K8s Storage Performance

Ansible playbooks to collect Storage performance metrics on an OpenShift cluster.

Note that: to assess if a storage provider, especially for new storage, is suitable to support ReadWriteOnce and ReadWriteMany volumes for your Cloud Pak workloads or not, check out this project at https://github.com/IBM/k8s-storage-tests, and perform the storage readiness tests provided there. It will give you a good assessment of the particular storage.

Prerequisites

• Ensure you have python 3.6 or later and pip 21.1.3 or later installed

  python --version

  pip --version

  NB: if your python interpreter is using python3 or python37 or other Python 3 executables, you can create a symlink for python using this command

  ln -s -f /usr/bin/python3 /usr/bin/python
  
  # OR depends on the Python 3 installation location
  
  ln -s -f /usr/local/bin/python3 /usr/local/bin/python

  NB: if pip is not available or is an older version, run the command below to upgrade it, and then check its version again. If pip command can't be found after the below command, add /usr/local/bin into your PATH ENV variable.

  python -m pip install --upgrade pip

• Install Ansible 2.10.5 or later

  pip install ansible==7.0.0

• Install ansible k8s modules

  pip install openshift

  ansible-galaxy collection install operator_sdk.util

  ansible-galaxy collection install kubernetes.core

  NB: the openshift package installation requires PyYAML >= 5.4.1, and if the existing PyYAML is an older version, then PyYAML's installation will fail. To overcome this issue, manually delete the exsiting PyYAML package as below (adjust the paths in the commands according to the your host environment):

  rm -rf /usr/lib64/python3.6/site-packages/yaml
  rm -f  /usr/lib64/python3.6/site-packages/PyYAML-*

• Install OpenShift Client 4.6 or later based on your OS.

• Access to the OpenShift Cluster (at least 3 compute nodes) setup with RWX and RWO storage classes with cluster admin access.

Setup

• Clone this git repo to your client

   git clone https://github.com/IBM/k8s-storage-perf

• Select the appropriate parameter yaml file for the level of data collection you would like. There are three versions of the params file

  • params.yml - Default. Will only run selected write tests that are considered in the CP4D Knowledge Center.
  • params-extended-metrics.yml - Extended writes. Will run all write tests.
  • params-all-metrics.yml - All tests. Will run all combinations of read and write tests.

• Update the params.yml file with your OCP URL and Credentials

  ocp_url: https://<required>:6443
  ocp_username: <required>
  ocp_password: <required>
  ocp_token: <required if user/password not available>
  ocp_apikey: <required if neither user/password or token not available>

  NB: update only one of the three login formats, that is, either ocp_username/ocp_password, or ocp_token, or ocp_apikey. Leave other two choices' lines intact, i.e. do NOT change the lines for the login formats that are not chosen to be used.

• Update the params.yml file for the required storage parameters

    run_storage_perf: true
    arch: amd64  # amd64, ppc64le

    storageClass_ReadWriteOnce: <required> 
    storageClass_ReadWriteMany: <required> 
    storage_validation_namespace: <required>

• Optionally, you can set/modify these label parameters to display in the final CSV report

    cluster_infrastructure: "self-cpd-cli managed" # optional label eg ibmcloud, aws, azure, vmware
    cluster_name: storage-performance-cluster      # optional labels
    storage_type: <storage vendor>

• Optionally you can run the tests with a "remote mode" where the performance jobs can run on a dedicated compute node. The compute node should be labelled with a defined key and value for this purpose and set in the params file.

    dedicated_compute_node:
      label_key: "<optional>"
      label_value: "<optional>"

  To label a node, you can use this command

    oc label node <node name> "<label_key>=<label_value>" --overwrite

Running the Playbook

• From the root of this repository, run:

  ansible-playbook main.yml --extra-vars "@./params.yml" | tee output.log

  NB: if the playbook fails to run due to SSL verification error, you can disable it by setting this environment variable before running the playbook

  export K8S_AUTH_VERIFY_SSL=no

• Storage performance role takes about an hour to run. When completed, a storage-perf.tar file will be generated for storage performance with the following contents

  • result.csv
  • nodes.csv
  • pods.csv
  • jobs.csv
  • params.log

Pulling and loading the required image in airgap environment

This storage performance test suite relies on a container image: quay.io/ibm-cp4d-public/xsysbench:1.1 This image may not be directly accessible on an aigap cluster. To resolve this, follow the steps below to download the image onto an intermediary host and then copy the image to the airgap cluster, and finally load it into the cluster's private registry.

 # on an intermediary host that can access the image
 podman pull quay.io/ibm-cp4d-public/xsysbench:1.1
 podman save -o xsysbench-1.1.tar quay.io/ibm-cp4d-public/xsysbench:1.1

 # copy the above .tar file onto the airgap cluster

 # on the airgap cluster
 podman load -i xsysbench-1.1.tar
 podman tag quay.io/ibm-cp4d-public/xsysbench:1.1 <private-registry>/ibm-cp4d-public/xsysbench:1.1
 podman tag quay.io/ibm-cp4d-public/xsysbench:1.1 <private-registry>/ibm-cp4d-public/xsysbench:1.1-amd64

 podman login -u <uaername> -p <password> <private-registry> --tls-verify=false

 podman push <private-registry>/ibm-cp4d-public/xsysbench:1.1
 podman push <private-registry>/ibm-cp4d-public/xsysbench:1.1-amd64

Next make sure that in the "params.yml" file, you modify the imageurl line to below:

imageurl: <private-registry>/ibm-cp4d-public/xsysbench:1.1

Finally just follow the previous sections to run the test suites.

Running the Playbook with the Container

Environment Setup

export dockerexe=podman # or docker
export container_name=k8s-storage-perf
export docker_image=icr.io/cpopen/cpd/k8s-storage-perf:v1.0.0

alias k8s_storage_perf_exec="${dockerexe} exec ${container_name}"
alias run_k8s_storage_perf="k8s_storage_perf_exec ansible-playbook main.yml --extra-vars \"@/tmp/work-dir/params.yml\" | tee output.log"
alias run_k8s_storage_perf_cleanup="k8s_storage_perf_exec cleanup.sh -n ${NAMESPACE} -d"

Start the Container

mkdir -p /tmp/k8s_storage_perf/work-dir
cp ./params.yml /tmp/k8s_storage_perf/work-dir/params.yml

${dockerexe} pull ${docker_image}
${dockerexe} run --name ${container_name} -d -v /tmp/k8s_storage_perf/work-dir:/tmp/work-dir ${docker_image}

Run the Playbook

run_k8s_storage_perf

Then to view the results:

mkdir /tmp/k8s_storage_perf/work-dir/data
${dockerexe} cp ${container_name}:/opt/ansible/storage-perf.tar /tmp/k8s_storage_perf/work-dir/data/storage-perf.tar
tar -xvf /tmp/k8s_storage_perf/work-dir/data/storage-perf.tar -C /tmp/k8s_storage_perf/work-dir/data

ls -l /tmp/k8s_storage_perf/work-dir/data
-rw-r--r-- 1 1001 root   190 Aug  2 17:18 jobs.csv
-rw-r--r-- 1 1001 root  5615 Aug  2 17:18 nodes.csv
-rw-r--r-- 1 1001 root   952 Aug  2 17:18 params.log
-rw-r--r-- 1 1001 root   484 Aug  2 17:18 pods.csv
-rw-r--r-- 1 1001 root  1453 Aug  2 17:18 result.csv
-rw-r--r-- 1 root root 20480 Aug  2 17:18 storage-perf.tar

Optional Cleanup the Cluster

run_k8s_storage_perf_cleanup

[INFO ] running clean up for namespace storage-validation-1 and the namespace will be deleted
[INFO ] please run the following command in a terminal that has access to the cluster to clean up after the ansible playbooks

oc get job -n storage-validation-1 -o name | xargs -I % -n 1 oc delete % -n storage-validation-1 && \
oc get pvc -n storage-validation-1 -o name | xargs -I % -n 1 oc delete % -n storage-validation-1 && \
oc get cm -n storage-validation-1 -o name | xargs -I % -n 1 oc delete % -n storage-validation-1 && \
oc delete ns storage-validation-1 --ignore-not-found

[INFO ] cleanup script finished with no errors

Clean-up Resources

With each run, delete the kuberbetes namespace that you created in Setup, you can delete the project

export STORAGE_PERF_NAMESPACE=<storage_perf_namespace>
oc delete project ${STORAGE_PERF_NAMESPACE}

OR delete the resources in the project individually

export STORAGE_PERF_NAMESPACE=<storage_perf_namespace>
oc delete job $(oc get jobs -n ${STORAGE_PERF_NAMESPACE} | grep -Ev NAME | awk '{ print $1 }') -n ${STORAGE_PERF_NAMESPACE}
oc delete pvc $(oc get pvc -n ${STORAGE_PERF_NAMESPACE} | grep -Ev NAME | awk '{ print $1 }') -n ${STORAGE_PERF_NAMESPACE}
# optionally
oc delete project ${STORAGE_PERF_NAMESPACE}