Inorder to make sure the code contributed by you, to OpenStack community, remains stable and to identify problems before it is too late, you need to setup a testing infrastructure. It should continuously test and give feedback/vote whenever changes get submitted to the community. Tests cannot be run by the openstack community CI infrastructure, because it won't have your/vendor's hardware/software. You need to setup your own testing infrastructure, probably in your company lab, in the deployment you prefer and test the changes that get submitted.
We had contributed NetScaler driver, inorder to make sure it is not broken we setup a test infrastructure. The following paragraphs captures the thought process of setting up a test infrasture, and provides code snipetts from the code that was used to setup the testing infrastructure at Citrix. Hoping, it will be useful for others who are setting up their own test infrastructure.
The starting point is http://ci.openstack.org/third_party.html
This document can be considered an addendum to the above link.
You have to be clear on certain things before jumping on to setup the testing infrastructure.
- What are you trying to qualify?
- This should have been obvious in most cases. It is most probably the plugin/driver that was written for your software/hardware.
- What changes to listen for?
- Make a note of the code areas that will impact your software, you would want to trigger tests only if those changes are submitted on that. For example, it could be the core modules of the openstack project.
- What tests to run?
- Most likely vendors run a subset of tempest tests, but you could in addition run your own tests as well.
- You might want to enable the test infrastructure to run tests on a particular changelist. This will be handy when you observe that a code submit event got lost and you want to trigger a test offline. Or during the test infrastructure development phase.
- You might not want to vote (-1) from the test infrastructure to start with. It is best if the testing infrastructure can send you an email on -1 cases. You can inspect files manually, once confident, you could trigger the the negatve vote manually.
Openstack has a gerrit code review system via which developers submit code. Your infrastructure has to listen to gerrit, and whenever relavant code patches are submitted to gerrit it should trigger tests.
Use pypi pygerrit to listen to events.
The example code of pygerrit is a good place to start https://github.com/sonyxperiadev/pygerrit/blob/master/example.py
Get the patchset and project details from the event.
if isinstance(event, PatchsetCreatedEvent):
change_ref = patchSetCreatedEvent.patchset.ref
submitted_project = patchSetCreatedEvent.change.project
Check if files match criteria?
Use pypi GitPython in combination with git command line to inspect the files in the patch. If you already have answered - What changes to listen for? - you should have the exact directories/files to check during code submission. Here is a code snippet that will check if files in a change submitted is of your interest.
def execute_command(command, delimiter=' '):
command_as_array = command.split(delimiter)
logging.debug("Executing command: " + str(command))
p = subprocess.Popen(command_as_array,stdout=subprocess.PIPE, stderr=subprocess.PIPE)
output, errors = p.communicate()
if p.returncode != 0:
logging.error("Error: Could not exuecute command " + str(command) + ". Failed with errors " + str(errors))
return False
logging.debug("Output command: " + str(output))
return True
"""
1. local_repo_path == /opt/stack/gerrit_depot/neutron # the git repository that will be used for file inspection
2. review_repo_name == gerrit_repo # git remote name in local_repo_path that is pointing to review.openstack.org repository
3. files_to_check == 'neutron/services/loadbalancer/drivers/netscaler' # files of interest
4. change_ref == refs/changes/24/57524/9 # changeref of a patchset submitted for a particular change
"""
def are_files_matching_criteria(local_repo_path, review_repo_name, files_to_check, change_ref):
""" Issue checkout using command line"""
logging.info("Fetching the changes submitted")
os.chdir(local_repo_path)
is_executed = execute_command("git checkout master")
if not is_executed:
return False, None
is_executed = execute_command("git fetch " + review_repo_name + " " + change_ref)
if not is_executed:
return False, None
is_executed = execute_command("git checkout FETCH_HEAD")
if not is_executed:
return False, None
""" Check the files and see if they are matching criteria using GitPython"""
repo = Repo(local_repo_path)
review_remote = None
for remote in repo.remotes:
if remote.name == review_repo_name:
review_remote=remote
break
if not review_remote:
logging.error("Unable to find review repo. It is used to check if files are matched")
return False, None
headcommit = repo.head.commit
commitid = headcommit.hexsha
submitted_files = headcommit.stats.files.keys()
for submitted_file in submitted_files:
for file_to_check in files_to_check:
if file_to_check in submitted_file:
logging.info("Some files changed match the test criteria")
return True, commitid
return False, None
Once the code submitted is found to be of interest, the tests idenified have to be run.
The first step is to setup the systems involved in testing. You should setup vendor specific systems in the deployment to clean slate, and also setup DevStack. To setup the former, you are the best person to know the steps. To setup the latter (devstack), following are the steps that are recommended
- Use an appropriate localrc with Devstack VM. Here is a full sample. It is recommended to use the following setting
RECLONE=YES # inorder to pull latest changes during every test cycle
DEST=/opt/stack/new # log collection scripts would be expecting devstack to be installed in this directory
- Run the following script to setup DevStack
cd $DEVSTACK_DIR
./unstack.sh > /tmp/unstack.out 2>&1
./stack.sh > /tmp/stack.out 2>&1
- Patch submitted code
function patch_submited_change
{
# patch the newly submittedfiles
if [ -n "$CHANGE_REF_PROJECT" ]
then
echo "Patching changeref submitted"
cd $ROOT_DIR/$CHANGE_REF_PROJECT
git checkout master
git fetch https://review.openstack.org/openstack/$CHANGE_REF_PROJECT $CHANGE_REF && git checkout FETCH_HEAD
else
echo "Nothing to be patched"
return
fi
}
Setup openstack configuration files to use your software
We had to patch the neutron.conf to include NetScaler driver
function configure_netscaler_driver
{
echo "Configuring NetScaler as the default LBaaS provider...."
sed -i 's!HaproxyOnHostPluginDriver:default!HaproxyOnHostPluginDriver\nservice_provider=LOADBALANCER:NetScaler:neutron.services.loadbalancer.drivers.netscaler.netscaler_driver.NetScalerPluginDriver:default!g' /etc/neutron/neutron.conf
}
Restart concerned openstack service
We had to restart neutron
function wait_till_port_open
{
PORT_NUM=$1
# Waiting 2 minutes for the service to be up
for i in {1..120}
do
port_open=`netstat -an | grep $PORT_NUM | wc -l | tr -d ' '`
if [ $port_open -eq '1' ]
then
echo "Port "$PORT_NUM" is UP"
return
fi
sleep 1
done
echo "Error: Timed out waiting for service to be UP on port "$PORT_NUM
exit 1
}
function restart_neutron
{
# restart neutron
PID=`ps ax | grep neutron-server | grep -v grep | awk '{print $1}'`
echo "Stopping neutron process: $PID"
kill -9 $PID
NL=`echo -ne '\015'`
screen -S stack -p 'q-svc' -X stuff 'cd /opt/stack/new/neutron && python /usr/local/bin/neutron-server --config-file /etc/neutron/neutron.conf --config-file /etc/neutron/plugins/ml2/ml2_conf.ini'$NL
# wait till neutron is up
wait_till_port_open 9696
}
We run the LBaaS API tests today. In future, LBaaS scenario tests will be included.
cd /opt/stack/new/tempest && testr init
cd /opt/stack/new/tempest && testr run tempest.api.network.test_load_balancer
The best way to collect logs from DevStack is to use cleanup_host function present in devstack-gate's functions.sh. In addition to collecting log files it also generates results in pretty format.
NOTE The above script is dependent on https://github.com/openstack-infra/config/blob/master/modules/jenkins/files/slave_scripts/subunit2html.py copy this to /usr/local/jenkins/slave_scripts/subunit2html.py
Upload the log files in the public domain. Only then, the community members will be able to have a look at the test results. At Citrix, we have used sharefile. We might be able to contribute space for community depending on the number of requests received. Please feel free to shoot a mail to me by next week (Jan-25).
The final step in the process is to vote (+1/-1) depending on the result. There are three kinds of voting available in the gerrit system. The 3rd party infrastructure is expected to execute the 'Verified' votes. Apply for a service account in openstack as per the details specificed in http://ci.openstack.org/third_party.html#requesting-a-service-account Use the ssh key to execute a Verified vote. An example is given below
$ ssh -p 29418 review.openstack.org gerrit review -m '"LBaaS API testing failed with NetScaler providing LBaaS. Please find logs at <http://....>"' --verified=-1 c0ff33111123313131
NOTE Vote should contain link to logs.