What Is an Uptime Service? How Monitoring Works
Key takeaways
- An uptime service bundles three parts: scheduled probes, alert routing to on-call channels, and a public status page.
- Protocol-level probes (TCP, DNS, database, WebSocket, SSL) catch failures a homepage HTTP ping misses.
- One-minute checks with two-strike confirmation from multiple regions is a solid default to avoid 3 AM false alarms.
- Heartbeats are the only way to detect silent failures like cron jobs that never run or workers that die.
- Vendor-reported uptime measures the probe, not the customer experience, so ask what URL, region, and interval it reflects.
An uptime service runs automated probes against your servers and APIs at fixed intervals, alerts your on-call channel when checks fail, and publishes the results to a public status page your customers can see. If you are paying $29-199/mo for Atlassian Statuspage or $10+/mo for Pingdom, here is how these tools work and how to pick one that matches your team size.
What is an uptime service and how does it actually work?
An uptime service is three things bundled together: monitoring probes that hit your infrastructure on a schedule, alerting that routes failures to your on-call channel, and a public status page your customers can check during an outage. Tools like Atlassian Statuspage (which charges $29-199/mo) charge heavily for that bundle, but the components are simple enough to evaluate piece by piece.
The three moving parts: probes, alerting, status page
Probes are automated checks run from outside your network at a fixed interval, typically 30-60 seconds. Alerting takes probe failures and turns them into PagerDuty calls, Slack messages, or SMS after a threshold is met. The status page publishes the results so customers see incidents without filing support tickets.
What happens during an outage, minute by minute
Minute 0: a probe fails. Minutes 1-3: retries from other locations confirm it's a real outage, not a false positive. Minute 3: your pager fires and the status page flips to "investigating." Minutes 4 onward: you fix and post updates. The same flow applies at microservices scale, which we cover in our guide to distributed reliability.
What does an uptime service monitor beyond simple HTTP pings?
A URL check only tells you that a page returned a 200. It says nothing about whether your API accepts writes, your database answers queries, or your nightly backup actually ran. An uptime service that goes beyond HTTP pings probes the protocols your stack actually depends on.
Protocol-level probes for API-first stacks
| Probe type | What it catches that a URL ping misses |
|---|---|
| TCP port | Service is down even though the load balancer still returns 200 |
| DNS | Record deleted, nameserver misconfigured, or propagation failure |
| Database (e.g. Postgres) | Connection pool exhausted or query hangs |
| WebSocket | Handshake fails while the HTTP endpoint looks healthy |
| SSL certificate | Expiry or chain errors before browsers start rejecting clients |
If your product is an API first, a homepage check is nearly useless. StatusPulse runs these protocol-level probes rather than URL checks alone, and multi-region checks reduce false positives from a single bad network path (see multi-region uptime checks: identify true downtime).
Heartbeats and cron job monitoring
Push-based heartbeats cover the failures pull checks cannot see: a cron job that never runs, a worker that dies silently, a queue that stops draining. If the expected signal does not arrive inside its window, you get an alert. This is the only way to detect "nothing happened" outages.
Why homepage-only checks produce false confidence
A green homepage while your signup endpoint or database is broken is worse than no monitoring: it teaches your team to ignore alerts. Monitor the critical paths customers pay for, not the vanity URL.
How do check intervals and probe locations affect false positives?
Interval math: detection time vs alert noise
A 30-second interval detects an outage in about 30 seconds on average; a 5-minute interval delays your downtime alert by up to 5 minutes. But short intervals raise false alarm rates on flaky networks, because one failed request triggers an alert. The fix is retry logic: confirm a failure with a second check before paging anyone. UptimeRobot's free tier runs checks every 5 minutes, with 30-second checks starting on paid plans (UptimeRobot pricing). If you're monitoring a payment endpoint rather than a marketing page, 1-minute checks with 2-strike confirmation is a reasonable default.
Multi-region confirmation to kill false positives
A single probe location can't distinguish "the site is down" from "this datacenter's route to the site is down." Multi-region confirmation requires probes in two or three locations to agree before an incident fires, which eliminates most false positives at the cost of a few seconds of detection time. StatusPulse runs real protocol-level probes from multiple regions for exactly this reason, and the same logic applies to DNS checks: a regional resolver failure looks like an outage until a second region confirms it (see 7 key DNS record types to monitor for better uptime). The honest tradeoff: more confirmation means slower detection, so tune retries per endpoint criticality instead of using one global rule.
What alerting and escalation features should an uptime service include?
Alert channels and confirmation windows
Require at least email, SMS, Slack, and a webhook, plus a PagerDuty integration if you already pay for it. A single failed probe should never page you: the service must retry from a second location and only alert after 2-3 consecutive failures, or a transient blip becomes a false alarm at 3 AM. Confirmation windows of 30-60 seconds are a reasonable default for a small team that cannot afford alert fatigue.
Escalation policies when you have no on-call rotation
With a solo or two-person team, escalation means: if the founder does not acknowledge within 5 minutes, alert the second channel, then repeat on a schedule until someone wakes up. Look for unacknowledged-alert re-notification and scheduled maintenance windows that suppress alerts during deploys. Anomaly-based alerting helps catch slow degradation that threshold checks miss, which we cover in more depth in our guide to AI anomaly detection and faster incident response.
Honest caveat: StatusPulse covers channels, retries, and webhooks well, but if you need multi-step rotation schedules for a 10-person on-call roster, PagerDuty remains the stronger escalation layer.
How do uptime services integrate with status pages and incident workflows?
Automatic incident creation from failed probes
When a probe fails, the uptime service should open an incident on your status page without you touching anything. You confirm or annotate it, and the page flips from "All systems operational" to a degraded state, timestamped. That matters at 3 AM, when you are fixing the problem and cannot also write customer updates. StatusPulse does this from probe failures; if you are evaluating other tools, check whether incident creation is automatic or a manual button press mid-outage.
Component mapping for partial outages
Your API can be down while your dashboard works fine. Component mapping ties each probe to a specific status page component, so a failed database check marks "API" as degraded instead of taking your whole page red. Verify the service supports per-component subscriber notifications, not just page-wide emails. Hyperping bundles status pages with its monitoring starting at $5/mo (pricing). Where StatusPulse is honest about falling short: subscriber notification customization is thinner than Atlassian Statuspage's, so if you need heavily branded subscriber emails, test that before switching.
For a deeper look at probing protocols that feed these incidents, see monitoring WebSocket uptime beyond the initial handshake.
How do you evaluate uptime services against your team size and stack?
The evaluation matrix by team profile
Score any uptime service on four axes: probe types (HTTP only, or TCP/DNS/SMTP too), check interval (60s minimum, 30s better), escalation (email only, or PagerDuty and on-call rotations), and status page depth (single page vs. component-level subscribers). A solo founder shipping one SaaS usually needs TCP and DNS probes on the database and mail endpoints, a 60-second interval, email plus Slack alerts, and one status page. A 1-10 person team adds PagerDuty escalation and per-component pages so API downtime doesn't read as "everything is down." An agency needs multi-tenant pages and white-labelling more than sub-minute intervals.
Running StatusPulse through the same criteria
StatusPulse covers the probe axis well: protocol-level checks against TCP, DNS, and SMTP rather than homepage pings, which catches the failure modes CronAlert described as monitoring the wrong things. Intervals, escalation, and component-level status pages match the tiny-team profile. Where it falls short honestly: no on-call rotation scheduling, so if you need a full PagerDuty replacement you will run both tools. And with a 5-probe free tier, an agency monitoring dozens of client stacks will outgrow it faster than a solo founder will.
Where the free tier is enough and where it is not
Five probes and one page cover a solo SaaS with one API, one database, and one marketing site. The moment you add a second environment, a worker queue, or a client, you are on Starter at $5/mo, and Pro at $19/mo once escalation matters.
What do uptime SLAs and reported uptime percentages really mean?
The math behind the nines
99.9% uptime allows 43 minutes of downtime per month; 99.99% allows about 4.3 minutes. Most services, including Better Stack's paid plans (which run $25-85/month), quote nines in their SLAs (Better Stack pricing). The number only becomes useful when paired with an error budget: at 99.9%, you can spend those 43 minutes on planned maintenance, failed deploys, or real incidents before you owe credits.
Whose uptime number is it, anyway?
Vendor-reported uptime measures the probe, not your customer. A check hitting your homepage every 60 seconds from one region can report 100% while your checkout endpoint is down for EU users. Ask three questions: what URL is being probed, from where, and at what interval. If the answer is "a homepage ping from one location every 5 minutes," the published percentage is measuring almost nothing your paying customers touch.
How do the main uptime services compare on price and capability?
Atlassian Statuspage is a status page first, not a monitor: it starts at $29/mo and reaches $199/mo, and you typically pair it with a separate monitoring tool. UptimeRobot covers basic HTTP and keyword checks cheaply, but protocol depth is limited compared with tools built around multi-protocol probes. Hyperping and Better Stack bundle monitoring with status pages; Better Stack's monitoring, logging, and status page bundle runs $25-85/mo. Instatus focuses on fast, good-looking status pages and sits adjacent to full monitoring suites.
StatusPulse runs the same comparison honestly: you get protocol-level probes (TCP, DNS, SMTP, not just URL pings) plus a hosted status page from $5/mo Starter or $19/mo Pro, with a free tier of 5 probes and 1 page. Where it falls short: Statuspage's incident subscriber lists, Atlassian ecosystem integrations, and brand recognition are stronger, and Hyperping's probe network is larger. If you need those, pay for them. If you need protocol checks, escalation, and a public page under $20/mo, the trade favours StatusPulse.
FAQ
Is a free uptime monitoring tier enough for a SaaS with paying customers?
Sometimes. Five probes cover one API, one database, and one marketing site, which is enough for a solo SaaS at launch. The moment you add a second environment, a worker queue, or a client, you outgrow it. Paid tiers also buy shorter check intervals and escalation, which matter once customers pay you.
Can I self-host Uptime Kuma instead of paying for an uptime service?
Yes, and it works if you can keep the monitor itself alive. The catch: self-hosting means your probes run from your own infrastructure, so a network or hosting failure can take down both your service and the tool watching it. You also have to build and maintain the public status page and alert routing yourself.
How many probes do I actually need for a small SaaS?
Plan for one probe per critical path: homepage, signup or checkout endpoint, API, database, and mail. That is five, which matches most free tiers exactly. Add heartbeats for cron jobs and workers, plus TCP and DNS checks on infrastructure endpoints, and you land around seven or eight once a second environment exists.
Do I need a separate status page tool if my monitor already sends alerts?
Alerts are for you; a status page is for customers. During an outage, an automatic incident on a public page answers support tickets before they arrive, and component mapping shows that your API is degraded while the dashboard still works. Some tools bundle both; otherwise you are pairing a monitor with Statuspage or similar.
What check interval should I use to avoid false alarms at night?
One-minute checks with two-strike confirmation: require a second failed probe, ideally from another region, before paging anyone. A single failed request on a flaky network should never wake you. Thirty-second intervals are reasonable for payment endpoints; five-minute intervals delay detection too much for anything customers depend on.