Case Study: One Monitoring Window for Office, Cloud and Kubernetes Servers

A logistics company ran its business on seven servers in four different places — and found out about problems the hard way: from customers, from warehouse staff, or by accident. We gave them one monitoring window with all servers and logs, built on Grafana, in two days.

The situation

• Public website and customer portal on rented servers in a data center.
• A new version of the portal running in containers on a two-node Kubernetes cluster.
• Warehouse management backend in AWS — dozens of scheduled jobs: carrier rates, labels, tracking.
• A second cloud server for another site and a test environment.
• A print-and-scales server in the warehouse office (RED Print).

Every server was checked on its own, through a separate login. Logs were read only after something had already gone wrong — and the first signal was usually a customer.

What we did

1. Monitoring server in the client’s own office — a virtual machine on their existing Proxmox host, so the data never leaves the company.
2. One light agent on every server (Grafana Alloy), memory capped so it can never hurt production. It works on new and old systems alike — including a server still on an older Ubuntu release.
3. Safe delivery from the cloud: cloud servers send data over encrypted HTTPS with a secret key through a Cloudflare tunnel — no ports opened into the office. The dashboard opens only from the company’s approved addresses.
4. Logs that matter, labeled: website requests and errors, PHP errors and slow requests, each scheduled job separately (noisy ones excluded), portal security events, Kubernetes application logs and the system journal.
5. Human names: “Website”, “Customer portal”, “Warehouse backend · AWS”, “Print server · office” — instead of IP addresses.
6. Storage: a year of metrics, 30 days of logs.

The result

• One window for all seven servers — office, data center, AWS and Kubernetes.
• One search across all logs: “show every error in the last hour” instead of seven logins.
• A year of history: disk growth, peak load and slow periods are visible before they become problems.
• Ready for alerts: the rules below send messages to the RED Cube Monitoring bot in Telegram and to email, with an owner for each.

Alert rules for this client

• Disk will be full within 24 hours — on every server.
• Server silent for 5 minutes — power, network or hardware.
• Website and portal errors above 1% for 5 minutes.
• A scheduled job in the warehouse backend did not finish or logged errors.
• Repeated failed logins to the portal.
• SSL certificate expiring in 14 days.

Want the same for your company? Read how Grafana monitoring works in plain words, or talk to us.