firepot 1.3.0

Added

• FIREPOT_PROBE_TIMEOUT sets the time limit for filesystem probes in seconds (default 10; a negative value is a usage error), so a stuck mount can no longer hang a run.

Fixed

• Fleet mode accepts remote reports that contain info checks. sessions is always info and swap is info without swap, so a --host run with the default check set now aggregates every node instead of marking each one as an invalid report.
• A remote that runs but exits with a code other than 0, 1, or 2 is reported as remote run failed with exit N; ssh failed is kept for ssh's own exit 255.
• Text-mode fleet output prints each failing check's summary decoded from JSON, so quotes and backslashes come through intact.
• Remote reports stream straight into the 1 MiB-bounded validator rather than being buffered in a shell variable first. ServerAliveInterval (the --ssh-timeout value) with three retries ends dead connections.
• df runs under a deadline (10 seconds, FIREPOT_PROBE_TIMEOUT), so a hung mount becomes a critical result instead of stalling the run. The default filesystem list uses df -l, matching the documented local-filesystem scope.
• Lists given to --services, --peers, --mounts, and the other list options are never glob-expanded against the working directory, and empty items are dropped.
• Load per core divides by online CPUs from /sys/devices/system/cpu/online. nproc (the fallback) follows the caller's CPU affinity and OMP_NUM_THREADS, which do not bound the system-wide load average.
• The gateway check follows the first default route: multipath next hops are pinged correctly, and a point-to-point default route (WireGuard, PPP), which has no next hop to ping, is reported as info.
• Mount points containing spaces or = keep their full names in the disk and inode metrics.
• String metrics (hostname, kernel, name, address, gateway) are always JSON strings, so every hostname, including all-digit names such as 007, yields valid JSON.
• Port lists accept leading zeros (022 matches port 22), reject numbers above 65535 as usage errors, and work with ss builds that lack -H.
• An empty /proc/loadavg or an incomplete /proc/meminfo is reported as a skipped check. Any unexpected runtime failure exits 3, as documented, rather than with a status that reads as a health state.
• Arguments after -- are rejected as usage errors instead of being ignored.

Changed

• A full local check starts 12 processes instead of 559 and finishes in about a tenth of the time (median 76 ms instead of 886 ms on the reference host). JSON escaping, thresholds, and percentages are computed in bash; the hostname, kernel release, CPU count, uptime, and memory figures are read from /proc and /sys; and each probe's output is parsed without awk, sed, sort, or tr.
• Configured peers are pinged in parallel, so unreachable peers cost one ping timeout in total rather than one each.
• Configured systemd units are checked with one systemctl is-active call, falling back to one call per unit if systemctl cannot answer for the whole list.
• The script runs with LC_ALL=C, so probe output parses the same way in every locale.
• Long options and config keys share one table, and fleet mode forwards the effective settings, whether they came from flags or --config.
• awk is no longer required.
• The deprecated node-healthcheck name keeps working; its stderr note now says it will be removed in a future release.

Full Changelog: v1.2.0...v1.3.0