CF Performance Tests Pipeline

This repository defines a Concourse pipeline that automatically runs the cf-performance-tests against new releases of cf-deployment and stores the results in this repo. The pipeline also generates .png charts that show how performance varies across different cf-deployment releases. These are regenerated each time a new results file is added, along with a coverage table that gives an overview of releases that have been tested.

The pipeline is triggered by new cf-deployment releases. It uses bbl to stand a bosh director, and then uses that director to deploy a Cloud Foundry foundation. Each pipeline run will run the tests twice against a specific release of cf-deployment - once with a Cloud Controller that has a postgres CCDB, and once with a mysql CCDB. After completion, the bbl test environment is torn down.

See the docs subdirectory for instructions on one-off manual setup steps that need to be run before running any tests, along with guidance on configuring new automated tests and manual setup steps for triggering a series of tests against old releases of cf-deployment.

Test results and charts

Results are stored automatically in the following paths in this repo, according to the Cloud Controller that was tested (the go proof-of-concept or the regular CC written in rails) and whether a postgres or mysql CCDB was used.

results
├── go #tests that swap in a proof-of-concept Cloud Controller written in go: https://github.com/cloudfoundry/go-cf-api (archived results only)
│   ├── mysql
│   │   ├── charts
│   │   └── results
│   └── postgres
│       ├── charts
│       └── results
└── rails #tests using the regular Cloud Controller
    ├── mysql
    │   ├── charts
    │   └── results
    └── postgres
        ├── charts
        └── results

Four types of chart are generated:

1. Detailed: contains the largest, shortest and average cf api execution time.
2. Detailed with most recent runs. Same as 1, but only contains the last 15 runs.
3. Simplified: Chart only contains the average cf api execution time.
4. Simplified with most recent runs: same as 3, only contains the last 15 runs.

General information

The AWS account and domain used to host the BBL and CF foundation is controlled by SAP, but may move to a community owned account in the future. See here for information on the manual steps that were followed to get the tests automated.

The pipelines are running on concourse.app-runtime-interfaces.ci.cloudfoundry.org. All secrets referenced by the pipelines (e.g. ((aws-pipeline-user-secret))) are stored in a CredHub server deployed alongside Concourse. Contributors who need access must contact the App Runtime Interfaces Working Group in order to get approval to be added to the working group's Google Cloud account. Once access is granted, you can login to CredHub with this script.

Access a performance test environment

Although the test environments are torn down automatically at the end of a successful run, you might need to access one to debug a failure.

1. Follow these instructions to install bbl locally.
2. Navigate to the s3 bucket cf-performance-tests and find the subfolder for the test environment you want to connect to and download the tarball bbl-state.tgz.
3. Extract the archive locally, and cd into the state directory.
4. Run eval "$(bbl print-env)"
5. Confirm you can now communicate with the bosh director by running bosh env
6. You should also now be able to access the director's CredHub with the credhub CLI.

Connect to the CCDB: MySQL

If you have deployed a test environment with MySQL as the Cloud Controller's database, you can open a tunnel to the jumpbox and then connect from there. Initialise bbl as explained above and then run:

ssh -4 -N -o "UserKnownHostsFile=/dev/null" -o "StrictHostKeyChecking=no" -o "ServerAliveInterval=30" -o "ServerAliveCountMax=10" -o "IPQoS=throughput" \
  -i "$JUMPBOX_PRIVATE_KEY" -L "3306:10.0.16.5:3306" jumpbox@<jumpbox ip from $BOSH_ALL_PROXY> &
# copy mysql client from "database" vm
bosh -d cf scp database:/usr/local/bin/mysql .
./mysql --host=127.0.0.1 --port=3306 --user=cloud_controller cloud_controller --password=<cc_database_password from credhub>

From the mysql command prompt, you can e.g. use source db.sql to read and execute statements from a file.

Connect to the CCDB: PostgreSQL

If you have deployed a test environment with PostgreSQL as the Cloud Controller's database, you can open a tunnel to the jumpbox and then connect from there. Initialise bbl as explained above and then run:

ssh -4 -N -o "UserKnownHostsFile=/dev/null" -o "StrictHostKeyChecking=no" -o "ServerAliveInterval=30" -o "ServerAliveCountMax=10" -o "IPQoS=throughput" \
  -i "$JUMPBOX_PRIVATE_KEY" -L "5524:10.0.16.5:5524" jumpbox@<jumpbox ip from $BOSH_ALL_PROXY> &

To get PostgreSQL run:

apt-get update
apt-get install postgresql

To connect to psql run:

psql --host=127.0.0.1 --port=5524 --user=cloud_controller cloud_controller 

Troubleshooting

The tests are fully automated, but some of the following techniques might help if you need to debug a failure:

A pipeline task fails

1. Login to Concourse on the commandline with fly --target bosh-cf login --concourse-url https://bosh.ci.cloudfoundry.org/
2. In your web browser, navigate to the latest failed build (only the container for the latest build is preserved), and copy the URL.
3. In your terminal, run fly --target bosh-cf hijack --url <BUILD_URL>. You should connect to the container.

If the step that failed was a task, a consider re-running its script from the container with -x to log out all commands that are run. For example, if the deploy-cf task fails, get the name of its yaml configuration file from the pipeline file (cf-deployment-concourse-tasks/bosh-deploy/task.yml), and look at for the path to the script. In this case, you'll need to run bash -x cf-deployment-concourse-tasks/bosh-deploy/task.sh

bbl-up or bbl-destroy fail

You can retrieve the debug logs of bbl, which include terraform logs and bosh CLI output, by hijacking the appropriate task container. The directory that your session starts in should contain a number of log files - bbl_plan.log, bbl_up.log or bbl_destroy.txt depending on the task.

If you're not sure whether the director actually exists, as a sanity check you might consider navigating to the AWS console and checking in eu-central-1 to see if a VM named bosh/0 with the appropriate tags for the test environment exists and is in a started state.

deploy-cf or bosh-delete-deployments fail

1. Access the director by following these instructions
2. When debugging a failure it's often useful to run bosh tasks --recent=<NUMBER> to find the number of a failed task, and then retrieve its debug logs with bosh task <TASK_NUMBER> --debug.
3. Also consider setting BOSH_LOG_LEVEL=debug when running other commands.

Automated teardown fails

Aside from the several dozen bosh-managed VMs created as part of a cf deployment, the pipeline creates a total of about 100 AWS resources per test in environment. Although the pipeline is written to destroy these resources upon successfully completing the tests, it also has a series of jobs that can be manually triggered to separately delete the cf bosh deployment, the bosh director, the base infrastructure (upon which the former both depend) - or all three reverse-order. These are defined in a pipeline group that you can view by clicking the manual-teardown button in the top-left of the Concourse UI when viewing the pipeline in question.

Warning: If you run manual-teardown-base-infra-only deletes the IAM user and associated policy that bbl depends on to manage resources in AWS. You should not run this before bbl has destroyed those resources.

If these manual jobs fail to clean up an environment, you will need to locate the test environment's subfolder in the cf-performance-tests S3 bucket and, depending on the failure, download either bbl-state.tgz or base-infra/terraform.tfstate.

• bbl-state.tgz:
  1. Extract the archive and change into the state directory
  2. Run eval "$(bbl print-env)" and then bosh vms to check if any bosh-managed vms still exist. If they do, try to delete them with bosh (if this fails, you'll have to do this manually in AWS)
  3. If there are no bosh-managed VMs left but you've still got a director, try to delete it with bbl destroy. If this fails, you'll have to delete the director and jumpbox VMs manually in AWS.
  4. If bbl destroy is unable to delete the director, or succeeds but then fails while destroying terraform resources, you'll have to try to delete these yourself.
  5. Download version 0.11.x of the terraform CLI, and move vars/terraform.tfstate and vars/bbl.tfvars into the terraform subdirectory.
  6. Run terraform init and provide the values for any required variables, then terraform destroy -var-file=bbl.tfvars. You may need to remove resources from the tls provider from the state file.
• base-infra/terraform.tfstate
  1. Download the file, place it in base-infra/terraform, and run terraform init with whatever version of the terraform CLI matches that used by the latest commit of the Concourse terraform resource when the pipeline last ran the base-infra job. Then terraform destroy.

In both cases, the last resort - and this should almost never be necessary - will be to manually delete any remaining resources from AWS