Skip to content
Docs

Changelog

Onagre ships three products on their own schedule: the Platform (web app, API and Hubs), the Agent you install in your networks, and the Mobile app. Each release has its own entry, labelled with its product and the exact version you see in the product.

Current versions

  1. Agent 2.0.0.5

    New Linux installer, logs on standard output and more reliable Sensor execution

    New

    • New Linux installer (curl -fsSL https://d.onag.re/install.sh | sudo bash): one onagent systemd service per host, transactional upgrades checked against their SHA-256 checksum with automatic rollback, and the --version, --rollback, --set-token and --uninstall options. Unattended installs use --token-file or --token-stdin with --yes.
    • onagent --version works offline, without a token or configuration.
    • HTTP_DIAG=1 turns on a network diagnostic that logs DNS, TCP, TLS, time to first byte and redirects for each HTTP, Keyword, JSON and XML check.
    • SENSOR_WORKERS (default 32) sets how many checks run at the same time, and SENSOR_HTTP2=0 turns HTTP/2 off for HTTP Sensors.
    • /status shows the execution pipeline (queue, running checks, rejected runs, wait time), and a [PIPELINE] summary is logged every minute, as a warning when the Agent is saturated.
    • LOG_FORMAT=json writes the logs as compact JSON lines.

    Improvements

    • A saturated Agent no longer loses Sensor runs: the Hub only sends what the Agent can handle and a refused run is rescheduled.
    • Response times no longer include the time a check waited in the Agent's queue.
    • HTTP Sensors reuse connections and negotiate HTTP/2, with a fallback to HTTP/1.1, for steadier response times.
    • A missing or refused token stops the Agent with exit status 78 instead of restarting it in a loop, and temporary API errors are retried.

    Fixes

    • The Follow redirects option of HTTP, Keyword, JSON and XML Sensors is honored; redirects were always followed before.
    • When a host has several addresses, the Agent tries each of them instead of failing on the first unreachable one, such as an IPv6 address without a route.
    • An unreachable Certificate validity Sensor no longer slows down the Agent's other Sensors.
    • A hung check is cancelled and reported as a failure (Timeout reached) instead of producing no measurement.

    Security

    • On Linux, the Agent no longer runs as root: it runs as the onagent user, with CAP_NET_RAW as its only privilege, in a hardened systemd service.
    • The token is never accepted as a command-line argument.

    Breaking changes / action required

    • The default port of the health endpoints changes from 8080 to 27520: update your health checks, or set PORT.
    • On Linux and Docker, logs go to standard output (journalctl -u onagent, docker logs) and no log file is written by default. Set LOG=<path> to keep one; Windows is unchanged.
    • Loki is no longer supported: LOKI_ENDPOINT, LOKI_ACCOUNT and LOKI_PASSWORD are ignored, with a warning at startup. Point your own log collector at journald or the container logs.
    • The onagent.sh script and its multi-instance onagent@<name> services are replaced by install.sh, with one Agent per host and no automatic migration: remove the old service, then run the installer. Use Docker to run several Agents on one host.
    • The Agent no longer reads a .env file from its working directory: pass settings as environment variables (/etc/onagent/onagent.env, docker --env-file).