Get an alert when a website is down or slow
Most sites don't go down with a bang. The server answers with an error page for twenty minutes after a deploy, the database behind it stops and every page turns into a 500, or everything still works but takes eight seconds. Someone notices in the end; the question is whether it is you or a customer. A check from outside, every few minutes, tells you first, and keeps a record of how the site has been doing.
Why you hear about it late
- The server is up, the site is not. Hosting dashboards watch the machine. A full disk, a stopped database or a bad release leave the machine healthy and the pages broken. A
503 Service Unavailablemeans the server is not ready to handle the request and is meant for temporary conditions; a temporary condition that nobody sees can last all night. - The page loads, but it is the wrong page. A maintenance notice, an empty product list or a login wall served with status 200 looks fine to anything that only reads the status code.
- It got slow, not broken. web.dev gives 0.8 seconds or less as a good time to first byte and more than 1.8 seconds as poor, as a rough guide. A site that drifts from 200 ms to 3 seconds still answers, so nothing fails, and visitors leave.
- Search engines notice too. Google documents that 5xx and 429 answers make its crawlers slow down, and that URLs which persistently return a server error are removed from its index. An outage you fix in an hour costs an hour; one you find after a weekend can cost more than that.
A check from outside, every few minutes
A TaskBeacon website check requests your address from our server every 1, 5, 15 or 60 minutes. There is nothing to install and nothing to add to the site: on Jobs, open the arrow next to + New job, choose Website or API check and enter the address. It works for a page, a health endpoint or an API, yours or someone else's that you depend on. You get:
- Failed when there is no answer within the timeout (10 seconds unless you change it), on a connection or certificate error, on a status outside the expected ones (by default anything other than 2xx or 3xx after redirects), or when the page lacks a word you said it must contain. A failure is checked once more about a minute later, and only a second failure alerts, so one dropped connection doesn't wake you.
- Odd data when the site answers far slower than usual twice in a row: at least three times its usual time and at least one second more.
- Back to OK with the next good answer, so you also know when it ended.
The alert says what went wrong, for example HTTP 503 Service Unavailable, and goes where your other alerts go: email, Slack and the other channels in Alerts. The exact rules are in When you get an alert.
Response time and uptime over time
Each check stores the status code and the response time, measured as the time until the answer starts, including DNS, connection and TLS. The job page draws the response time of the recent checks with a line for the usual time, lists every check with its code and time, and shows the share of checks that succeeded in the last 30 days, such as “100% up · 964 checks”. A slow hour last Tuesday or a site that is quietly getting slower shows up there before anyone complains.
How far back the single checks go depends on the plan: the last 100 checks on Free, more on the paid plans. See Pricing. For an https:// address the same check also reads the certificate's expiry date; that part is described in SSL certificate and domain expiry alerts.
The site and the scripts behind it, in one place
A site is rarely only a web server. There is a nightly import that fills the prices, a backup, a job that sends the orders on. When the page shows yesterday's numbers the site is “up” and still wrong. In TaskBeacon the website check sits in the same project as those jobs: the import pings when it finishes and reports its row count, the backup reports its size, and a script that never started shows up as missed. One list tells you whether the site answers and whether what feeds it ran.
What it is not
- One location. We check from one server, which is why a failure is confirmed a minute later before it alerts. If you need to know that the site is reachable from five continents, use a multi-region monitoring service.
- Not a browser. The response time is the server's answer, not the time until a page with its images and scripts is ready in a visitor's browser.
- Public addresses only. Addresses that lead to a private or local network are refused. For an intranet site, run a small scheduled script on your side that does the request and pings TaskBeacon, as in heartbeat monitoring.
Free includes 3 website checks, every 5 minutes or less often; 1-minute checks are part of Pro, open to everyone for now.
Without TaskBeacon
Schedule a small script on a machine that is not the web server: request the page with curl --fail, which exits with code 22 when the answer is an HTTP error, and send yourself an email when the exit code is not 0. Add a text search on the downloaded page if a wrong page with status 200 worries you. It works, with two gaps to close yourself: the script needs something to notice when it stops running, and it tells you about now, not about last week, unless you also write each result to a file and look at it.
Connected Claude or ChatGPT to TaskBeacon? Ask it "what failed last night, and why?" and it answers from your jobs. How that works.
Related
In the Docs: Website and API checks · When you get an alert · Alerts.
- Get an alert before an SSL certificate or domain expiresShorter certificates and no more reminder emails from Let's Encrypt.
- Heartbeat monitoring (a dead man's switch) for scheduled scriptsThe job checks in on every run; silence is the alert.
- Check that last night's backup exists and isn't emptyThe backup ran; the file it made is 0 KB.
TaskBeacon is free to start, with every feature open for now and no credit card. Create a job, paste the code, and the first ping shows up on the job page.
Start free View live demoSources
- Google Search Central: How HTTP status codes affect Google's crawlers (undated; checked 2026-11-16)
- MDN: 503 Service Unavailable (2026-06-22)
- web.dev: Time to First Byte (TTFB) (2021-10-26, updated 2025-11-18)
- curl man page: -f, --fail (curl 8.23.1, undated; checked 2026-11-16)
Page checked against its sources on 2026-11-16. Menus and limits of other products change; if something here no longer matches, write to support@taskbeacon.io.