Skip to content
Justin White
← Writeups

Monitoring on a too-new Debian: the Zabbix-on-Trixie saga

I wanted Zabbix watching the Proxmox nodes and the Cisco switch. The host was on Debian 13 (Trixie) before Zabbix shipped a Trixie repo, so apt had nothing, the build-from-source fallback went sideways, and PostgreSQL wasn't even installed where the guide assumed. A monitoring setup that fought back.

4 min read Home lab: Home lab Info zabbixmonitoringdebianpostgresqlhomelab

Running things on a too-new distro is a recurring home-lab tax: you get the shiny new base before the ecosystem on top of it has caught up, and routine installs turn into projects. This is the Zabbix-on-Trixie version of that story, with notes on where it stuck.

The goal

Zabbix watching the Proxmox cluster (host metrics via the agent) and the Cisco switch (over SNMP). Ordinary home-lab observability, paired with the Wazuh/Graylog side that handles security telemetry.

Problem one: there’s no repo yet

The host was on Debian 13 (Trixie). Zabbix publishes a repo per Debian release, and at the time there simply wasn’t a Trixie one. apt knew nothing was installed:

nope@zabbix:~$ systemctl status | grep zabbix
(nothing)

and grabbing the release package by its usual name just 404s, which then cascades into dpkg failing on a file that was never downloaded:

2026-04-11 22:53:28 ERROR 404: Not Found.
dpkg: error: cannot access archive 'zabbix-release_latest_*.deb': No such file or directory

On a too-new distro there are two pragmatic moves: point at the previous release’s repo (Bookworm) and trust the binaries to run, or build from source. I ended up doing both, one for the agent and one for the server.

The agent: take the Bookworm repo

The Proxmox hosts only need the agent, and that’s the easy path: zabbix-agent2 from the Bookworm repo runs fine on Trixie. Install it, enable it, and confirm it’s listening on its port:

sudo apt install -y zabbix-agent2
sudo systemctl enable --now zabbix-agent2
ss -tulnp | grep 10050        # agent2 listening on 10050?

Then prove it from the server side rather than assuming. Ask the agent a question and see if it answers:

zabbix_get -s <proxmox-ip> -k system.hostname

If that returns the node’s hostname, the agent half is done.

The server: building from source, and the postgres surprise

The server is where the time went. A from-source ./configure in ~/zabbix was cheerfully reporting that it was going to build almost nothing:

Enable server:  no
Enable proxy:   no
Enable agent:   no

The real wall was the database. Zabbix wants PostgreSQL, and the host didn’t have it the way the install scripts assumed:

sudo: unknown user postgres
id: 'postgres': no such user
Unit postgresql.service could not be found.

That trio is unambiguous once you stop reading it as a permissions bug: there is no postgres user and no postgresql.service because PostgreSQL isn’t installed at all. The Zabbix database-setup steps assume a packaged PostgreSQL that created the system user and the service unit for you. Skip that, or land on a distro where it didn’t happen, and every sudo -u postgres ... in the guide collapses. The fix is install and start PostgreSQL first (it’ll listen on localhost:5432), create the zabbix database and the zabbix role, then point the server at it. The bootstrap is straightforward, but the source build won’t do it for you.

The switch: SNMPv3, proven before the UI

The Cisco switch is monitored over SNMP with Zabbix’s stock Cisco IOS by SNMP template. The discipline that saves an afternoon here is to prove the credentials and crypto with a raw walk before wiring anything into the Zabbix UI. Query the device’s own sysName OID and see if it answers with its hostname:

snmpwalk -v3 -l authPriv -u zabbix -a SHA -A '<auth-pass>' -x AES -X '<priv-pass>' \
    192.168.3.10 1.3.6.1.2.1.1.5.0

A hostname coming back means the SNMPv3 user, the auth (SHA) secret, and the priv (AES) secret all line up, so when Zabbix later says “no data,” you already know it’s a template or item problem and not a credentials one. (Real secrets redacted here; don’t ship the placeholders.)

Honest status

The agent side works and is useful. The full Zabbix server is “mostly stood up” rather than a finished dashboards-and-alerts deployment. The too-early Trixie timing turned a fifteen-minute apt install into a build-and-bootstrap project, and there’s no polished NOC behind it yet.

Takeaways

  • A too-new distro means owning the ecosystem gap. Trixie before Zabbix’s Trixie repo meant apt had nothing and the release .deb 404’d.
  • The previous release’s repo is the pragmatic fallback for agents. zabbix-agent2 from Bookworm runs fine on Trixie.
  • unknown user postgres / postgresql.service could not be found is a missing dependency, not a config bug. Install PostgreSQL first; the source build won’t bootstrap the DB for you.
  • Prove SNMP with a snmpwalk of the sysName OID before opening the monitoring UI. It cleanly separates “credentials/crypto wrong” from “template wrong.”