log_cpu_mem_in_docker.sh

#!/bin/bash

# Running inside a docker container, periodically read the container's CPU/memory usage counters
# and log them to standard error. Fields:
#
#     cpu_pct        average user %CPU usage over the most recent period
#     mem_MiB        container's current RSS (excludes file cache), in mebibytes (= 2**20 bytes)
#     cpu_total_s    container's user CPU time consumption since this script started, in seconds
#     elapsed_s      wall time elapsed since this script started, in seconds
#
#
# YMMV: host OS version & configuration may affect availability of the needed counters under
#       /sys/fs/cgroup

log_cpu_mem_in_docker() {
    PERIOD_SECS=${1:-10}  # logging period (default 10s)
    JIFFIES_PER_SEC=100   # see http://man7.org/linux/man-pages/man7/time.7.html
    T_0=$(date +%s)

    cpu_user_jiffies() {
        cut -f2 -d ' ' /sys/fs/cgroup/cpuacct/cpuacct.stat | head -n 1
    }
    user_jiffies_0=$(cpu_user_jiffies)
    user_jffies_last=$user_jiffies_0
    t_last=$T_0

    while true; do
        sleep "$PERIOD_SECS"
        t=$(date +%s)
        wall_secs=$(( t - T_0 ))

        user_jiffies=$(cpu_user_jiffies)
        user_pct=$(( 100*(user_jiffies - user_jffies_last)/JIFFIES_PER_SEC/(t - t_last) ))
        user_secs=$(( (user_jiffies - user_jiffies_0)/ JIFFIES_PER_SEC ))

        user_jffies_last=$user_jiffies
        t_last=$t

        rss_bytes=$(awk -F ' ' '$1 == "total_rss" { print $2 }'  /sys/fs/cgroup/memory/memory.stat)

        >&2 echo "container_usage :: cpu_pct: ${user_pct}, mem_MiB: $(( rss_bytes/1048576 )), cpu_total_s: ${user_secs}, elapsed_s: ${wall_secs}"
    done
}

log_cpu_mem_in_docker "$1"