About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
When a hosting proposal says 24/7 monitoring, you should expect to be able to get four things in writing: the list of what is watched, the condition that raises an alert, the named person or team who receives it, and the action that follows. Without those four, the phrase describes an intention, not a service you can hold anyone to.
This guide is for business owners and IT administrators comparing managed hosting offers. It does not tell you which tools a provider should use. It gives you a way to turn a claim into a checklist, using Google's Site Reliability Engineering (SRE) book and Microsoft's Azure Monitor documentation for vocabulary. Anything about what a managed customer should expect is our recommendation, not a standard.
Why it matters
Monitoring is how a fault becomes known. If nobody notices, the first alarm is a customer ringing you. If everybody is alerted about everything, the real alarm is lost among the noise.
Three different things sit behind the one word:
- Collecting measurements, such as processor load, free disk space and response times.
- Deciding when a measurement is a problem, which is the alert rule.
- Responding, which is people and procedures, and is where most of the value lies (our reading).
A claim that covers the first and says nothing of the third is the gap to look for. Monitoring also differs from backup, replication, failover and disaster recovery. Monitoring tells you something is wrong. Backup keeps older copies, replication keeps a current second copy, failover switches to it, and disaster recovery is the wider plan. A monitoring service does not by itself fix anything.
What is watched: symptoms and causes
The SRE book separates two ways of watching. White-box monitoring looks inside a system, at things such as logs and internal counters. Black-box monitoring tests the system from outside the way a user would. The book says it combines "heavy use of white-box monitoring with modest but critical uses of black-box monitoring". Black-box checks are symptom-oriented: they tell you users are affected now. White-box measurements often help find the cause.
The same chapter gives a short list for what a user-facing service should measure: "The four golden signals of monitoring are latency, traffic, errors, and saturation." In plain terms: how long requests take, how much demand there is, how many requests fail, and how full the service is. A single signal can mislead. The book's example is that a request can fail quickly or return a success code with the wrong content, and notes that "only end-to-end system tests can detect that you're serving the wrong content".
For a managed hosting customer, this suggests asking at two levels (our recommendation):
- Symptoms: is the application reachable, does it respond within a sensible time, and does a typical user journey work?
- Resources and causes: servers up, disk space, memory and processor pressure, storage health, network links, backup jobs completing, certificates approaching expiry, and security-relevant events.
A list that is all resources and no symptoms may report green while users cannot sign in. A list that is all symptoms tells you something broke but not where to look.
What counts as an alert
An alert is a notification that a rule has been met. The SRE book classes alerts by how they reach a person: they are pushed to a ticket queue, an email alias or a pager, and these are called tickets, email alerts and pages respectively.
That matters because the channel carries a promise about urgency. A page interrupts someone, at any hour. A ticket can wait for a working day. Which of your items go where is a decision, not a technical detail.
The book offers a test for whether something should interrupt a human: does the rule detect "an otherwise undetected condition that is urgent, actionable, and actively or imminently user-visible"? It also states: "Every page should be actionable." And it gives the reason noise is a fault in itself: "Effective alerting systems have good signal and very low noise." When pages occur too frequently, the book says, people second-guess, skim or ignore them.
For a customer, the useful reading (our recommendation) is: ask which items would wake someone, which would raise a ticket, and which are only visible on a report. A provider that pages on everything has probably not decided. For a service sold as 24/7, ask for an identified responder at every contracted hour and a route to that responder that has been tested. That can be an on-call person who is paged, or a team that is staffed continuously and sees the alert on a screen; the book does not require paging for every staffing model. A service where nobody is named for the night hours, or where the only route is an email nobody reads until morning, has not provided 24/7 response, whatever the sales page says.
Who sees the alert and what happens next
Azure Monitor's documentation shows how a system can separate the rule from the response. A rule monitors data and fires when a condition is met. It then runs an action group: "Action groups define who gets notified and what actions are taken when an alert fires." The notification types include voice call, SMS, push notifications and email, and the automated actions include a webhook, a function or a workflow (Azure Monitor action groups, updated 21 July 2026).
The same documentation describes a lifecycle. A fired alert has a system-set condition (fired, then resolved) and a user response that a person sets (new, acknowledged or closed) (Azure Monitor alerts overview, updated 8 July 2026). Stateful alerts, it says, "fire when the rule conditions are met, and will not fire again or trigger any more actions until the conditions are resolved". Alerts are kept for a limited time: "Alerts are stored for 30 days and are deleted after the 30-day retention period."
You do not need Azure to take the lesson. Any monitoring arrangement has the same parts, and you can ask about each (our recommendation):
- Who is notified, by name or role, and through which channel.
- What acknowledgement means, and how quickly someone should take ownership.
- What the next step is: a check, a restart, a call to you, or an escalation to a specialist.
- How resolution is recorded, so that you can see what happened afterwards.
- How long the history is kept, because a monthly report is only as good as the data behind it. The 30-day figure above is Microsoft's alert retention; your provider's may differ.
Reading the claim: scope, hours and responsibility
Here are the questions behind a phrase such as 24/7 monitoring. They are our recommendation.
- Scope. Which systems, and which layers: hardware, operating system, database, application, backup jobs, certificates, security logs? Servers you manage yourself, or software you install, may be outside the list.
- Hours. Is the watching continuous but the response only in business hours? Both can be called 24/7.
- Detection or response. Does the provider fix, or only notify you? If only notify, who is on call on your side?
- Thresholds. Who chooses the level at which a disk is considered full, and can you change it?
- Maintenance. How are planned outages handled so they do not trigger alerts, and how are you told?
- Reports. What do you receive, how often, and does it show alerts and responses or only uptime?
- Gaps. What is not monitored, for example custom software or a third-party service, and who is responsible there?
The SRE book gives one last test for the list itself, aimed at operational metrics: "Signals that are collected, but not exposed in any prebaked dashboard nor used by any alert, are candidates for removal." Applied to a proposal, a metric that is collected but is neither displayed nor alerted on is not giving you early warning, so ask why it is collected. The book calls such signals candidates for removal, not worthless: data kept on purpose for later investigation, audit or retention, such as security logs, can have value without a dashboard or alert, as long as the proposal says that is why it is kept and for how long. The book's test concerns immediate detection, and it does not say that data without a dashboard has no value.
Questions to put to your provider
- Can you give us the list of monitored items, with the condition that raises each alert?
- For each item, which channel does it use (page, ticket, email, report only), and which hours does that apply to?
- Who receives the alert on your side and on ours, and what happens if they do not acknowledge it?
- What is the first action taken for each item, and who is allowed to take it without asking us?
- How do we see alert history, and for how long is it kept?
- How are false alarms, planned work and repeated alerts handled?
- Which parts of the hosted environment are not monitored, and who is watching them?
- How do you check that the monitoring itself is working?
A monitored-item list you can ask for
This table is a template for the list in the first question. Rows are invented examples, not a claim about any provider.
| Item | Condition | Channel | Hours | Who receives it | First action |
|---|---|---|---|---|---|
| Website reachable | Fails from outside, repeated checks | Page | Continuous | On-call technician | Check service and network, then call the customer contact |
| Disk space on file server | Below an agreed free-space level | Ticket | Business hours | Support queue | Review growth, agree action with customer |
| Nightly backup | Job fails or does not start | Ticket, escalate if two nights | Business hours | Backup team | Re-run, report cause |
| Certificate expiry | Fewer than an agreed number of days left | Email and ticket | Business hours | Support and customer contact | Renew and install |
A worked example (illustrative)
The following is invented. A manufacturer is offered managed hosting for its ordering system, with 24/7 monitoring in the quote. It asks for the monitored-item list. The list contains server up or down, processor, memory and disk. It does not contain the ordering website, the database job that sends orders to the warehouse, or the nightly backup. The provider explains that it watches the servers and notifies the manufacturer's office manager by email, with no phone call. The manufacturer asks for a website check, a backup job check and a phone call for server down. The provider agrees the first two and quotes extra for the third. The firm now knows what it is buying, and records the decision.
Read next
Related articles you may want next.
- HostingSynthetic sign-in checks for a hosted application: what a probe provesWhat a scripted sign-in test catches that server monitoring misses, where it falls short, and how to choose one user journey to probe.Read article
- HostingManaged, co-managed or self-managed hosting: choosing how much of the server you runHow a business decides which layers of a hosted server it hands to a provider and which it keeps, using staff time, the need for administrator access and the evidence it must produce.Read article
- HostingSeverity levels, response and resolution times: how to read a support scheduleWhat a support response time commits to, how severity is decided, and which questions to ask before you sign a hosting support schedule.Read article
Risks and limits
- A list of monitored items shows intent, not performance. You still need evidence that alerts reach someone and that they act.
- The SRE book describes Google's own practice at scale. A smaller provider may reasonably do less. Its test questions are still useful for any size (our reading).
- The Azure pages describe Microsoft's tooling and are used here for vocabulary. A hosted provider may use other products.
- More alerts are not better. A long list that pages constantly may be worse than a short, well-chosen one.
- Monitoring does not fix, back up or recover. See the backup and recovery topics for those.
Next step
Ask your provider for the monitored-item list, with owners and actions, using the table above as a starting point. If you are reviewing a managed hosting arrangement, the Managed Services page is a starting point for that conversation; ask for the scope in writing.
Sources
- Monitoring Distributed Systems (chapter 6), Site Reliability Engineering, Google, O'Reilly Media, copyright 2017.
- Overview of Azure Monitor alerts, Microsoft Learn, last updated 8 July 2026.
- Create and manage action groups in Azure Monitor, Microsoft Learn, last updated 21 July 2026.
All accessed 15 November 2026.
Learning notes
- Monitoring tells you something is wrong; it does not repair or restore anything.
- Ask for the list: what is watched, what raises an alert, who receives it, what happens next.
- A page should be urgent and actionable. Constant noise hides the real alarm.
- Watching users' experience and watching resources answer different questions. You need both.
- 24/7 can mean watching all hours but responding only in business hours. Ask which.
This article was originally published on .