SMB infra monitoring without DevOps: 5 signals that matter
A customer writes: “The cabinet won’t open.” The admin is on leave. Hosting is green: “server pings.” An hour later you learn it was not “the hardware” — login to the cabinet failed, and the accounting server disk was already at 98% yesterday. For an office of 30–150 people without in-house DevOps, this is normal: monitoring exists on paper, and the first signal comes from humans, not from the system.
Below is a frame for SMB CEOs and IT admins: infrastructure monitoring without DevOps as five signals that should wake you before the customer call. No Grafana “for looks,” no “bank-grade” promises.
Key takeaway. Start with five signals: money/accounting path down, disk filling, backup silent, latency/errors rising, SSL/license dates burning. Three Telegram alerts beat a hundred ownerless charts.
Five signals beat a pretty dashboard nobody reacts to
Why “server pings” is not monitoring
Ping and “homepage opens” miss:
- cabinet or web accounting returns 500 while the storefront is green;
- disk is full — tomorrow morning the database will not accept documents;
- nightly backup has failed for two weeks while the panel checkbox lies;
- SSL expires on a weekend — browsers scare customers with a red lock.
For production website metrics (latency, errors, capacity headroom) see production monitoring: 4 metrics. Here the focus is different: SMB infrastructure without a dedicated SRE — accounting servers, access, backups, and expiry dates.
Signal 1. Money or accounting path is down
Answer first: check from the outside not “server alive,” but the URL where the business loses money or stops work — customer cabinet, checkout, web accounting, lead form.
Every 1–5 minutes an external check (UptimeRobot, Better Stack, host monitor, or a simple script) hits Telegram if the status is not 200 or it times out. Pick two or three URLs, not twenty. Human names in alerts: “Customer cabinet,” not an IP.
If you learn about outages from customer chat, this signal does not exist yet — even with “99.9% uptime” in the hosting panel.
Signal 2. Disk is filling up
Answer first: a full disk kills accounting, mail, and sites more quietly and more often than a “broken CPU.” Threshold is not 99%, but 80–85% used, with room for a day of logs and backups.
Alert “disk >85%” on accounting, site DB, and file-share servers. Without DevOps this is often one host agent or out-of-the-box Zabbix/Netdata plus Telegram. Cost: an hour of admin time; cost of silence: a day of database downtime.
Signal 3. Backup is silent or failing
Answer first: alert on “no successful backup for N hours” or a failed job — not on “job is scheduled.” Otherwise “everything is backed up” lives until the first real incident.
For 3-2-1 and restore drills see SMB backups and DR. Monitoring here is the watchdog: if the copy did not appear, someone learns today, not on ransomware day.
Signal 4. Slowdowns and errors on the business path
Answer first: watch 5xx share and response time on the same URLs as signal 1. “Average is fine” lies — worst cases and error spikes matter.
SMB starting thresholds: response over 2–3 seconds for a noticeable share of requests, or a 5xx spike — “important within an hour” alert. Bridge to a stress test before ad peaks: monitoring catches degradation; the test proves headroom before the campaign.
Signal 5. Expiry dates: SSL, domain, licenses
Answer first: calendar bombs without DevOps explode on weekends. Reminders at 30 / 14 / 7 days before SSL, domain, accounting license, or a critical SaaS expires.
Not a “server metric,” but part of infrastructure monitoring: without it the director learns from a customer screenshot of “your connection is not private.” Keep one expiry table plus a bot/calendar with escalation to two people.
Where alerts go and who gets woken
- Urgent — money/accounting path down, mass 5xx → Telegram + call the on-call
- Today — disk >85%, backup failed, SSL <14 days → IT chat, escalate to the director by evening
- Morning — “disk is growing,” one-off errors, SSL <30 days → no 3 a.m. call
Two owners on critical alerts (primary + backup). One person on leave with no deputy means monitoring is off. On how many “nines” you actually need — see SLA 99.9% vs 99.99%.
What silence costs
SMB ballpark: an hour of site or accounting downtime often 50–200k+ RUB in lost revenue and payroll; a full day is another order of magnitude. A five-signal stack usually costs tens of thousands of RUB to set up and pennies for external checks — cheaper than one quiet incident. Full math: cost of one hour of downtime.
Checklist: infrastructure monitoring without DevOps
- 2–3 money/accounting URLs with external checks every 1–5 minutes.
- Disk alert on accounting/DB/files at >80–85% used.
- “No successful backup” within the agreed window + date of last restore test.
- 5xx and latency thresholds on those URLs — not ping alone.
- SSL / domain / license calendar with 30/14/7-day reminders.
- Two people get critical alerts; urgent / today / morning are separated.
- Quarterly drill: disable a check on purpose — did the alert arrive within 5 minutes?
When to hire DevOps instead of “one more free monitor”
Five signals stop “learning from chat.” Next level: ad peaks, several sites, contractual SLA with customers, or 24/7 on-call. Then you need a shared dashboard, error budget, and a real DevOps setup — not a tenth Telegram bot.
If the checklist above is red, start this week with money-path, disk, and backup alerts. Grow the rest over a month. Need help standing up the stack — estimate the work or message @MozziDev.
Related services
Executive FAQ — monitoring without a DevOps hire
No. For an office of 30–150 people, start with external checks on key URLs, disk alerts on accounting/web servers, “backup did not succeed,” response-time thresholds, and SSL/license reminders. Grafana and Prometheus come later — after alerts work and someone owns them.
Hosting uptime on the homepage can stay green while the customer cabinet, web access to accounting, or checkout already returns errors. Infrastructure monitoring watches money and operations paths plus disk, backups, and expiry — not only an ICMP ping.
1) money or accounting path is down; 2) disk is filling up; 3) backup is silent or failing; 4) errors/latency spike; 5) SSL, domain, or license is about to expire. Those five cover most “we learned from a customer chat” incidents.
A shared Telegram chat for IT plus the director for critical events, and a separate “morning” channel for trends. Urgent = downtime and mass errors; disk and backup = business hours with escalation if not fixed the same day. Without priority tiers, the team burns out and ignores signals.
An hour of site or accounting downtime often costs tens to hundreds of thousands of RUB in revenue and payroll; a full day can reach hundreds of thousands to millions plus slipped deals. A basic alert stack is usually cheaper than one quiet incident. Use the same logic as in the downtime-cost article.
Want to apply this in practice?
Tell us about your system — we’ll propose a work plan and the metrics worth fixing in an SLA/SLO.
Related articles
Code from AI Agents: Checklist Before Production
AI agents write code: a pre-production control checklist for CTOs — leak, license, and silent-bug risks, review gates, and incident cost ranges in ₽.
Read ArticleProduction monitoring: 4 metrics anyone on the team can understand
Production monitoring in plain language: site speed, errors, traffic, and server headroom. What to check before ads and how not to learn about outages from angry customers. DevOps, Grafana, Prometheus.
Read ArticleDevOps and CI/CD in Production: What to Set Up First
DevOps services for business: build pipeline, staging, zero-downtime deploy, monitoring and rollback — priorities for the first 4–6 weeks.
Read ArticleKubernetes in Production: A CTO Checklist Before Launching a Cluster
Production Kubernetes setup: RBAC, resources, Ingress, GitOps, monitoring, and common mistakes — a checklist before going live.
Read Article