why the contianer memory limit in cgroup not accurate

Question

The container set resource limit：

resources: limits: cpu: "1" memory: 1G requests: cpu: "1" memory: 1G

the cgroup memory limit:

cat /sys/fs/cgroup/memory/kubepods.slice/kubepods-podaace5b66_c7d0_11e9_ba2a_dcf401d01e81.slice/memory.limit_in_bytes 999997440

1GB= 1*1024*1024*1024=1,073,741,824B

k8s version:1.14.4

docker version: docker-ce-18.09.6 OS: ubuntu 18.04

prometherion · Accepted Answer · 2019-09-02 17:55:12Z

1

Because you declared a GibiByte using the G scale: if you expect a GigaByte, you should use Gi

# k get deployments.apps memory -o yaml|grep -i limits -C 1 resources: limits: memory: 1Gi $ cat /sys/fs/cgroup/memory/kubepods/burstable/pod15dff9ec-7815-48c0-bfce-453be77413ad/memory.limit_in_bytes 1073741824

answered Sep 2, 2019 at 17:55

prometherion

2,31914 silver badges22 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

mbx Over a year ago

Gibi is binary scale: (2^10)^3, while Giga is decimal scale (10^3)^3. Your solution is correct, your explanation isn't ;-) And certain folks still call (2^10)^3 Giga (i.e. GB) instead of Gibi (GiB) - looking at you, MicroSoft. see: kubernetes.io/docs/concepts/configuration/…

wow qing Over a year ago

thank you, the account is 999997440 not 1000000000?

prometherion Over a year ago

yeah, that's weird because the byte-based value is correct... need to investigate further! :)

Mark · Accepted Answer · 2019-09-05 09:53:37Z

I have performed some tests.

For the values between 999997440 B (976560 KB) and 1000000000 B (as in your example) you will have the same results in memory.limit_in_bytes = 999997440 B. Till you reach the next (integer) number of bytes divisible by your pagesize (default 4096). In my example it was 1000001536 B (976564K).

I am not linux expert but according to the documentation:

A successful write to this file does not guarantee a successful setting of this limit to the value written into the file. This can be due to a number of factors, such as rounding up to page boundaries or the total availability of memory on the system. The user is required to re-read this file after a write to guarantee the value committed by the kernel.

I would like to suggest use Gi notation instead as mentioned by prometherion to have more control about resources limits.

Hope this help.

Collectives™ on Stack Overflow

why the contianer memory limit in cgroup not accurate

2 Answers 2

3 Comments

Comments

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

3 Comments

Comments

Related