- Game hosts: a "game query" now complements ping. The agent queries the game server itself and confirms that a game is really running there: server name, game, map, player count. Requires agent 2.29.0.
- Latency to a game host is visible even when the server does not answer ICMP: it comes from the game query.
- "Game hosts" panel: game query status under the name, an "unconfirmed only" filter, search by server name and game.
- Steam Datagram Relay relays are recognised automatically and named "Steam SDR" with the PoP name.
- The route server of an internet exchange (IX) is also detected from PeeringDB data and can be set manually.
Changelog
What's new
New versions of the center (server and panel) and of the argus-probe agent. Newest first.
Center and panel
- BGP notifications are shorter. A session with your own probe is marked, an IX route server is shown by IX name and AS number, other neighbours by a short AS name. The same applies to flap reports.
- The uplink shows a short AS name in brackets, for example "AS3356 (Lumen)".
- Game hosts: the host's PTR record, search by PTR.
- Game hosts do not send a notification per target. When the problem is on the probe or hosting side, one summary message covers several probes.
- A game host that has not answered for 24 h becomes inactive and is pinged less often. As soon as it answers, it is pinged as before.
- New "Game hosts" panel page: game servers your subscribers play on. Hosts are found automatically, taken from a catalogue or added by hand. Each one shows country, AS, subscriber count for 24 h and 7 days, uplink in both directions with shares, latency and loss from probes in your AS, and a 24 h chart.
- Subscriber count: static subscribers are counted exactly from flow, CGNAT subscribers exactly from an A10 session snapshot (agent 2.26.1+) or as a lower bound.
- Ping to a game host nobody has visited for 14 days is paused and resumes by itself once subscribers come back.
- Notifications when latency to a game host grows or loss appears on probes in your AS, with the uplink. A daily digest of changes.
- The uplink, meaning which AS or IX carries traffic to and from a target, is now shown for regular targets too, labelled with the interface or AS name.
- Session snapshot requests no longer overload the A10: the next one starts only after the previous one finishes, and "busy" replies are respected.
- Telegram chat topics by category: BGP, network, CGNAT, quarantine, classifier, BRAS, monitoring. Set them in the panel: Settings → Telegram channel card.
- A staff comment on a manual A10 action appears in the Telegram message about its result.
- Agent update state (2.26.0+) is shown on the probe card: pending, blocked or failed, with the reason.
- BGP session flaps are folded. The first two changes arrive at once, then a summary report at a growing interval, and a final result at the end: "stabilised" or "not established".
- Auto-quarantine: the infection case state is shown in the status. For subscribers stuck without relapses there is a suggestion to resume softening, with a button in Telegram. A standing hold list of addresses that the automation never softens or releases.
- After each action, auto-quarantine checks that the subscriber really goes through the quarantine NAT pool on the A10. On a mismatch it repeats the move, reconciles hourly and notifies the operator if it fails. Requires agent 2.25.0+.
- Monitoring of quarantined subscribers refreshes every 15 minutes with live session and NAT pool data.
- A10 commands carry an expiry time: agent 2.25.1+ will not run an expired command after coming back online.
- Every bot message in Telegram ends with the time it was sent or last updated.
- A10 LSN drop alerts count only real drops. Traffic dropped by the quarantine itself is shown separately on the chart.
- The infection state is cleared automatically only on a fresh clean session snapshot. If the subscriber was silent for a long time, two clean snapshots some time apart are needed. A subscriber with only empty snapshots is treated as offline and the state stays.
- Subscriber card, "DNS for 24 h" section: grey context rows for ArgusDNS anomalies (reflected DNS responses on :53, excessive query rate) with time, duration and servers. This is context, not evidence: the subscriber state does not change.
- Periods when the DNS server did not answer or had no data are visible in the subscriber's DNS summary. Source silence is no longer read as subscriber silence.
- Resilience to ArgusDNS outages: a connection failure is marked as "source error" rather than "subscriber not seen in DNS".
- An infection state that rests only on DNS evidence is not cleared automatically while the DNS source is unavailable, and for a while after it recovers.
- Muting Telegram notifications by event kind. Bulk changes and route withdrawals are muted by default: they stay in the journal and the panel but do not reach the chat.
- An infection state is not cleared automatically while its evidence comes from a new rule that is still maturing, or right after a quarantine relapse.
Versions 0.1–0.58.33 are described in the Ukrainian changelog. Open the full history
Agent argus-probe
The agent for Linux, FreeBSD and Windows (also Docker / Kubernetes and a MikroTik RouterOS container) updates itself as soon as a version is published. Agent version requirements for center features are listed in the center entries.
- More reliable updates on Windows: the updater service is restarted through the Windows Service Control Manager, and the agent checks that it came up on the new version.
- The installer over an existing agent stops the services and the tray cleanly before replacing files. If a file is held by another program, the installer names it and brings the agent back on the previous version.
- The installer window no longer closes right after success or an error, so the result can be read.
- Leftover files from earlier versions are cleaned up automatically after an update.
- New "game query" check: the agent queries the game server itself and returns the server name, game, map and player count. Supported protocols: A2S (Source), SA-MP, Minecraft, RakNet, FiveM and Quake 3; the protocol is picked by port. Requires center 0.58.86+.
- Manual check from the console:
argus-probe check game. - Windows: online installer for the 2.x line. It downloads the current version from the center, verifies its SHA-256 and signature, then installs the agent or updates an existing one.
- The tray icon shows the installed, available and target versions. The "Update" button appears only when the center allows that version; the tray never updates above the center's target.
- Agent status window (double-click the icon) with the latest log lines. The token and keys are never shown there.
- The agent and updater services watch each other: a hung or stopped service is restarted automatically.
- The Windows agent now shares one codebase and one version number with the Linux and FreeBSD agent. It installs over 0.2.x and 0.3.x and carries over data and registration.
- Windows: Windows services, Windows Firewall rules, a tray icon and Start menu entries. The flow and BGP modules are not available on Windows.
- The agent never installs a version lower than the current one without explicit permission from the center.
- Ping checks honour the check interval set in the center. A target with a 60 s interval is pinged once a minute rather than every cycle, which cuts packets and stored rows several times over.
- The agent status shows the update state: pending, blocked or failed, with the reason.
- Flow: anonymised count of how many of your subscribers reach game hosts. A10 CGNAT: sessions by destination address for the same count.
- A10 CGNAT: a new operation that moves a subscriber back onto the NAT pool of their profile; NAT pool name in responses.
- Stopping or updating the agent no longer cuts an A10 action in the middle: the running action completes and no new ones start.
- A10 commands carry an expiry time: after coming back online the agent will not run a command that is no longer relevant.
- The agent log never contains passwords, keys or command payloads.
- An agent running as the main process of a container (PID 1) no longer accumulates zombie processes after trace timeouts.
- The agent status shows process count, limit and zombies; the center warns when the agent hits the process limit.
- The official Docker image includes an init process.
- The flow module counts hourly "subscriber ↔ peer" pairs for addresses under manual watch (requires center 0.58.31+).
- A10 CGNAT: batch profile (LID) change for up to 20 addresses in one operation, with a per-address result check.
- A10 CGNAT: support for the DNS-redirect exemption profile (LID 7).
- The agent never touches class-list entries it did not create.
- The agent-managed firewall detects the actual SSH ports (sshd config, systemd
ssh.socket, open sockets) and rechecks them every minute, so changing the SSH port never locks you out of the server. - SSH ports are shown in the probe firewall status in the panel.