GitCloud.pl
Contact
Service · subscription

On-call adminnight duty for a Proxmox and OPNsense server

For a small business running a Proxmox server and an OPNsense firewall. Night alarms are handled by an on-call program: it fixes known faults by itself, takes a copy before every change, and sends me whatever it cannot settle as a card with options on my phone.

Running on my own infrastructure since October 2026. Package for businesses since 14 November 2026, install tested on a clean Debian 12

How it works

An alarm without a recipe, or a change that cannot be rolled back: a card goes straight to me, no action taken.

What it does

Night dutyNetdata alarms: host and guest services, disks, memory, reboot after a kernel update, websites and certificates. Every alarm has a recipe: what to check, what to do, when it counts as fixed.
Safe changesA copy of the machine before a change. If the service does not come back or the connection drops after it, the system returns to the state before the change by itself.
Weekly reviewPackages with security vulnerabilities, disk and memory forecast, services that stopped working, age of backups.
Decisions by cardWhen a recipe is not enough or a change is risky, I get a card: what happened, what is already done, 2 to 4 possible steps. I choose and I am responsible for the decision.

Since November 2026 the on-call program also handles alarms from the OPNsense firewall and errors from Loki logs.

Rules of the on-call program

By itself
  • re-checks the alarm against the live metric
  • restarts a failed service (except networking, SSH and the Proxmox cluster)
  • checks services inside Proxmox machines and on OPNsense
  • tests websites, ports and certificates
Only in the night window
  • heavier steps that may briefly interrupt work
  • we set the window together, 02:00-05:00 by default
Only with consent
  • server reboot after an update
  • limiting machines that eat memory
  • anything a recipe does not cover
Never
  • does not delete data
  • does not change the network or firewall
  • does not touch keys or payments
  • does not write to people on your behalf

A recipe that failed twice within a day is not repeated: the case comes to me. Every step is written to a log on your server (90 days).

A report every week

On Monday at 7:00 you get a one-page e-mail in Polish. Example:

Dyżurny admin · raport · Firma Przykład · serwer pve1 · 07.11-14.11.2026

W skrócie: 672 przeglądy, alarmów: 3, zamknięte bez Państwa udziału: 2, czeka na decyzję: 1.

Czeka na decyzję:
• ram_available: czeka na decyzję (od 13.11 22:05)

Co się działo:
• 09.11 03:12 systemd_service_unit_failed_state: restart usługi → rozwiązane
• 11.11 14:40 disk_space_usage: kontrola miejsca na dysku → sprawdzone, bez zagrożenia

If Netdata stops responding, I get a message straight away and the report says so plainly: a week without checks does not pretend to be a quiet week.

GitCloud Shield: included in the subscription

GitCloud’s own security system, included in the subscription. It checks messages and mail before anyone opens them and scores server security from 0 to 100.

Mail and messengersPhishing, scams (“my own courier”, payment “before shipping”, BLIK or SMS codes), fake banks, OLX, InPost and gov.pl. The warning arrives without the text and without links.
Attachments“invoice.pdf.exe”, programs, disk images, login pages, macros. Contents are scanned by ClamAV, an infected file goes to quarantine.
Network block515k phishing domains (CERT Polska and open lists) blocked on the firewall DNS: a fake site will not open on any device in the company.
Score 0 to 100Firewall, fail2ban, file integrity, backups, updates and the shield in one view with a “to fix” list. A falling score is an alarm for the on-call program.
515kphishing domains blocked, updated hourly
1 of 1057messages from a month flagged as dangerous: a real OLX scam

What stays with you

  • The on-call program is a systemd service on your server, its Python code lies open in /opt/dyzurny.
  • Data, logs and backups stay on your hardware.
  • Only e-mails leave the network: a decision request to me and the weekly report to you.
  • Settings in one file: /etc/dyzurny/dyzurny.toml.
  • Switched off and removed with one command: install.sh --uninstall.

A real example

Environment: one Proxmox server with containers and virtual machines, an OPNsense firewall, Netdata monitoring. As of 12 November 2026:

1summary in the morning instead of notifications at night
02:00-05:00update and reboot window, a check after every step
  • A service did not start after an update: restart, metric check, alarm closed without waking anyone.
  • Forecast "disk full in a few days": the on-call program checked free space right away. It was a short spike of temporary files and there was enough space. It reported this and deleted nothing.
  • A new kernel after an update: host reboot only in the night window and with consent, then a check that the shutdown was clean and services came back.

Price

setup1500 PLNOne Proxmox server and an OPNsense firewall, a week of configuration and testing.
subscription390 PLN / monthNight duty, weekly reviews and up to 2 hours of work on decisions per month.
each extra server+150 PLN / monthEvery further server under the same duty.

A fixed price for the scope, agreed in writing before the start. Contract and invoice on request.

How to start

  1. 1
    WriteHow many servers you have and what runs on them.
  2. 2
    AssessmentI check the environment, price and scope in writing.
  3. 3
    SetupOne week: the package installs in one evening, then recipes for your services and alarm tests.
  4. 4
    SubscriptionDuty every night, a report every week.
Send a message →

Questions

What if the on-call program does not know the alarm?

It changes nothing and sends me a card. Once solved, I add a recipe, so next time it handles it by itself.

Can the program break something?

The rules are written in the code: it does not delete data or change the network or firewall, heavier steps run only in the window, a server reboot only with consent. Configuration changes I make with a copy and automatic rollback.

What if monitoring itself goes down?

The on-call program checks Netdata before every run. If it does not respond, I get an e-mail straight away and the weekly report shows the gap.

Does it replace backups?

No. The review checks that backups are fresh, and a missing one is an alarm.

Is everything done remotely?

Yes. I do not do hardware repairs or on-site visits.

Esc

↑↓ select · Enter open · Esc close