Linux System Monitor

A Prometheus exporter that reads host telemetry straight from /proc and /sys. Most agents wrap the kernel in layers; this one parses the raw files the kernel already exposes, runs a poller thread per subsystem, and serves the numbers on a /metrics endpoint.

Single C++17 file, no dependencies beyond libc and pthreads. Builds to one static-ish binary you can drop on any Linux host.

How it works

/proc, /sys files  ->  collectors  ->  poller threads  ->  registry  ->  /metrics

Each poller reads its source on a fixed interval and writes the latest values into a mutex-guarded registry. The HTTP server renders whatever is in the registry and handles each scrape on its own thread, so a scrape never blocks on a file read or on another scrape.

Metrics

Names follow node_exporter where one exists, so dashboards and alerts built for it work here too.

Source	Metrics
/proc/stat	node_cpu_seconds_total, node_context_switches_total, node_boot_time_seconds, node_forks_total, node_procs_running, node_procs_blocked
/proc/meminfo	node_memory_*_bytes
/proc/loadavg	node_load1, node_load5, node_load15
/proc/net/dev	node_network_receive_*_total, node_network_transmit_*_total
/proc/diskstats	node_disk_reads_completed_total, node_disk_read_bytes_total, node_disk_writes_completed_total, node_disk_written_bytes_total, node_disk_io_time_seconds_total
/proc/mounts + statvfs	node_filesystem_size_bytes, node_filesystem_free_bytes, node_filesystem_avail_bytes, node_filesystem_files, node_filesystem_files_free
/sys/class/thermal	node_thermal_zone_temp

Build and run

make
./monitor --listen :9101 --interval 5
curl -s localhost:9101/metrics

Or without make:

c++ -std=c++17 -O2 -pthread -o monitor monitor.cpp

Flags:

• --listen host:port to bind (default :9101)
• --interval seconds between polls (default 5)
• --procfs procfs root (default /proc)
• --sysfs sysfs root (default /sys)

The --procfs and --sysfs flags let you point at a mounted host root from inside a container, or at the test fixtures on a non-Linux machine.

Test

make test        # builds, then runs ./monitor --selftest

The self-test parses the sample files under fixtures/ and checks the parsed numbers and the rendered output. It needs no Linux and no network, so it runs on any dev machine.

Scrape config

scrape_configs:
  - job_name: linux-system-monitor
    static_configs:
      - targets: ["localhost:9101"]

Deploy

Build, copy the binary to /opt/linux-system-monitor, and install the unit:

make
sudo mkdir -p /opt/linux-system-monitor
sudo cp monitor /opt/linux-system-monitor/
sudo cp deploy/linux-system-monitor.service /etc/systemd/system/
sudo systemctl enable --now linux-system-monitor

Notes

• A collector whose source file is missing returns nothing instead of failing, so a wrong kernel or a missing mount just omits that data.
• loop and ram block devices are filtered from disk stats; pseudo filesystems (tmpfs, cgroup, and similar) are filtered from filesystem stats.
• The HTTP server reads one request buffer per connection, which is all a Prometheus GET needs. It is a scrape target, not a general web server.
• All values are doubles, matching Prometheus, which stores every sample as a float64.