Skip to content


This collector polls resource usage by Harvest pollers on the local system. Collector might be extended in the future to monitor any local or remote process.

Target System

The machine where Harvest is running ("localhost").


Collector requires any OS where the proc-filesystem is available. If you are a developer, you are welcome to add support for other platforms. Currently, supported platforms includes most Unix/Unix-like systems:

  • Android / Termux
  • DragonFly BSD
  • FreeBSD
  • Linux
  • NetBSD
  • Plan9
  • Solaris

(On FreeBSD and NetBSD the proc-filesystem needs to be manually mounted).


parameter type description default
mount_point string, optional path to the proc filesystem `/proc


The Collector follows the Linux proc(5) manual to parse a static set of metrics. Unless otherwise stated, the metric has a scalar value:

metric type unit description
start_time counter, float64 seconds process uptime
cpu_percent gauge, float64 percent CPU used since last poll
memory_percent gauge, float64 percent Memory used (RSS) since last poll
cpu histogram, float64 seconds CPU used since last poll (system, user, iowait)
memory histogram, uint64 kB Memory used since last poll (rss, vms, swap, etc)
io histogram, uint64
IOs performed by process:
rchar, wchar, read_bytes, write_bytes - read/write IOs
syscr, syscw - syscalls for IO operations
net histogram, uint64 count/byte Different IO operations over network devices
ctx histogram, uint64 count Number of context switched (voluntary, involuntary)
threads counter, uint64 count Number of threads
fds counter, uint64 count Number of file descriptors

Additionally, the collector provides the following instance labels:

label description
poller name of the poller
pid PID of the poller


  • Collector will fail on WSL because some non-critical files, in the proc-filesystem, are not present.