nurse leverages Go's pprof tooling to monitor various system and application metrics. The operator can configure per-resource thresholds which, when surpassed, trigger the gathering of a suite of metrics which are then dumped to disk along with a nurse.log file describing the reason for the dump.
To enable nurse, set the NURSE_TOML_PATH environment variable and create a nurse.toml file with the following schema:
profile-root = "/tmp/nurse" # where the profiles will be dumped
poll-interval = "1m" # frequency with which to check thresholds
gather-duration = "60s" # how long to gather samples when profiling
max-profile-size = "100mb" # maximum size of the profile
cpu-profile-rate = "5" # samples per second
mem-profile-rate = "5" # samples per second
block-profile-rate = "..." # samples per second
mutex-profile-fraction = "..." # fraction of mutex events that are reports (1 / rate)
mem-threshold = "..." # maximum memory usage of the process before a dump is triggered
goroutine-threshold = "..." # maximum number of goroutines before a dump is triggered