make this a single API call. If only one location fails, the problem is between that location and the site, not the site.
- Require two consecutive failures. A single timeout during a deploy or a cache purge is normal. Two in a row, confirmed from elsewhere, is an incident.
- Maintenance windows. If updates run at 02:00 on Tuesdays, the monitor should know that. Suppress alerts in the window but keep recording, so the uptime number stays honest.
Where alerts go
The alert that goes to an inbox with four hundred unread emails did not happen. Route by severity:
- Site down, confirmed: Slack or a webhook into whatever on-call tool you use, immediately. Email as well, for the record.
- SSL expiring, domain expiring, slow response: email or a daily digest. These are "this week" problems.
- Recovered: the same channel as the outage, so the thread closes itself.
Group alerts by client, not by site. "Three of Northgate's sites are down" is one message and one phone call; three separate alerts are a Slack channel nobody reads. And decide, in writing, who gets woken up for which clients. A monitor with no on-call rota is a monitor that alerts whoever happens to be awake.
Status pages
A public status page per site, or per client, does two things. It gives the client somewhere to look instead of calling you, and it gives you a place to say "we know, here is what is happening". It also, quietly, shows the client that someone is watching their site all day, which is the thing they are paying for and cannot otherwise see. An uptime badge on your own site, pulling from the same data, does the same job for prospects.
What to promise
Uptime percentages sound precise and mostly are not. The arithmetic for a 30-day month:
| Uptime | Downtime allowed per month |
|---|---|
| 99% | 7 hours 12 minutes |
| 99.5% | 3 hours 36 minutes |
| 99.9% | 43 minutes |
| 99.95% | 21 minutes |
| 99.99% | 4 minutes |
Shared hosting cannot do 99.99% and neither can you on top of it, whatever the host's own page says. 99.9% is a reasonable promise for a managed host with a CDN in front; 99.5% is honest for cheaper hosting. Whatever you promise, measure it with your own monitor rather than the host's, report it monthly, and count the maintenance windows in the number.
After an outage, write three lines: what broke, how long it took to notice, and what changes. The second number is the one monitoring improves. If the client told you first, the monitoring failed, regardless of why the site did.
The short version
- Check every minute: status, response time, content, SSL expiry.
- Confirm from other locations before alerting, and require two failures.
- Respect maintenance windows but keep recording.
- Route by severity: Slack for outages, email for everything else, grouped by client.
- Give each client a status page.
- Promise 99.9% on good hosting, measure it yourself, report it monthly.
WP Admin Online does most of this for the sites you connect: uptime checks at the interval you choose, down to one minute, confirmed with Globalping before an alert goes out, response-time percentiles, SSL and domain expiry, cron health, email, Slack and webhook alerts, and a public status page per site. It does not pause checks for maintenance windows. SSL and domain alerts are on the free plan; uptime monitoring is part of every paid plan.