rke: cri_dockerd as included in rke-tools has extremely bad performance

RKE version: 1.3.11 - v1.23.6

Docker version: (docker version, docker info preferred) 20.10

Operating system and kernel: (cat /etc/os-release, uname -r preferred) Ubuntu 20.04 LTS

Type/provider of hosts: (VirtualBox/Bare-metal/AWS/GCE/DO) Azure

cluster.yml file:

  rke_config {
    kubernetes_version = var.k8version
    ignore_docker_version = false
    enable_cri_dockerd = true
    }
    network {
      plugin = "flannel"
    }

Steps to Reproduce: Set enable_cri_dockerd will lead to constant, near 100% CPU consumption from an empty cluster on a 4vCPU VM (Standard_D4s_v3)

Results: With 1.24 approaching fast, we should provide a working option for the installed base that wants to continue with RKE1 and Docker.

About this issue

  • Original URL
  • State: closed
  • Created 2 years ago
  • Comments: 17 (6 by maintainers)

Most upvoted comments

Ticket #2938 - Test Results

Reproduced with rke v1.3.11 and k8s v1.23.6:

  1. Create a single EC-2 instance of t3.medium
  2. Using rke v1.3.11, run rke up on instance with cluster.yml ignore_docker_version = false and enable_cri_dockerd = true configurations set - [k8s v1.23.6-rancher1-1]
  3. Once cluster successfully builds, ssh into the node and run top to view CPU performace
  4. Reproduced - CPU consumption EXTREMELY high from empty cluster

Screenshot:

Step 4 Screen Shot 2022-06-13 at 1 35 09 PM

@Raboo You are right; the fix is not in rancher 2.6.6 but will be in 2.6.7 ( as the milestone indicates) which is planned to be released this week.

@Raboo The plan is in Rancher ~2.6.6~ 2.6.7