If you ask a sysadmin what the most important thing in running an infrastructure is, they are either going to say “backups” or “monitoring”. Yet, a surprising amount of systems only have rudimentary monitoring at best.
When I go into a system to do an audit, I usually discover that if they even have a monitoring, they usually run with some default settings that do no good whatsoever.
So how do we design a monitoring system? No, we are not going to talk about specific software like Icinga, Prometheus or Grafana, we are going to talk about the things you need to think about when setting up a monitoring system.
The questions monitoring answers
Before we dive in, we need to ask ourselves the question: What is the purpose of our monitoring system? What should we monitor?
The obvious answer I hear from a lot of business owners is: “Everything!” But nothing could be further from the truth.
In my eyes monitoring has two purposes. First of all, it needs to tell you when something happens in the system that impairs the business. Nobody wants to get up at 2 AM in the morning for something that doesn’t affect the business.
Second, it should help with forecasting. It should tell you about things that will soon become an issue.
If, for example, your site is down, you may want to react immediately. In contrast, if one of the disks is soon going to die, you don’t care at 2 AM in the morning. You do, however, care in the long run, as you may need to order more disks.
This brings us to the next important point about monitoring: trust. A lot of monitoring systems are set up in a way that they alert for absolutely everything.
If you get alerts for irrelevant things, or even worse, something that isn’t even an issue, you will lose trust in your monitoring system. You will stop paying attention, or even disable alerts.
If your monitoring system alerts for something it shouldn’t, you need to go fix it, or alarm fatigue will set in.
Types of monitoring
Monitoring can be done in a variety of ways. When business owners think of monitoring, they think of it from a top-down perspective: does the system work. They will want to run a check every couple of minutes, checking if the site loads, if you can do a purchase, etc.
However, these kinds of checks only tell you if a system is working or not, they don’t tell you the reason. So you need to dig deeper. The situation is similar to the difference between integration tests and unit tests: the former tells you if the system is working, the latter tells you what’s wrong with it.
Active vs. passive
A different way of looking at monitoring is the source of the data. Systems with low traffic volume will want to go with an active check: the monitoring software will actively connect to the service and perform a check of its functionality, such as load the website. This is great, because it actively checks, even if users are not hitting the website, but if there is a temporary failure, it may go uncovered if it lasts less than the period between checks.
The other type of checks is passive, for example by collecting data. If you have enough traffic on your website, or enough purchases on your webshop, you can set up a system that collects data and alerts if a certain number reaches a high or low point. For example, you could monitor 5xx HTTP response codes. Or you could monitor that there are at at least 10 purchases per hour.
However, with thresholds like this you need to be careful. For example, you will want to set a different threshold during the night than during the day, otherwise you will find that you get lots of false alarms, or on the flip side, you don’t discover issues quickly enough.
Now, one important aspect that is often neglected when it comes to monitoring systems is the backups and disaster recovery of it. If your monitoring system crashes, you are flying blind. You absolutely have to make backups of your monitoring system, and you have to be able to restore operations quickly in case of a failure.
This applies to third party monitoring providers as well. A misconfiguration on their part means that all your carefully set monitoring configuration could be lost. To prevent that you either need to back up your configuration, or use Infrastructure as Code tools (e.g. Terraform, Ansible) to create that configuration in the first place.
Monitoring is hard. Sometimes mistakes happen, and you get false positives or don’t discover an outage in time. That’s normal. Don’t, however, allow yourself to normalize false positives and missing or inconclusive alerts. If a mistake happens, make sure to fix it. ∎