Skip to content

FleetPilot v1.8.0-beta.2

Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 18 Oct 12:40
· 66 commits to main since this release
ad979c7

Pre-release for testers. Everything below is new since 1.7.0. Please try it on your own hardware and report what breaks.

Security

  • Fan, disk and CheckMK pages put names and the monitoring token into inline JavaScript as JSON now (|tojson); a test checks every template for unencoded values in scripts.
  • A new test sends a request without signing in to every route, add-on routes included, and fails if anything but the short list of public pages answers.
  • Fixes for CodeQL's security-extended findings:
    • Proxmox, Veeam and Duplicati: API paths now only use names the device itself reported (node, guest type, power action, job, disk). A request can no longer put arbitrary text into the URL; before, the node and guest type from the form went into the Proxmox API path unchecked.
    • Logs: host, device and path names have their line breaks escaped, so they cannot forge log entries.
    • File permissions: the generated public SSH key, the proxy routes and the HAProxy candidate configuration are readable by their owner only.
    • Data sync: the optional sync checks the peer's certificate by default (FLEETPILOT_SYNC_CA for a self-signed peer).
  • VM power actions answered a rejected node, guest type or action with the exception's text; they now return only fixed messages (CodeQL py/stack-trace-exposure).
  • /vm/<id>/disks stored disks without their node, so the disks of a cluster's nodes overwrote each other there too.

Added

  • Translation report: python scripts/missing_translations.py lists untranslated literal template strings and per-language totals without starting the application; --lang checks one language.

  • Windows PCs and servers, every release Microsoft still supports, all over OpenSSH (see WINDOWS_SUPPORT.md):

    • covers Windows 11 up to 26H2, Windows 10/11 LTSC 2019/2021/2024 (including IoT and evaluation editions) and Windows Server 2016, 2019, 2022 and 2025 (Desktop or Core), in any display language
    • Detect OS names the exact release and edition and shows Microsoft's support end date; the host list warns about ended support
    • updates use the built-in Windows Update Agent, run once as SYSTEM through a scheduled task (Windows does not let an SSH session install updates), never restart the machine, and work with WSUS, LTSC and Server Core; a full update also runs winget upgrade where winget exists
    • an installation is never cut off: after two hours FleetPilot stops waiting while Windows carries on, and the next update run (or one after a lost connection) follows the running installation to its result
    • hardware overview and SMART (with smartmontools for Windows) for Windows hosts; Install SSH key puts the key where Windows OpenSSH reads it for administrators
    • all commands are encoded PowerShell, so they work with cmd.exe or PowerShell as the SSH shell
  • Android app (beta), now working with FleetPilot itself:

    • signs in with a personal API token
    • trusts FleetPilot's self-signed certificate after you confirm its fingerprint, and then only that certificate
    • shows the dashboard, servers, VMs, HW monitor and fans
    • wakes and shuts down servers, runs VM power actions and sets fan speeds, within the token's scope and your role
    • the API gains GET /api/vms, POST /api/vms/<id>/action and POST /api/fans/<id>/set_fan, and /api/hosts now reports each host's state
  • Enforced coding standard and strict warnings: ruff (PEP 8, rules in ruff.toml) fails CI on any finding, the test suite treats every warning as an error, and CI measures statement and branch coverage with a floor that only goes up.

  • License headers: every source file states SPDX-License-Identifier: GPL-3.0-or-later and its copyright holder; a test keeps it that way.

  • Project documents for the OpenSSF silver level: a code of conduct (Contributor Covenant 2.1), a roadmap, code review standards in CONTRIBUTING.md, continuity rules in GOVERNANCE.md, and a security assurance case.

  • Pull requests are fuzzed with ClusterFuzzLite (.clusterfuzzlite/, Atheris), focused on the changed code.

  • Signed releases with build provenance: each release now carries the source archive, the SBOM, SHA256SUMS and Sigstore-signed SLSA provenance (gh attestation verify). It is assembled as a draft and published in one step.

  • Hardened CI:

    • Least-privilege permissions per job and no stored checkout credentials.
    • Concurrency limits for every workflow.
    • CodeQL with the security-extended queries, now also for the workflows themselves.
    • A zizmor audit of all workflows.
    • Dependabot groups, a 7-day cooldown, and updates for the fuzzing image.
    • CODEOWNERS.
    • The unused "Notify New Version Available" workflow is removed.
  • Vulnerability scan (Servers → Vulnerabilities): known CVEs in the installed packages of every Linux server (Debian, Ubuntu, Raspberry Pi OS, Proxmox VE/Backup Server, TrueNAS SCALE, Alpine, Rocky Linux, AlmaLinux) from OSV.dev, Docker container images through Trivy when it is installed, pfSense/OPNsense pkg audit, and FleetPilot's own Python packages and server. Shows severity (CVSS), the fixing version and where each vulnerability is; filters; scan one target or all; optional daily scan with notifications about new high/critical findings that have a fix; GET /api/vulnerabilities, Prometheus metrics and a Security center check. Only package names and versions are sent to OSV.dev; nothing in offline mode; FLEETPILOT_OSV_URL for a mirror.

  • Live demo you can try before installing (https://christianhandy.github.io/FleetPilot/): the real FleetPilot runs in the browser with Pyodide against an emulated homelab of Proxmox nodes, TrueNAS, Proxmox Backup Server, Linux and Windows machines, FRITZ!Box/Repeater, OpenWrt and OPNsense. Updates, Wake-on-LAN, VM power, fan control, SMART alerts and router checks all work; a "Virtual lab" panel switches devices off, fails disks or publishes updates. The previous static demo is now the lighter tour at /tour/. See demo/README.md.

  • Releases: pre-release tags (v1.8.0-beta.2) are published by the release workflow too, as pre-releases. Files attached to a draft release before the tag is pushed, such as the Android APK, get checksums and signed provenance like the source archive.

  • Android app in every release: the release workflow builds the APK from the tag, signs it with the release key, checks the signing certificate and attaches it to beta and stable releases, covered by SHA256SUMS and the signed provenance.

Fixed

  • SSH connections leaked when a command or read timed out: the HW monitor's polling, the hardware overview, the live host stats, shutdown and the remote-disk plugin closed the connection only on success, and a failed connect left its socket open. On a slow Windows host FleetPilot held 193 connections until Windows' sshd refused every new one ("Exceeded MaxStartups"). Every path now closes the connection.
  • An SSH server that closes the connection before greeting is reported as overloaded, out of memory or with too many logins waiting, not as paramiko's "Error reading SSH protocol banner". WINDOWS_SUPPORT.md explains how to recognise and clear an sshd that has too many connections waiting. A Windows Update that Windows ends by restarting (task result 0x41306) now says so.
  • Windows Update no longer stops with "The process cannot access the file … update.log" when the log is being written while FleetPilot reads it (seen after an hour of installing on Server 2025).
  • SSH connections send keepalives every 30 seconds, so a long Windows Update whose host stops answering (a firewall that starts dropping packets) ends with an error instead of waiting forever.
  • Windows hosts were detected as Linux, could not be shut down (FleetPilot sent the Linux shutdown -h now), and updates failed when the SSH shell was PowerShell or the PowerShell Gallery was unreachable.
  • Editing a host dropped every field the form does not show (custom image, reachability state, detected OS) and, when it was renamed, its last-update and last-seen times.
  • HW Monitor never collected live metrics: the collector script sent to each host had broken indentation since an earlier clean-up, so it failed on every host and every server showed as unreachable. A test now runs the script.
  • The hardware stress test kept going after the GPU dropped off the bus: the GPU and combined phases set a local running flag instead of the global one.
  • SQLite connections opened with with get_db() as conn: were committed but never closed, leaking file handles until garbage collection; they are now closed at the end of the block.
  • A test of the network allowlist's lockout guard never checked its result; it does now.
  • SMART data from Proxmox nodes was never read: FleetPilot asked the API for an empty disk name instead of the devpath Proxmox reports.
  • In a Proxmox cluster, disks with the same name on different nodes (every node's sda) overwrote each other in the SMART registry; they are now stored as node:sda.
  • Fan controllers (lm-sensors) showed the sensor limits (temp1_max, temp1_crit) as extra temperature readings; only measured values are shown now.
  • HW Monitor showed German labels ("Netz", "CPU Last") in the English interface.
  • Ping button (example plugin): the round-trip time started before the port 80 attempt, so a host without a web server was charged that attempt's full timeout even when SSH answered in a millisecond; every attempt is timed on its own and the time of the one that answered is reported. A refused connection also counts as an answer (the host is up, only the port is closed), so a host with neither port 80 nor port 22 open is no longer shown as "Timeout". The button shows 3 ms (22) or the reason a host did not answer (timeout, dns, unreachable) instead of always "Timeout".

Install or update: see the installation guide.

Verify a download: gh attestation verify fleetpilot-1.8.0-beta.2.tar.gz --repo ChristianHandy/FleetPilot