System, web and mail integrations
The agent recognises installed software and collects its metrics, with Datadog-compatible names. Most integrations turn themselves on.
How it works
enabled: auto: the integration turns on after a first successful read. Software absent: state "absent", no check, new attempt every 10 minutes.enabled: trueforces the integration,enabled: falseturns it off.- Each integration publishes a
<name>.can_connectcheck and never breaks the agent if the tool is missing. - The state of each one is shown in Settings › Agents.
Settings go under integrations.<name> in the configuration. Databases have their own page: Databases.
Web servers and load balancers
| Integration | Source | Main metrics | Activation |
|---|---|---|---|
nginx | stub_status page | nginx.requests, nginx.connections.active, .reading, .writing, .waiting | enabled: true and url |
apache | mod_status (?auto) | apache.net.hits, .request_per_s, .bytes_per_s, apache.performance.busy_workers, .idle_workers, apache.conns_total | auto |
haproxy | admin socket (show stat) or CSV page | haproxy.frontend.*, haproxy.backend.*, haproxy.server.*, haproxy.backend_up, haproxy.server.up | auto |
iis (Windows) | performance counters | iis.requests_per_s, iis.current_connections | auto |
For nginx, open the status page to the loopback only:
location /nginx_status { stub_status; allow 127.0.0.1; deny all; }
integrations:
nginx: {enabled: true, url: http://127.0.0.1/nginx_status}
apache: {enabled: auto, url: "http://127.0.0.1/server-status?auto"}
haproxy: {enabled: auto, socket: /run/haproxy/admin.sock} # or url: http://127.0.0.1:8404/stats;csv
nginx, Apache and HAProxy access logs are parsed on arrival: method, path, HTTP code, duration, client address. See Logs.
PHP-FPM, Varnish, Squid, Lighttpd, Gunicorn and Supervisor, as well as PowerDNS, OpenLDAP, Keycloak, Nexus, Harbor, ZooKeeper and Ceph, have their dedicated probes: see Applications and services.
Jenkins, GitHub (and GitHub Enterprise Cloud), the Zot registry and Odoo: see CI, forges and business apps.
DNS
bind reads the JSON statistics channel (statistics-channels { inet 127.0.0.1 port 8053; };), otherwise rndc status. Metrics: bind.queries by type, bind.responses.nxdomain, bind.responses.servfail, bind.zone.serial per zone, bind.recursive_clients.
Pi-hole, AdGuard Home and Unbound, as well as Caddy, MinIO, Mosquitto, Proxmox Backup Server, Borg and restic, have their own page: Network, storage and backups.
| Integration | Source | Metrics and checks |
|---|---|---|
postfix | postqueue -j and delivery logs | postfix.queue.size per queue (active, deferred, hold, incoming), postfix.delivery.sent, .deferred, .bounced |
dovecot | doveadm stats dump, doveadm who | dovecot.auth.success, .failures, dovecot.connections |
rspamd | local /stat page | rspamd.messages.scanned, .learned, .spam, .ham, rspamd.actions per action, rspamd.connections |
exchange (Windows) | performance counters | exchange.queue.<state> per transport depot, exchange.database.mounted (and check), exchange.imap.connections; checks for poison and suspended messages |
To check mail end to end (SMTP banner, IMAP, send then fetch, SPF, DKIM, DMARC, blocklists), see mail synthetic tests.
Security and access
openvpn: status files (server: connected clients; client: tunnel throughput):openvpn.clients.connected,openvpn.client.bytes_in,.bytes_out. A file frozen for 10 minutes turnsopenvpn.can_connectcritical.fail2ban:fail2ban-client status, read every 5 minutes:fail2ban.banned,fail2ban.failedper jail.- Authentication logs:
sshd,sudo,suandfail2banlines are analysed on arrival (evt.name,evt.outcome,usr.name,network.client.ip). The "SSH brute-force attack" template relies on them. ssh: availability and banner of SSH servers, without authentication (the agent reads the banner, sends its own and closes; the server logs a connection closed before authentication). Onessh.can_connectcheck per target: critical if the port does not answer or does not speak SSH, warning if the banner does not matchexpect_banneror announces protocol 1; metricssh.response_time. Inactive until targets are given:
integrations:
ssh:
targets:
- {name: bastion, address: "bastion.example:22", expect_banner: "^SSH-2.0-OpenSSH_9"}
- {address: 192.0.2.10} # port 22 by default
wazuh: REST API of the Wazuh manager (port 55000), with an API account holding thereadonlyrole (url,username,password,insecure: truefor its self-signed certificate). Manager daemons (wazuh.manager.daemons: critical if analysisd, remoted, wazuh-db, modulesd… is stopped), agents by connection status (wazuh.agents,statustag), events received, processed and dropped, alerts written, analysisd and remoted queue usage, remoted sessions and bytes.suricata: theeve.jsonfile (default/var/log/suricata/eve.json), followed as it grows: lateststatsrecord (captured and dropped packets, drop sharesuricata.capture.drop_rate, decoding, flows, TCP, detection, flows per application protocol, memory) and alerts counted by severity and action (suricata.alerts).suricata.can_readcheck: critical if the lateststatsrecord is older thanstale_after(5 min; Suricata stopped, orstatsmissing from theeve-logtypes). On the first pass, only the last 4 MiB of the file are read.
Certificates
x509 watches the certificates installed on the host: tls.cert.days_left and the tls.cert.expiry check (warning under 21 days, critical under 7 days), read every hour.
- Paths read by default:
/etc/letsencrypt/live/*/cert.pem,/etc/ssl/*/fullchain.pem. - Discovery (
discover, on by default): the certificates declared by Apache, nginx, Postfix, Dovecot and HAProxy. - Windows: the computer store (
stores: [My]), without certificates replaced by a newer one or expired for more than 30 days.
The Certificates page of the interface gathers every certificate collected by agents and synthetic tests, sorted by expiry.
Hardware and system
| Integration | Source | Metrics and checks |
|---|---|---|
smart | smartctl, every 30 min, without waking a sleeping disk | smart.health, .temperature, .reallocated_sectors, .power_on_hours, .wear_level |
hwmon | /sys/class/hwmon | system.temperature, system.fan.rpm |
ntp | chronyc, timedatectl or ntpq; w32tm on Windows | ntp.offset, ntp.in_sync check |
updates | package manager cache (no apt update), read every hour | system.updates.pending, .security, system.reboot_required; Windows: windows.updates.pending check |
systemd_timers | systemctl | systemd.timer check: critical if the last run failed, warning if an active timer has not run for two periods |
virtualbox | accounts' VirtualBox.xml and processes | virtualbox.vms.running, .total, virtualbox.vm.running check for autostart VMs |
btrfs | /sys/fs/btrfs (error counters: kernel 5.14 and later, otherwise btrfs device stats) | per mounted filesystem: btrfs.allocated, .used, .usage per type (usage_type) and profile (replication_type), btrfs.unallocated and .unallocated.pct, btrfs.device.errors; btrfs.health check: critical if a device is missing, warning if error counters are not zero |
nfs | client: /proc/1/mountstats; server: /proc/net/rpc/nfsd | client, per mount: operations, bytes read and written, average round-trip and execution times, retransmissions, timeouts, usage; nfs.mount check (critical if statfs does not answer within timeout, 5 s: hung mount). Server: nfsd threads, RPC calls and bad calls, bytes, operations per version, exports; nfs.server check |
An NFS server in a container keeps its statistics in its own network namespace: integrations.nfs.server_stats: /proc/<container pid>/net/rpc/nfsd. nfs.mounts restricts the followed mounts (patterns).
Processes, ports and pages
Without any specific integration, the agent can also check:
- that a process is present (
checks.processes):process.upcheck andprocess.countmetric; - an HTTP page (
checks.http) or a TCP port (checks.tcp), from the host's network.
Your own metrics
The agent listens for StatsD in DogStatsD format on 127.0.0.1:8125/udp, aggregates over 15 seconds and sends everything as regular metrics:
echo "shop.orders:1|c|#env:prod" | nc -u -w0 127.0.0.1 8125
For an application that already publishes Prometheus metrics, Nagios or Centreon plugins, or logs sent over syslog, see Generic collection.
Counters (c) become sums, gauges (g) values, timers (ms, h) distributions (.avg, .max, .p50, .p95, .p99, .count). For OpenTelemetry metrics, see Traces.
Updated: 20 November 2026