Minikube: status: show guest VM health (cpu/ram/disk)

[tstromberg]

The status command should make it obvious when one is out of guest VM resources - particularly disk space.

[afbjorklund]

You can use ssh, to run some simple ad-hoc Linux commands to do this monitoring...

$ minikube ssh -- top -b -n 1 | head -n 4
top - 20:47:27 up  1:01,  1 user,  load average: 0.00, 0.01, 0.00
Tasks: 142 total,   1 running, 141 sleeping,   0 stopped,   0 zombie
%Cpu0  :   0.0/6.2     6[||||||                                                                                              ]
%Cpu1  :   0.0/0.0     0[                                                                                                    ]
$ minikube ssh -- free -m
              total        used        free      shared  buff/cache   available
Mem:           1990          72        1127          16         790        1871
Swap:           999           0         999
$ minikube ssh -- df -h /var/lib/docker
Filesystem      Size  Used Avail Use% Mounted on
/dev/sda1        17G   49M   16G   1% /var/lib/docker

Nothing wrong with having the basic information available with a simple command, though.

[afbjorklund]

Maybe this will muddy the "status" command output, and better off as a separate cmd (or extra flag) ?

This library looked quite useful, for making a health check binary that can be deployed on the VM:

https://github.com/shirou/gopsutil

Strange that there is no built-in monitoring of node disk usage, only cpu and ram ? (cAdvisor has it)

https://github.com/google/cadvisor/tree/master/deploy/kubernetes

[fejta-bot]

Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle stale

[afbjorklund]

/remove-lifecycle stale

[afbjorklund]

Here's a little hack to show the available information:

package main

import (
    "fmt"
    "os"
    "time"

    "github.com/shirou/gopsutil/cpu"
    "github.com/shirou/gopsutil/disk"
    "github.com/shirou/gopsutil/mem"
)

func busy(percent []float64) float64 {
    return percent[0] // single
}

func idle(percent []float64) float64 {
    return 100.0 - busy(percent)
}

func megs(bytes uint64) float64 {
    return float64(bytes) / 1024.0 / 1024.0
}

func main() {
    interval := time.Duration(1) * time.Second
    path := "/var/lib/minikube"
    if _, err := os.Stat(path); os.IsNotExist(err) {
        path = "/"
    }

    i, _ := cpu.Info()
    p, _ := cpu.Percent(interval, false)
    v, _ := mem.VirtualMemory()
    s, _ := mem.SwapMemory()
    d, _ := disk.Usage(path)

    fmt.Printf("CPU\tNumber:%d, Idle:%.1f%%, Busy:%.1f%%\n",
        len(i), idle(p), busy(p))
    fmt.Printf("MEM\tTotal:%.0f, Available:%.0f, Used:%.1f%%\n",
        megs(v.Total), megs(v.Available), v.UsedPercent)
    fmt.Printf("SWP\tTotal:%.0f, Free:%.0f, Used:%.1f%%\n",
        megs(s.Total), megs(s.Free), s.UsedPercent)
    fmt.Printf("HDD\tTotal:%.0f, Free:%.0f, Used:%.1f%%\n",
        megs(d.Total), megs(d.Free), d.UsedPercent)
}

Typical output:

$ ./health 
CPU Number:2, Idle:95.2%, Busy:4.8%
MEM Total:1991, Available:1285, Used:24.9%
SWP Total:0, Free:0, Used:0.0%
HDD Total:17368, Free:14847, Used:9.2%

[tstromberg]

I noticed today that kubectl describe node shows health data that almost certainly has an API call:

Conditions:
  Type             Status  LastHeartbeatTime                 LastTransitionTime                Reason       
                Message
  ----             ------  -----------------                 ------------------                ------       
                -------
  MemoryPressure   False   Fri, 20 Sep 2019 13:38:48 -0700   Fri, 20 Sep 2019 11:47:38 -0700   KubeletHasSuf
ficientMemory   kubelet has sufficient memory available
  DiskPressure     False   Fri, 20 Sep 2019 13:38:48 -0700   Fri, 20 Sep 2019 11:47:38 -0700   KubeletHasNoD
iskPressure     kubelet has no disk pressure
  PIDPressure      False   Fri, 20 Sep 2019 13:38:48 -0700   Fri, 20 Sep 2019 11:47:38 -0700   KubeletHasSuf
ficientPID      kubelet has sufficient PID available
  Ready            True    Fri, 20 Sep 2019 13:38:48 -0700   Fri, 20 Sep 2019 11:50:07 -0700   KubeletReady 
                kubelet is posting ready status

[tstromberg]

This issue is still relevent in minikube v1.6. It'd be nice if someone plumbed kubectl describe get node calls into minikube status. help wanted!

[prasadkatti]

This issue is still relevent in minikube v1.6. It'd be nice if someone plumbed kubectl describe get node calls into minikube status. help wanted!

@tstromberg - kubectl desc node -A is now part of minikube logs, merged in #7105

[afbjorklund]

We can get some of this from the metrics-server, by using kubectl top node.

But it would be nice to be able to get the "raw" data from the VM as well ?
Using SSH could be a fallback, but it should be possible to ask hypervisor.

http://www.virtualbox.org/manual/ch08.html#vboxmanage-metrics
VBoxManage metrics query minikube

Some available:

Object          Metric                                   Values
--------------- ---------------------------------------- --------------------------------------------
minikube        Guest/CPU/Load/User                      8.00%
minikube        Guest/CPU/Load/Kernel                    4.00%
minikube        Guest/CPU/Load/Idle                      86.00%
minikube        Guest/RAM/Usage/Total                    5954124 kB
minikube        Guest/RAM/Usage/Free                     5166640 kB
minikube        Guest/RAM/Usage/Cache                    1079048 kB

When using KIC, we need to make sure to not display the data for the host.
So we need to ask the container runtime used, what applies to the container.

$ docker stats docker --no-stream --format '{{json .}}'
{"BlockIO":"245MB / 0B","CPUPerc":"23.86%","Container":"minikube","ID":"6c28dceb5e35","MemPerc":"9.91%","MemUsage":"792.8MiB / 7.812GiB","Name":"docker","NetIO":"92.1kB / 297kB","PIDs":"455"}

[mr-bjerre]

What to do when running out of space?

> minikube ssh -- df -h /var/lib/docker

Filesystem      Size  Used Avail Use% Mounted on
/dev/vda1        10G  9.9G     0 100% /var/lib/docker

I know I can run minikube delete of course but it is really time consuming to rebuild everything.

[tstromberg]

@mr-bjerre - if you are on Windows/macOS, try giving Docker more space:

https://docs.docker.com/docker-for-mac/space/

[mr-bjerre]

Skaffold more space, right?

What I usually do is

minikube ssh - docker system prune

[afbjorklund]

I added some parsers for free -m and df -m output, for showing the total available.
They _could_ be used for getting the "available" and "/var/lib/docker" (or containers) too

https://github.com/kubernetes/minikube/commit/41979352b927b85e12574cb3bbbf4917ef1a06a0

The CPU usage is trickier, since /proc/stat only has counters (reason for "sleep" above)
Probably easier to just call some existing program, I do believe that most have "vmstat"

vmstat 1 2

  The first report produced gives averages since the last reboot.   Addi‐
  tional  reports  give information on a sampling period of length delay.
  The process and memory reports are instantaneous in either case.

procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa st
 1  0      0 3501344  12540 1763764    0    0    30  7024 1491 2562  6  4 89  0  0
 2  0      0 3500864  12540 1763764    0    0     0    48 2417 4753  5  3 92  0  0

5% user, 3% system, 92% idle

[afbjorklund]

Added implementation of a vmstat parser as well, in 4c78a65b8

        count := runtime.NumCPU()
        busy, idle, _ := util.LocalCPU()
        fmt.Printf("Local CPU: #%d %d%% %d%%\n", count, busy, idle)

        rr, _ := cr.RunCmd(exec.Command("nproc"))
        count, _ = strconv.Atoi(strings.TrimSpace(rr.Stdout.String()))
        rr, _ = cr.RunCmd(exec.Command("vmstat", "1", "2"))
        busy, idle, _ = util.ParseVMStat(rr.Stdout.String())
        fmt.Printf("Remote CPU: #%d %d%% %d%%\n", count, busy, idle)

Since it is based on integers, the numbers don't always add up:

Local CPU: #8 7% 92%
Remote CPU: #2 4% 97%

The local implementation just uses gopsutil, per above: 03d950feb

Put the code in "generic-status" for now, needs better status cmd...