Release LoadGen Insight Agent: 1.0.0.12432

Linux agents only: every change in this release applies to the Linux agent. Windows agents are unchanged.

New Features

• Package Inventory and Outdated Package Reporting - Linux agents now report the software packages installed on each host and which of them the host's own package manager considers upgradable, published as the new package_inventory and package_summary measurements. Debian and Ubuntu (dpkg and apt), Red Hat family (rpm and dnf) and Alpine (apk) package managers are supported, and each source is collected independently so one failing source does not lose the others. package_inventory carries one entry per package with its installed version, available version and an outdated flag; a package is reported when it is first seen, when its version or update status changes and when it is removed, and the full inventory is resent once a day. package_summary carries the per-host counts of installed and outdated packages and is sent on every collection, so a host without a summary is one the agent could not inventory. Collection is on by default and runs every 12 hours. The agent reports inventory only: no scanner and no vulnerability database is installed on monitored hosts.

Enhancements

• Linux agents are now built with Go 1.26, keeping the agent runtime on a release line that still receives security fixes
• Hardened the self-update path: update archives are validated before extraction, only regular files are accepted for the staged archive and the extracted binaries, and leftover temporary files and abandoned archives from earlier update attempts are cleaned up automatically
• Hardened the permissions of the local socket used between the LoadGen Control and LoadGen Insight services

Bug Fixes

• Fixed the Alpine Package Installing No Services - The Alpine package shipped the agent binaries without service definitions, so an Alpine installation left nothing running. It now installs OpenRC service definitions for both the LoadGen Control and LoadGen Insight services, registers them and starts them when the host is already enrolled; on a host that is not yet enrolled the installer prints the enrollment commands instead. Both services run as root under OpenRC supervision and are restarted automatically if they stop, uninstalling removes them cleanly, and agent self-update works on Alpine hosts.
• Fixed a Network Statistics Query Stalling All Metrics - Collecting per-connection TCP statistics could wait indefinitely for a reply from the kernel, which stalled every measurement the agent produces, not just the network ones. The query is now time-bounded and, if it fails, the affected connection counters are reported as zero while all other metrics continue.
• Fixed a Malformed Machine Group Configuration Blocking Updates - A machine group configuration the agent could not parse also stopped it from processing a pending agent update, which is exactly the situation where pushing a fixed version is the remedy. The agent now keeps its previous configuration, logs the problem and still applies pending updates.
• Fixed Self-Update Never Retrying After an Interrupted Hand-Off - When an update hand-off did not complete, for example on a manually installed host without the update trigger, the agent never tried again until the service was restarted. It now retries after 30 minutes, and immediately when the Platform points at a different version.
• Fixed Silent Measurement Loss When No InfluxDB Destination Is Configured - An agent without an InfluxDB destination reported every write as successful while discarding the measurements. Dropped measurements are now counted and a warning is logged periodically. The deliberate back-off after a failed InfluxDB write is also no longer logged as a write failure on every batch; measurements keep spooling to disk and are sent when the destination recovers.
• Fixed Process Filters Being Bypassed - The process include and exclude filters were not applied to process start and stop events, so a machine group monitoring a few processes received lifecycle events for every process on the host, and network connections that could not be attributed to a process were reported even when filters were in force. Both now honour the filters.
• Fixed Missing Process Events When a Process ID Is Reused - When the kernel reused a process ID between two polls, neither the stop of the old process nor the start of the new one was reported, so the series silently continued under the same ID for a different program. The agent now detects the reuse and reports both events.
• Fixed a Newly Enrolled Host Waiting for Its First Inventory - Service and package inventories are now collected as soon as the first configuration arrives instead of at the next scheduled collection, so a newly enrolled host appears with its inventory right away.
• Fixed measurements being lost when a batch failed to encode; they are now re-queued in the same way as after a failed send
• Fixed a batch being lost during a reconnect between the LoadGen Control and LoadGen Insight services, and a stalled connection no longer blocks other senders or the shutdown flush
• Fixed process creation times that could not be read being reported as 1970-01-01; they are now omitted
• Fixed system_specs being written as a near-empty point when hardware enumeration fails, for example in a container where disk enumeration is blocked; failed probes are now logged and the point is skipped only when nothing could be collected
• Startup readiness and service handshake errors are now logged instead of being discarded