nbari / pg_exporter / 29278675245

push

github

add --collector.system host CPU/memory collector (#26)

Opt-in, Linux/FreeBSD-only collector exposing node_exporter-style host
metrics for the machine running the exporter. CPU time is a single
per-core counter, pg_system_cpu_seconds_total{cpu,mode} (mirroring
node_cpu_seconds_total), alongside load average and memory/swap byte
gauges. It reads only the OS (/proc/stat on Linux, the kern.cp_times
sysctl on FreeBSD, sysinfo for memory/load) and never queries
PostgreSQL, so it adds zero query or connection load.

- Per-core CPU by default; aggregate host utilization is derived in
  PromQL (sum without(cpu) ...). This gives per-core hotspot visibility
  while cardinality stays bounded per host (modes x cores) and does not
  scale with the number of databases.
- Warns at startup when enabled against a non-loopback DSN (co-located
  only; misleading on managed services such as RDS/Aurora).
- Adds a collapsed "Host CPU / Memory" Grafana dashboard row, the
  collector README, a main-README entry, a CHANGELOG [0.17.0] entry, and
  --collector.system to the "just watch" recipe.
- Makes the dashboard "Memory & Swap Usage" panel easier to read: total
  memory renders as a dashed reference line while used-memory and swap
  fills are no longer merged into one stacked gradient.
- Refreshes Cargo.lock (mio, rustls, zmij patch updates) via cargo update.
- Clarifies the "Exporter Process CPU Usage" panel: the solid line is
  relabeled "CPU % of one core" (100% = one core, can exceed 100%) and
  the dashed line "% of whole host (normalized)", removing the ambiguous
  "per-core" wording, with a rewritten description.

641 of 681 new or added lines in 6 files covered. (94.13%)

16413 of 18114 relevant lines covered (90.61%)

136.08 hits per line