Proposal:

Periodically (by default once per minute, but configurable) check for breaches of a list of kernel limits. This would be a linux-only feature (at least initially), and checks on other operating systems would be no-ops.

Log a warning if within 30% of OS limits.

Current behavior:
Users run out of file handles / mmaps with no warning

Desired behavior:
Log a warning if within 30% of OS limits.

Unix commands for checking these are given below - these unix commands should be translated to go-specific idioms.

File Handles

Check cat /proc/sys/fs/file-nr - this gives 3 numbers. The first is the total file handles on the system, and the last is the maximum.

Check cat /proc/$(pidof influxd)/limits for the per-process file handle limit, and ls /proc/$(pidof influxd)/fd | wc -l for the number of file handles the process owns.

MMAP regions

Check cat /proc/sys/vm/max_map_count for the per-process limit on memory maps, and cat /proc/$(pidof influxd)/maps | wc -l for the total number of maps used by influxd.

Alternatives considered:
N/A

Use case:
Often users run out of file handles or mmaps, both of which are limited in various places within the kernel, and the error messages are not good.