docker run "--cpus" flag confusing - docker

I am referring to the --cpus flag of docker run.
The documentation states that the number provided in the flag amounts for the number of CPUs available, which I have found to be untrue.
From the documentation:
--cpus=0.000 Number of CPUs. Number is a fractional number. 0.000 means no limit.
From the limit resources documentation:
--cpus=<value> Specify how much of the available CPU resources a container can use. For instance, if the host machine has two CPUs and you set --cpus="1.5", the container is guaranteed at most one and a half of the CPUs. This is the equivalent of setting --cpu-period="100000" and --cpu-quota="150000". Available in Docker 1.13 and higher.
I have tried the exact above in a docker container with 2 cpus, which resulted in CPU% usage of 100% in docker stats (which should have been 75%). Same result with following declarations:
docker run --cpus 1.5 ...
docker run --cpus=1.5 ...
docker run --cpus="1.5" ...
When I tried setting a value of 0.5:
docker run --cpus 0.5 ...
The CPU resources were limited to 50% in docker stats, which does not comply with the documentation provided by docker, since the docker stats CPU% result should have been 25% (2 CPUs, 0.5 CPU available results in 25% of the CPU capacity)
I am trying to figure out why my perception of the documentation seems to differ from my results, may it be a misinterpretation on my side of the CPU% in docker stats, or wrong documentation provided by docker, which in the latter case, I can report back to Docker.
I am running version Version 18.03.1-ce-mac65 (24312) on my Mac.

docker stat command shows calculation multiplied by CPU kernel count, Ex: if your host has 4 cores and any application utilizing 100% CPU then docker stats output like 400%, in your case it's reporting correct values, as you restricted to 0.5 cores means it 1/2 CORE (50% in 1 core)

Related

Why does docker have more cpus available than the limit I set?

I'm running my container with
docker run --rm -it --cpus=1 --memory "8G" -v "$pwd":"/code/" 'algolab' bash
This means according to the docs that the container will only use one cpu core.
--cpus=0.000 Number of CPUs.
Number is a fractional number. 0.000 means no limit.
But when I run nproc inside the container, it tells me it sees 8 cores - the same number as on my host.
This answer supports that claim. But it also mentions --cpuset-cpus="0-2". When I use that, in addition to --cpus=1, I get the result 3 from nproc.
Why is docker ignoring --cpus=1?
And how do I make it stop ignoring it?
For now, I can use --cpuset-cpus="0", but I don't understand why that's necessary.
$ docker --version
Docker version 19.03.6, build 369ce74a3c
The --cpus option sets a quota on the CPU usage - this does not restrict which CPUs can be used nor how many at a time; this restricts the total slice of time they can be running on your CPU. So if you pass --cpus 1 on a machine with 8 cores, the container will be limited to running on 1/8th of your CPU time. This is covered on a different section of the docs.

How can I make Docker trigger higher CPU frequencies

OK, so my title may not actually be linked to a possible solution, however this is my problem.
I am running a Python 3 Jupyter notebook inside a docker container in from my windows 10 kaby-lake (2 physical cores, 4 virtual cores) laptop.
I noticed while doing heavy computing from there, my CPU usage seen in the task monitor is very low (~15%).
When going on the details for each process, the VBoxHeadless.exe actually uses 24% of the processor, which matches docker stats command which yields 97-100% CPU usage, and therefore makes sense from a single-core operation point of view.
My actual issue is that even though on thread is filled in terms of CPU time, windows (I guess) does not decide that it may actually be useful to speed up the CPU, and therefore it runs at 1.7GHz (with other apps in high performance mode, I usually hit the max 3.5GHz that the computer is capable of).
Therefore, how can I induce the higher clock speeds (nominal 2.7GHz or max 3.5GHZ) (considering that they would probably double my single threaded speed) from docker itself or inside windows 10?
You need to configure the docker machine running docker. If you haven't created a custom one, the default docker machine named 'default' will only have access to one cpu.
You can check all the configuration for this docker-machine by running:
docker-machine inspect default
You need to purge this default machine and recreate it:
docker-machine rm default
docker-machine create -d virtualbox --virtualbox-disk-size "400000" --virtualbox-cpu-count "2" --virtualbox-memory "2048" default
You can check all the avaible configuration options for the machine by running:
docker-machine create --help
Defining CPU Shares can help you but not exactly.
CPU limits are based on shares as these shares are a weight between how much processing time one process should get compared to another. If a CPU is idle, then the process will use all the available resources. If a second process requires the CPU then the available CPU time will be shared based on the weighting.
e.g. The --cpu-shares parameter defines a share between 0-768. If a container defines a share of 768, while another defines a share of 256, the first container will have 50% share while the other one having 25% of the available share total.
Below the first container will be allowed to have 75% of the share. The second container will be limited to 25%.
docker run -d --name p1 --cpuset-cpus 0 --cpu-shares 768 image_name
docker run -d --name p2 --cpuset-cpus 0 --cpu-shares 256 image_name
sleep 5
docker stats --no-stream
docker rm -f p1 p2
It's important to note that a process can have 100% of the share, no matter defined weight, if no other processes are running.

What unit does the docker run "--memory" option expect?

I'd like to constrain the memory of a Docker container to 1 GB. According to the documentation, we can specify the desired memory limit using the --memory option:
$ docker run --memory <size> ...
However, the documentation does not describe the format or units for the argument anywhere on the page:
--memory , -m Memory limit
What units should I supply to --memory and other related options like --memory-reservation and --memory-swap? Just bytes?
Classic case of RTFM on my part. The --memory option supports a unit suffix so we don't need to calculate the exact byte number:
-m, --memory=""
Memory limit (format: <number>[<unit>], where unit = b, k, m or g)
Allows you to constrain the memory available to a container. If the
host supports swap memory, then the -m memory setting can be larger
than physical RAM. If a limit of 0 is specified (not using -m), the
container's memory is not limited. The actual limit may be rounded up
to a multiple of the operating system's page size (the value would be
very large, that's millions of trillions).
So, to start a container with a 1 GB memory limit as described in the question, both of these commands will work:
$ docker run --memory 1g ...
$ docker run --memory 1073741824 ...
The --memory-reservation and --memory-swap options also support this convention.
Taken from the docker documentation:
Limit a container’s access to memory Docker can enforce hard memory
limits, which allow the container to use no more than a given amount
of user or system memory, or soft limits, which allow the container to
use as much memory as it needs unless certain conditions are met, such
as when the kernel detects low memory or contention on the host
machine. Some of these options have different effects when used alone
or when more than one option is set.
Most of these options take a positive integer, followed by a suffix of
b, k, m, g, to indicate bytes, kilobytes, megabytes, or gigabytes.
This page also includes some extra information about memory limits when running docker on Windows.
docker run -m 50m <imageId> <command...>
This is how it should be given. This forces the docker container to use 50m of memory. As soon as it tries to use more than that, it will be shut down.
However using free -m you won't be able to see anything related to the container memory usage. you have to go inside it to see allowed memory.

Docker container CPU allocation

I have created a container:
docker run -c=20 -i -t ubuntu:latest /bin/bash
I tried to use -c flag to control CPU usage and maximize it in 50 %. When I am running md5sum /dev/urandom inside container, it use up 100 % CPU in host machine.
The -c flag for docker run command modifies the container’s CPU share weighting relative to the weighting of all other running containers.
It does not restrict the container's use of CPU from the host machine.
You can use the --cpu-quota flag to limit CPU usage, for example:
$ docker run -ti --cpu-quota=50000 ubuntu:latest /bin/bash
The --cpu-quota is usually used in conjunction with --cpu-period. Please see more details on the Docker run reference document:
https://docs.docker.com/reference/run/#runtime-constraints-on-resources
It seems that you are running a single container, so this is the expected result.
You might find this blog post helpful.
Every new container will have 1024 shares of CPU by default. This
value does not mean anything, when speaking of it alone. But if we
start two containers and both will use 100% CPU, the CPU time will be
divided equally between the two containers because they both have the
same CPU shares (for the sake of simplicity I assume that there are no
other processes running).
Take a look here, this is apparently what you were looking for:
https://docs.docker.com/engine/reference/run/#cpu-period-constraint
The default CPU CFS (Completely Fair Scheduler) period is 100ms. We can use --cpu-period to set the period of CPUs to limit the container’s CPU usage. And usually --cpu-period should work with --cpu-quota.
Examples:
$ docker run -it --cpu-period=50000 --cpu-quota=25000 ubuntu:14.04 /bin/bash
If there is 1 CPU, this means the container can get 50% CPU worth of run-time every 50ms.
period and quota definition:
Within
each given "period" (microseconds), a group is allowed to consume only up to
"quota" microseconds of CPU time. When the CPU bandwidth consumption of a
group exceeds this limit (for that period), the tasks belonging to its
hierarchy will be throttled and are not allowed to run again until the next
period.

Docker CPU percentage

Is there any way that I can get the cpu percentage inside docker container and not outside of it?! docker stats DOCKER_ID shows the percentage which is exactly what I need but I need it as variable. I need to get cpu percentage inside the container itself and do some operation with it.
I have looked into different stuff such as cgroup and docker rest API, but they do not provide cpu percentage. If there is a way to get the cpu percentage inside the container and not outside of it will be perfect. I found one solution provided by someone in below link, which is still outside the container by the rest api, however I did not really get it how to calculate the percentage.
Get Docker Container CPU Usage as Percentage
You can install Google cAdvisor with Axibase Time-Series Database storage driver. It will collect and store CPU utilization measured both in core units as well as in percentages.
Screenshots with examples of how CPU is reported are located at the bottom of the page: https://axibase.com/products/axibase-time-series-database/writing-data/docker-cadvisor/
In a centralized configuration, the ATSD container itself can ingest metrics from multiple cAdvisor instances installed on multiple docker hosts.
EDIT 1: One liner to compute total CPU usage of all processes running inside the container. Adjust -d parameter to change the interval between samples to smooth spikes out:
top -b -d 5 -n 2 | awk '$1 == "PID" {block_num++; next} block_num == 2 {sum += $9;} END {print sum}'
I have used ctop which gives a more graphical way than docker_stats
But I found that it was showing CPU percentage way higher than what Top was showing for the system. Basically it is showing relative to the root process. Docker containers run as child process
To illustrate with an example
First find the root process under which all the containers run
docker-containerd-shim -
..the Docker architecture is broken into four components: Docker engine, containerd, containerd-shm and runC. The binaries are respectively called docker, docker-containerd, docker-containerd-shim, and docker-runc.
- https://hackernoon.com/docker-containerd-standalone-runtimes-heres-what-you-should-know-b834ef155426
root 1843 1918 0 Aug31 ? 00:00:00 docker-containerd-shim 611bd9... /var/run/docker/libcontainerd/611bd92.... docker-runc
You can see all the containers that are running using the command
pstree -p 1918
Now say that we are interested in seeing the CPU consumption of fluentdb.
Easy way to get the child pid of this is
pstree -p 1918 |grep fluentd
Which gives 21670
Now you can run top -p 21670 to see the CPU share of this child process also top -p 1918 to see the overall CPU of the parent process.
With cadvisor collecting to Promethus and view in Grafana, this was the closest and most accurate representation of the actual CPU percentage used by the container; in relation to the host machine. This diagram illustrates this.
cTop and docker stats give 23% as the CPU percentage. Actual CPU percentage of the docker parent process is around 2% and cAdvisor output from Grafana shows the most 'accurate' value of the container CPU percentage related to host.

Resources