SCALE-CI-PIPELINE

Automates the installation of OCP on various cloud platforms and runs performance and scale tests related to kubelet density, control plane, http, storage, prometheus and cluster limits.

NOTE: This is also maintained at openshift/aos-cd-jobs repo.

Dependencies

- Running Jenkins instance.

Components

• Properties files
• Pipeline scripts
• Scale-ci-watcher
• Scale-ci-linter

Properties files

The parameters for each job in the scale-ci-pipeline are supplied through a job specific properties file. It contains key=value pairs, sample properties for all the supported jobs are hosted in scale-ci-pipeline/properties-files.

The properties for the OCP large scale and per sprint test runs are maintained at scale-ci-properties.

Pipeline scripts

These scripts are responsible for parsing the properties files and building the respective job.

Scale-ci-watcher

This looks for changes to the JJB templates or new templates and updates/onboards the jobs into the scale-ci-pipeline. The watcher also supports xml format, it has the support to convert them to JJB format. The scale-ci-watcher gets kicked off as the first job in the pipeline, it will pick up the changes made to the templates if there are any and applies them to the respective job.

It gives us an option to choose whether to update the Jobs in Jenkins and it does is by organizing the templates into dynamic and static directories inside of scale-ci-pipeline/jjb directory.

• Dynamic

  This directory contains the dynamic job templates managed by scale-ci-watcher for the scale-ci-pipeline which is used to test OpenShift at scale.

• Static

  This directory contains the static job templates in the scale-ci-pipeline which are not managed by scale-ci-watcher.

Scale-ci-linter

Validates scale-ci templates to analyze them for potential errors for every commit. This can be run locally as well in case we want to check before pushing the commit:

$ pip install yamllint
$ ./scale-ci-linter.sh <path-to-the-template>

Scale-CI Jobs

Job	OCP component/category	Description	Managed by scale-ci-watcher	OCP-3.X	OCP-4.X
Scale-ci-pipeline	Orchestrator	Orchestrates scale-ci job stages	✔️	✔️	✔️
Watcher	Onboards/updates workloads	Creates/updates jobs based on changes to the templates	✔️	✔️	✔️
OpenShift install on AWS	Installer	Installs and configures OCP cluster	✔️	❌	✔️
OpenShift install on Azure	Installer	Installs and configures OCP cluster	✔️	❌	✔️
Tooling	Tooling	Sets up pbench-agents to collect perf data	✔️	✔️	✔️
Conformance	OCP/kubernetes e2e tests	Checks if the OCP cluster is sane	✔️	❌	✔️
Scaleup	scaleup	Scales up OCP cluster to a desired node count	✔️	✔️	✔️
Node Vertical	Kubelet performance and Cluster Limits	Creates max pods per compute node	✔️	✔️	✔️
Master Vertical	Control plane density	Creates bunch of objects to stress ApiServer, Etcd and Controller	✔️	✔️	✔️
Networking	SDN	uperf benchmarks testing node to node, pod to pod, and svc to svc throughput and latency for TCP and UDP protocols	In progress	✔️	✔️
Prometheus	Monitoring	prometheus density focused test that creates pods, and nodes and some workloads especially for openshift-monitoring	In progress	✔️	✔️
HTTP	Router	Data-plane workload generator that runs http requests through HAProxy into deployed pods	✔️	✔️	✔️
Pgbench	Storage	Checks how well Postgresql performs with RHOCS for glusterfs and gluster-block storage backend	In progress	✔️	✔️
MongoDB	Storage	Checks how MongoDB performs with RHOCS for glusterfs and gluster-block storage backend	In progess	✔️	✔️
Pod Vertical	Cluster Limits	Tests pods per namespace limit	✔️	✔️	✔️
Deployments per namespaces	Cluster Limits	Tests deployments per namespace limit	✔️	✔️	✔️
Services per namespace	Cluster Limits	Tests maximum number of services possible per namespace	✔️	✔️	✔️
Namespaces per cluster	Cluster Limits	Tests namespaces per cluster limit	✔️	✔️	✔️
Baseline	Base/Idle cluster	Captures performance data of the base cluster with no user created objects	✔️	NA	✔️
Uperf	Networking	uperf benchmarks testing node to node, pod to pod, and svc to svc throughput and latency for TCP and UDP protocols	✔️	❌	✔️

Modifying/Adding new workloads to the scale-ci-pipeline

Adding new workloads is managed by scale-ci-watcher and it expects the following:

• JJB template: It is the job definition in yaml ( xml is also supported, watcher takes care of converting it to yaml )
• Pipeline-build script for the Job and a properties file: The script is responsible for building the job as a stage looking with the configutation/parameters specified in the properties file.

There are many existing jobs in scale-ci-pipeline which can serve as an example/template job. One example is the NodeVertical scale test. To create a new workload using NodeVertical as a template, follow these steps:

$ # step 1: copy one of the existing job template and edit it with the workload specific stuff
$ cp scale-ci-pipeline/jjb/dynamic/scale-ci_nodevertical.yml scale-ci-pipeline/jjb/dynamic/scale-ci_$workload.yaml
$ # step 2: copy one of the existing pipeline-build script and edit it with the workload specific stuff
$ cp scale-ci-pipeline/pipeline-scripts/nodevertical.groovy scale-ci-pipeline/pipeline-scripts/$workload.groovy
$ # step 3: copy the properties file and edit it with the workload specific stuff
$ cp scale-ci-pipeline/properties-files/nodevertical.properties scale-ci-pipeline/properties-files/$workload.properties
$ # step 4: add the new workload to the Jenkinsfile to load the pipeline-build script ( Applicable only in the case of a new workload, this is not needed for a existing workload )
$ vi scale-ci-pipeline/Jenkinsfile
$ # step 5: add the workoad to the scale-ci-pipeline ( workload option and properties file path vars, applicable only in the case of a new workload )
$ vi scale-ci-pipeline/jjb/dynamic/scale-ci-pipeline
$ # step 6: update the Scale-CI jobs section in the readme.

NOTE: Modifying the existing workload just needs changes to the existing template, build script and properties file.

Test setup/Migrating scale-ci jobs to any Jenkins

The following instructions will help you setup a jenkins test environment super quick so as to play around and test the changes locally before pushing them to production.

$ docker run -d --name scale-ci-jenkins -p 8080:8080 jenkins/jenkins:lts

Grab the admin password from the scale-ci-jenkins container logs:

$ docker logs scale-ci-jenkins

Access the jenkins at http://:8080 and proceed with the initial setup. A new user instead of admin user and installing all the suggested plugins is highly recommended. Once we have the jenkins up and running, run the scale-ci-watcher to push your templates to the test jenkins instance:

$ git clone https://github.com/openshift-scale/scale-ci-pipeline.git
$ cd scale-ci-pipeline
$ # Set the variables in the scale-ci-watcher-env.sh and source it
$ source scale-ci-watcher-env.sh
$ # Set the jenkins user name, url and passwd in the scale-ci-watcher config file
$ vi config/jjb.ini
$ # Run the scale-ci-watcher and check the jobs in the jenkins ( Make sure python six module version is >=1.10.0 )
$ ./scale-ci-watcher.sh
$ # Running the scale-ci-watcher when we don't have the SSL certs needed to verify https when hitting Jenkins api ( check if needed )
$ export PYTHONHTTPSVERIFY=0; ./scale-ci-watcher.sh

Scale-CI Jenkins agents/slaves

We use bunch of dedicated slaves part of OpenShift-QE Jenkins to run Scale-CI jobs. The slaves are always online instead of the on demand slaves which take time to get ready before scheduling the jobs. There are multiple slaves and executors, so we should be able to run jobs in parallel.

If a slave with different dependencies for example ansible version is needed by the job, please open an issue and we will create a slave which will meet the requirements.

Here is the list of current scale-ci labeled slaves:

Slaves	Number of executors	Label	Ansible version
scale-ci-slave-1	3	scale-ci	2.8.2
scale-ci-slave-2	3	scale-ci	2.8.2
scale-ci-slave-3	3	scale-ci	2.8.2
scale-ci-slave-4	3	scale-ci	2.8.2
scale-ci-slave-5	3	scale-ci	2.8.2
scale-ci-slave-6	3	scale-ci	2.8.2

Credits

Thanks to Mary Shakshober for designing the logo.