Services

Server and application monitoring setup

Monitoring that wakes a person instead of drawing pretty graphs. Alerting, dashboards, log aggregation and calibrated thresholds.

Timelinefrom 1 week

What's included

  • Availability and scenario checks
  • Server and application metrics
  • Alerting to chat, email, SMS
  • Grafana dashboards
  • Centralised log collection
  • Threshold tuning against false alarms
  • Incident history and postmortems

Monitoring exists for exactly one reason: so that you hear about a problem before your customers do. Everything else — graphs, dashboards, a nice panel on an office screen — is a pleasant side effect. If the system didn't wake the right person at the right moment, it failed, however detailed the dashboard looked.

That leads to the two classic mistakes. The first is watching only whether the homepage loads: a site can be perfectly available while payments have been failing for a week. The second is configuring so many alerts that nobody reads them. A hundred notifications a day guarantees the real outage gets scrolled past with everything else.

How we work

  1. Establish what's critical. We work out which failures actually cost money: payments, the enquiry form, the customer area, a data exchange with an external system. We monitor business functions, not just the server.
  2. Scenario checks. Beyond an availability ping, we script the key journeys: a test order, a form submission, a login.
  3. Infrastructure metrics. CPU, memory, disk, response times, error rates, queue depth and background job status. Disk space belongs on that list — a full disk takes systems down more often than people expect.
  4. Alert routing. We define the paths: what goes to chat, what is mirrored to email, what has to ring a phone at night. Different severities wake different people.
  5. Threshold tuning. The longest part of the job. Thresholds are set from several weeks of real data so alerts fire on genuine problems rather than on ordinary load variation.
  6. Logs and review. We centralise logs so there's somewhere to look after an incident, and set up a postmortem template — a short write-up for each outage.

What you get

A system that catches problems ahead of your customers, and a clear history: what happened, when, and for how long. A dashboard that shows service health without anyone having to ask. More importantly, an absence of noise — alerts arrive rarely, and each one means something needs doing.

Setting up monitoring is one-off work: we build the system and hand it over. If you need not just monitoring but someone who responds to what it reports, that's technical support with an on-call engineer, or 24/7 cover with an SLA if that response has to happen at night. For applications with queues and background jobs see web application support.

Timeline

Basic setup takes a few days to a week: availability checks, metrics, alert routing. Then comes calibration — two or three weeks of observation during which thresholds are tuned until the alerts stop lying. That stage can't be skipped: uncalibrated monitoring quickly turns into noise, and noise gets switched off.

A typical scenario

Consider a service with a homepage availability check that faithfully reports one hundred percent uptime. Meanwhile the background job that sends notifications stopped three days ago because the disk filled up. Formally everything is green; in practice part of the product is dead. Properly configured monitoring catches that within the first hour, because it watches for jobs completing successfully rather than for whether a web server answers.

FAQ

What does monitoring setup cost?
It depends on how many services are involved and how deep it goes: an availability check on one site and a full stack with metrics, logs and scenario checks are very different jobs. We quote after talking through what's critical for you.
How is this different from a free uptime checker?
A free ping answers whether the site loads. It won't see that payments are failing, that a data exchange has stopped, or that a queue is backing up. For a simple brochure site a ping is enough; for a store or an application it isn't.
Who responds to the alerts?
By default you or your sysadmin — we build the system and hand it over. If there's nobody to respond, you need support with an on-call engineer: then the alerts come to us and we handle them.
Will we be buried in notifications?
You shouldn't be, and that's exactly why calibration is part of the work. Thresholds are tuned to real load over the first few weeks. Monitoring that cries wolf gets muted within a month, which is a worse outcome than having none.
What do you build it on?
Usually Prometheus and Grafana where you want full control and your own infrastructure, or hosted services where the project is small and self-hosting would be overkill. We choose by the task rather than by habit.
Can you monitor the server as well as the site?
Yes, that's standard: CPU, memory, disk, network and service status. Free disk space deserves its own mention — a full disk brings systems down regularly and always unexpectedly.
What is the accumulated data for?
Two things: investigating incidents after the fact, and seeing degradation coming. A slow creep in response time is visible on a graph weeks before users start noticing it.
Other services