TarleyNode 0.4.0
Released 8 November 2026
Added
- PostgreSQL checks also read active sessions, transactions per second, the cache hit ratio, deadlocks, blocked sessions, the longest running query and the database's size. SQL Server checks also read active requests, batch requests per second, page life expectancy, the buffer cache hit ratio, blocked sessions, deadlocks, the longest running request and the database's size. The check's summary shows them, and query text is never in the readings or history. A reading the login may not take is left out and the check still passes. PostgreSQL logins need the pg_monitor role and SQL Server logins need VIEW SERVER STATE to read everything. See docs/first-monitor.md.
- Database activity now also opens on PostgreSQL and SQL Server checks, with the same privacy rules as MySQL: running queries, top statements, space by database and the largest tables. Each part says what the login needs when it cannot read it, including pg_stat_statements on PostgreSQL. A connected site on an older version is told to update.
- PostgreSQL, SQL Server, MySQL and MariaDB checks say whether the server is a primary or a secondary, and how far behind the main server a secondary is, in the check's summary and on its page: Primary, Secondary, 12 seconds behind main, or Secondary, replication stopped. The readings are is_replica, replicating and replication_lag_seconds. Lag never reads as 0 when replication has stopped or the lag is unknown. Log shipping is read if the login may, and SQL Server database mirroring is not. MySQL logins need REPLICATION CLIENT and MariaDB logins need SLAVE MONITOR; without it the role shows as not readable and the check still passes. A PostgreSQL standby that only restores from a WAL archive shows lag unknown and never fails. A MySQL or MariaDB server with read_only off reads as Primary only when both replication threads are stopped on every channel with no error, as a server promoted without RESET REPLICA ALL looks the same; set read_only on secondaries to avoid it.
- Replication that stopped on a secondary now fails the check and raises a problem by the check's rule, after three failed readings by default, without touching primaries. A rule on replication_lag_seconds alerts on lag, and also opens when replication stops. This includes existing MySQL and MariaDB checks whose login may read the replication status: after updating, a read-only secondary stopped on purpose raises a problem, until you pause the check or turn its rule off.
- Connected sites need this version to take these readings and to answer Database activity for PostgreSQL and SQL Server. A site updated before its server keeps reporting.
- A database sensor's page has a switch above the graph, with the same time range. Lines stays the default. Overview shows the role and lag, a gauge for the cache hit ratio and for the fullest tablespace, and current sessions, throughput, blocked sessions, longest query, database size and page life with a small graph and the change over the range. Activity shows sessions and throughput as filled areas, and deadlocks, slow queries and failed connections as bars of how many in each period, worked out from each step's average and the check's interval and marked as approximate when a step holds several readings. Heatmap shows query and connect time by hour of day and day, hourly whatever the range, and shows a week when the range is under three days. PNG saves the view shown. A reading the database does not report is left out.
- The Network map now always shows every device, grouped by site and subnet, even with no LLDP, no network equipment or no SNMP at all. LLDP and CDP links are still solid and exact, and Dependencies are dashed as before. Other links are dashed best guesses and say why when you hover: a switch port that learned the device's address, its default gateway read over SNMP, the most common gateway of its subnet, the subnet's one router or firewall, or the one device whose route leaves the site's ranges, as that is most likely its router. With nothing to go on, devices hang from a labelled group for their subnet, or for no address. The Devices without links list is gone, as they are all on the map. Maps saved before this keep their look; a map saved from now on keeps the guesses and shows new devices. With more than 2000 devices the map says it shows the first 2000. Connected sites need this version to read gateways; a site or server on an older version keeps working without them. Linux servers set up with snmpd need the two new view lines from docs/linux-snmpd.conf (.1.3.6.1.2.1.4.21 and .1.3.6.1.2.1.4.24) to report a gateway. Every device on the map shows its name, or its address when it has none, spaced so typical host names do not touch, and a subnet's label sits above it. See docs/feature-scope.md.
Fixed
- Adding the same check to a site twice from Devices › Add checks to site no longer gives every device a second copy. A device that already has the same check (same kind, target, service, process or OIDs), however it was added, is skipped, and the review says how many will get it.
- A device that does not answer SNMP no longer reads as one with nothing on it. "Rescan for changes" now says "Could not read the device", the collector logs why, and the device keeps its interfaces, sensors, neighbors and profile. Before, the rescan said "No changes", and the next read listed every sensor as no longer reported and cleared the profile, every 5 minutes. A site still on 0.3.9 does this until it is updated. The site's regular SNMP reads and Rescan also ask a device twice before counting it as not answering.
- With custom branding, emails, text messages, chat messages and notifications now use the product name, including report emails and their attachment names, the name inside PDF reports, and the alert a cluster sends while no server is active. Licences without custom branding keep TarleyNode, as the console does. Ticket and thread keys are unchanged, so open tickets and threads carry on.
- A server removed from a high-availability cluster no longer stays in the cluster status with its last version. The next rolling update waited 45 minutes for it to install the release, then stopped with a message about disk space and left the servers on mixed versions. A rolling update also waits for every standby there was when it started, so one whose cluster agent restarts while it installs no longer lets the handover start early.
- A planned handover between high-availability servers no longer opens a critical "site is not reporting" incident for the server that handed over, which also alerted people. A server that really fails still opens it. That includes the rolling update to this version, which the older release hands over: the server taking over marks the old site itself.
- A "site is not reporting" incident now also closes when its site is revoked, as a restore does to the other servers' sites, or when the site was stood down and its devices have moved to another site. `monitorctl doctor` no longer counts such a site as stale.
- An update with less than 10 GB of free disk now stops before it starts and says how much is free, in Settings › Software updates and on the command line. Before, it stopped at the package install after the backup, leaving the package half-configured.
- The steps for going back to an earlier release in docs/upgrading.md now work as written, with commands to paste. The old steps could not work: the old package refuses a newer database, and `tarleynode restore` does not take the folder an update backs up to.
- A queue that can no longer write to disk now shows: the server reports not ready and `monitorctl doctor` says what is wrong and fails, instead of both reporting ready while results wait on the connected sites. Restart the queue to recover.
- A NUL or invalid UTF-8 in a request's path or query now gets a 400 instead of a 503 that looked like the database being down.
- Importing a MIB file whose names all resolve no longer blanks the Monitoring tab in Settings for good. A file imported that way before now opens normally and can be removed.
- Discovery adds the SNMP login only to devices that answered SNMP; rescan or assign it on the device page otherwise.
- A device a scan found answering SNMP on another port, such as 1161, is now monitored on that port. Before, its SNMP checks, the sensors, ports and template checks that follow them, and its neighbor scan used 161 and failed. Choosing another SNMP login when monitoring its ports also keeps its port.
- A Linux server a scan found with port 22 closed no longer gets the site's SSH login and a failing Linux health check when added, nor a Windows server with WinRM's port 5985 closed the Windows login and a failing Windows health check. A server the scan did not check that port on still gets it; assign it on the device page otherwise.
- A person in an access group sees only their own pages in the menu straight after signing in, including with two-step and directory (LDAP) sign-in. Before, the full menu showed until they reloaded; the server still refused the data.
- Deleting an event rule closes the incidents it left open and cancels their pending alerts. Incidents left open by rules deleted before this update are closed on upgrade.
- The public status page counts only the time a service has been monitored. Days before that show as no data, so a new service no longer shows 100% over 90 days. The page also follows the dark theme.
- On phones and narrow screens, pages no longer scroll sideways: the top bar shows only the page name, and button rows, time ranges, card notes and pagination wrap. Sign out, Getting started and Send feedback stay in the side rail at 900 px wide or less, and on short screens.
- The Audit trail button shows only to administrators, who are the only ones who can open it. Operators and viewers got "Page unavailable".
- A change to something that no longer exists now says Not found instead of saying it worked, and adds nothing to the audit trail. Removing a device that was already removed in another tab just closes the dialog. Removing a device records its name, and the reason when one is given. High-availability settings say "Enable high availability first" when there is no cluster.
- On a server installed without the host agent, the Network, Server disk, Monitor this server, Software updates and Backups cards now say the agent never reported, or when it last did, instead of waiting forever or showing a wrong message.
- Two logins can no longer share a name, whatever the capitals or spaces at the ends. Installations that already have two keep them, and each can still be saved under its own name.
- The console answers HEAD requests like GET, so uptime monitors and load balancers that check with HEAD no longer see it as down.
- Pause delivery asks for confirmation first, and says alerts wait and are sent once delivery is turned back on.
- Software updates shows "Not checked yet" instead of "Up to date" before the first check.
- A password reset link opened in a tab already on the sign-in page now shows the reset form.
- The sign-in page and, for access group members, the Devices page no longer log errors in the browser for requests they did not need.
- The Rules, Maintenance and Dependencies pages show their empty state instead of a blank page when the server sends no list.
- The status page address, event log and source, and webhook header name fields check what you type again in current browsers.
- Graph axes no longer show the same number twice, Max and Min labels stay inside the chart, and units such as Mbit/s no longer wrap.
- The installer now also fetches the backup and restore images, so backups and restores later work without internet access.
- The forgot-password help names Settings › People & sign-in, and the how-tos say that problems during a maintenance window still open incidents, marked Suppressed; only the alerts are held back.
- A site's scans, scheduled ones included, can ask SNMP on another UDP port than 161: set it under Discovery › Logins for the site. Before, scheduled scans always asked on 161, so a device that answers SNMP only on another port was added with Ping only, and its next scheduled scan replaced what a custom scan had found.
- Testing desktop alerts when no browser has them turned on now answers 409 "no_browsers" and says to turn them on in a browser first; a push service that refuses the test still answers 502. Single sign-on start and callback, and SAML start and the SAML reply address, answer 404 "sso_not_configured" when that sign-in is not set up, and 503 only for real failures, such as an identity provider that cannot be reached, so monitors and proxies no longer read the server as unhealthy.
Security
- The HTTPS gateway moves to Caddy 2.11.9, built with Go 1.27.3, which fixes security vulnerabilities in the earlier version.
- NATS moves to 2.14.9 and VictoriaMetrics, vmbackup and vmrestore to 1.155.0, which fix security vulnerabilities in their earlier versions.
- The PostgreSQL 16 and Debian 12 images are rebuilt with current system packages.
- Sign-in and compression libraries are updated: go-jose 4.1.9 and compress 1.20.3.
- The customer email and commercial fields of the license are for administrators only, by session or API key; the banner's own message still reaches everyone.
- SAML sign-in refuses an assertion in which an element or XML attribute it reads, such as the subject name, an attribute and its name, a subject confirmation or a condition, is outside the SAML namespace, or in which the subject name, issuer, subject or conditions appear more than once (GO-2026-6909). Such an element could otherwise change the name, the groups that set the role, or the ID that stops a replay.
- Responses carry a Permissions-Policy header that turns off the camera, microphone, location, USB and payment features, none of which TarleyNode uses, and the HTTPS gateway no longer sends a Via header.
- Database activity masks passwords in statement text, such as in CREATE ROLE, IDENTIFIED BY and connection addresses, before it leaves the collector. Text from a connected site is masked once that site is updated.