Skip to main content

Monitoring and alerts

How to watch the health and performance of your hosts, VMs and storage, and get alerted before users notice a problem.

🛰️ With Xen Orchestra​

Xen Orchestra is the natural place to monitor a whole XCP-ng infrastructure:

  • Dashboards and stats: every pool, host, VM and SR has a Stats view (CPU, RAM, network, disk I/O) with history, plus a global dashboard with health indicators (orphaned VDIs, missing patches, unhealthy VDI chains...).
  • Performance alerts: the perf-alert plugin emails you when CPU usage, memory usage or SR usage crosses a threshold you define, for the objects you select.
  • Backup reports: every backup job can send a detailed success/failure report (email, and messaging platforms via plugins).
  • Audit trail: records who did what on your infrastructure.

See the Xen Orchestra documentation for the details of each feature.

📊 Host-level metrics (RRDs)​

Each host continuously records metrics (CPU, memory, network, storage, per-VM and per-host) in round-robin databases (RRDs), with several months of history at decreasing granularity. This is what powers the graphs in Xen Orchestra and XO Lite.

You can query these metrics yourself:

root@xcp-ng-host — Host-level metrics (RRDs)
# List the metrics recorded for a host, then fetch values
# xe host-data-source-list host=<host>
# xe host-data-source-query host=<host> data-source=cpu_avg

# Same for a VM
# xe vm-data-source-list vm=<vm>
# xe vm-data-source-query vm=<vm> data-source=memory_internal_free

Individual data sources can be enabled or disabled (xe host-data-source-record/-forget, same for VMs), and the rrd2csv tool (shipped with XAPI) streams selected metrics as CSV for quick exports.

The raw RRDs are also exposed over HTTP for external collectors: https://<host>/rrd_updates?start=<unix-timestamp>&host=true returns everything that changed since the given time, in XML. This endpoint is a simple way to feed XCP-ng metrics into your own tooling without installing anything in dom0.

🖥️ Live view on the host​

For an instant view directly on a host (SSH or console):

  • xentop: per-VM CPU, memory, network and disk usage, as seen from the hypervisor. The %CPU column is relative to one physical CPU.
  • xl info: global hypervisor information (total/free memory, number of pCPUs...).
  • Standard Linux tools (top, iostat...) inside dom0 show the control domain's own resource usage. Remember that dom0 only sees its own slice of the machine: use xentop for the whole picture.

High dom0 load, or dom0 running out of memory, degrades every VM on the host: see the dom0 memory guide.

📡 Netdata​

For detailed real-time dashboards on a host, XCP-ng provides netdata packages, and Xen Orchestra Premium can install and integrate them in one click (host view, Advanced tab). Netdata gives you per-second metrics with a modern web UI, and can stream to a central Netdata parent if you have one.

You can also install it manually as an additional package, keeping in mind the general rule: dom0 is not a regular Linux box, keep extra software minimal.

🔔 SNMP and external supervision​

If your organization has a central supervision system (Zabbix, Centreon, LibreNMS, Prometheus...), a few options:

  • Agentless: scrape the RRD HTTP endpoint (above), or use the supervision system's API integration with Xen Orchestra.
  • SNMP: net-snmp is available in the XCP-ng repositories as an additional package:
root@xcp-ng-host — SNMP and external supervision
# yum install net-snmp
# configure a read-only community in /etc/snmp/snmpd.conf, e.g.: rocommunity monitoring 192.0.2.0/24
# systemctl enable --now snmpd

Then open UDP port 161 in the dom0 firewall (add the rule to /etc/sysconfig/iptables and restart iptables, so it survives reboots). Keep the community/users read-only.

  • NRPE: XCP-ng doesn't ship a Nagios NRPE integration. For Nagios-family systems, use SNMP (above), or netdata's collectors.
  • Dedicated agents (Zabbix agent...) can be installed in dom0 as additional packages, with the usual caution about keeping dom0 minimal.
warning

Whatever you install, never let a monitoring agent modify dom0's configuration or update its packages: see the additional packages rules.

🚨 What should I watch?​

A reasonable minimal set of alerts for an XCP-ng infrastructure:

  • SR free space: a full SR stops VMs from writing. Watch thin-provisioned SRs closely, snapshots and coalesce can consume space quickly.
  • Host memory: pool capacity to absorb a host failure (HA needs headroom).
  • dom0 CPU and memory: an overloaded control domain slows every VM.
  • Backup job results: a backup that silently stopped working is only discovered when you need it.
  • Hardware: use your server vendor's out-of-band management (iDRAC, iLO...) for disks, PSUs and fans; XCP-ng won't see everything the BMC sees.