Read a sample: 1. Monitoring Starts With a Question
A support message arrives on a Tuesday morning: “The booking page is slow.”
The message is useful, but it does not yet tell the support team what to fix. Is the page slow for everyone or only one customer? Did the delay begin a few minutes ago, or has it been building since yesterday? Are requests failing, or are they completing more slowly than usual? Is the application itself busy, or is it waiting for another service?
Monitoring helps turn those questions into an investigation. It does not automatically explain every problem, and it does not replace the judgment of the people supporting an application. Its purpose is to make useful evidence available when someone needs to understand what is happening and decide what to do.
Beginners sometimes approach monitoring by looking for a list of numbers to collect. Processor usage, request counts, and error messages can all be valuable, but none is automatically useful in every situation. A better starting point is an operational question: a question about the behavior, condition, or use of a system that someone may need to answer.
For the support message above, a good first question might be:
> Has the booking application become slower for many requests, or is this report isolated?
That question suggests what evidence to seek. It also leaves room for more than one answer. If many requests are slow, the team may investigate a shared problem. If the available data shows normal response times, the team may ask the customer for more detail while considering whether its monitoring has missed part of the experience.
This chapter introduces a small application that will serve as a running example throughout the course. You will use it to practice separating what people experience from what monitoring records and from what a support team chooses to do.
Imagine a small organization called Cedar Grove Workshops. It offers classes that customers can browse and book online. Its application runs in Azure and has three main responsibilities:
- Show customers a list of available workshops.
- Accept booking requests.
- Store booking information so staff can review it.
The application uses a web application and a database. A small support team is responsible for keeping the service usable. The team does not have someone watching a screen at all times, and its members also have other work to do.
From a customer’s point of view, the important outcomes are straightforward: pages should open, workshop information should appear, and a booking should be accepted without an unreasonable wait. From the support team’s point of view, those outcomes create practical questions:
- Can customers reach the application?
- Are booking requests completing successfully?
- Are pages and requests taking longer than usual?
- When failures occur, do they share a useful pattern?
- Is the problem still happening, and who should investigate it?
These questions are related, but they are not identical. An application can be reachable while bookings fail. It can accept bookings while responding slowly. It can produce an error that affects one request without causing a broad outage. Treating all of these situations as simply “the app is up” or “the app is down” would hide information the team needs.
The exact Azure services used by Cedar Grove are less important for now than the relationships among the parts. Customers send requests to the application. The application processes those requests and may read or write information in the database. Monitoring data can describe aspects of that activity and the resources involved. Later chapters will add more detail about how Azure organizes and collects that data.
Monitoring is the ongoing practice of gathering and examining information about a system’s behavior and condition. The information can help a team notice a problem, investigate an incident, understand normal patterns, or check whether a change had the intended effect.
Consider four moments in the life of the booking application.
Before a problem: The team observes ordinary patterns, such as busier periods when new workshops are announced. This context helps the team interpret future changes. A high request count may be expected during an announcement rather than evidence of a fault.
As a problem develops: A measurement shows that requests are taking longer. The team may use that signal to look for an issue before receiving a large number of customer reports.
During an investigation: The team compares the timing of slow requests with records of failures and other relevant measurements. The aim is not merely to confirm that something looks unusual, but to narrow the possible explanations.
After a change: The team checks whether the application’s behavior improved. If a configuration adjustment was intended to reduce delays, the team needs evidence about response times after the adjustment as well as before it.
Monitoring supports each moment, but the questions differ. “Is the application slow right now?” calls for a different view from “Did the adjustment improve response times over the afternoon?” A useful monitoring setup makes those questions easier to answer without assuming that every answer should produce an immediate notification.
Three terms will help you reason about monitoring throughout this book: symptom, signal, and action.
A symptom is an observed effect that suggests something may need attention. It often describes the experience of a customer or the behavior of a system: a page is slow, a booking fails, or a service cannot be reached.
A signal is…