Question 1

What does 99.9% uptime actually allow? Show the nines.

Accepted Answer

Pure arithmetic on a 30-day month, which holds 43,200 minutes. Two nines - 99% - allows 432 minutes of downtime, about 7.2 hours. Three nines - 99.9% - allows 43.2 minutes. Four nines - 99.99% - allows 4.3 minutes. Five nines - 99.999% - allows about 26 seconds. There is no opinion anywhere in that table; every line is a percentage of 43,200. Which nine you need depends on what a dark minute costs you. A site running paid campaigns bleeds media spend from the first minute, so the practical questions are how fast a failure is detected and how fast the right person knows. That is why revenue-critical endpoints get the 30-second cadence: the 43-minute budget should be spent fixing, not discovering.

Question 2

Will we get flooded with false alarms?

Accepted Answer

No. The Second-Region Rule applies: a failure seen by one probe is re-checked from a second region before any alert fires, which filters out the local network blips that plague single-point monitoring. When your phone buzzes, something is genuinely wrong. Alerting is also tuned by endpoint. The checkout failing pages people immediately; a marketing microsite can wait for morning. The fastest way to make a team ignore alerts is to send ones that do not matter.

Question 3

What does a check actually look at?

Accepted Answer

Three layers. An HTTP check confirms the server answers with a healthy status code. A keyword check confirms the page rendered the content it should - catching the broken template that returns 200 with an empty page. A transaction check scripts a real journey, like submitting the enquiry form and seeing the thank-you page. Most outages that hurt are not full outages. The site is "up" but the form is broken, the payment step errors, or a deploy emptied a template. Layered checks exist for exactly those.

Question 4

Our host already promises 99.9% uptime. Why monitor?

Accepted Answer

The nines table above is the first answer: 99.9% still allows roughly 43 minutes a month, and 99% allows over seven hours. The second answer: an SLA is a refund policy, not a detection system - it pays you back pennies after the damage, and it does not tell you the site is down right now. A host SLA also covers their infrastructure, not your stack. DNS, the SSL certificate, the application, a bad deploy, a third-party script - most real-world failures happen above the layer the hosting SLA measures.

Question 5

How will we find out when something breaks?

Accepted Answer

However you want, in whatever order: email, SMS, WhatsApp and Slack, routed by severity and time of day. If the first person does not acknowledge within the agreed window, the alert escalates to the next name on the ladder automatically. The routing is agreed up front - who gets woken for checkout failures, what waits until morning - so alerts arrive where they will be acted on, not in a mailbox nobody reads at 2 a.m.

Question 6

Do you fix the outage, or just report it?

Accepted Answer

Detection and first response are ours: confirm the failure, classify it, trigger the runbook and notify the right people with what we already know - which regions, which endpoints, since when. That diagnosis usually saves the first half hour of the fix. Whether we make the fix depends on scope. For sites we built or maintain, our team takes it end to end. Where another team owns the hosting, we hand them a precise incident report rather than a vague "site down" - and we stay on it until the recovery is verified.

Question 7

What is SSL monitoring and why does it matter?

Accepted Answer

Browsers block pages with expired security certificates behind a full-screen warning most visitors will not click through - so an expired certificate is functionally an outage, even though the server is running fine. It is also entirely preventable. We track expiry dates on every certificate and raise warnings from 30 days out, then verify the renewal actually deployed. Auto-renewal fails more often than anyone admits, and never on a convenient day.

Question 8

What does downtime actually cost while campaigns run?

Accepted Answer

Three ways at once. Paid clicks keep billing while the page is dark - the ad platforms do not know your server died. The leads who bounced rarely come back; they call whoever ranked next. And the buyers who hit an error remember it longer than you do. Search engines notice too. Google documents that pages returning persistent server errors get crawled less and can eventually drop from the index - one bad hour will not hurt, but chronic downtime compounds.

Question 9

Do we get a status page and reports?

Accepted Answer

Yes. A status page shows current state and incident history, so anyone in the company can check it before forwarding a "site down?" email. Every incident is logged with timestamps, duration, cause where known, and resolution. The monthly report turns the log into a decision document: measured uptime against the 43-minute budget, the incident pattern, response times, and what we recommend changing. Three outages from the same plugin in a quarter is not bad luck - it is a backlog item with a date.