Cleaning up frameworks on DC/OS


You will have to run the following steps after destroying a service to clean up all the resources in DC/OS. We are going to clean up the portworx-cassandra service in this example. These steps can be used to clean up any service in DC/OS including the Portworx service.

Shutdown the service

Find the ID of the service that you want to cleanup. The service should be in inactive state, i.e. ACTIVE should be set to False.

dcos service --inactive
NAME                         HOST               ACTIVE  TASKS  CPU    MEM      DISK   ID
portworx-cassandra  ip-10-0-2-15.ec2.internal   False    2    6.7  27530.0  59890.0  cc3a8927-1aec-4a8a-90d6-a9c317f9e8c6-0051
marathon                  10.0.4.203            True     4    3.1   3200.0    0.0    cc3a8927-1aec-4a8a-90d6-a9c317f9e8c6-0001
metronome                 10.0.4.203            True     0    0.0    0.0      0.0    cc3a8927-1aec-4a8a-90d6-a9c317f9e8c6-0000
portworx            ip-10-0-2-15.ec2.internal   True     6    1.8   5632.0  16384.0  cc3a8927-1aec-4a8a-90d6-a9c317f9e8c6-0050

Shutdown the service if you find it in inactive state. If you don’t find the service, there is no need to shutdown.

dcos service shutdown cc3a8927-1aec-4a8a-90d6-a9c317f9e8c6-0051

Run janitor script

The janitor script will clean up the reserved resources as well as any state stored in Zookeeper.

SERVICE_NAME=portworx-cassandra
PRE_RESERVED_ROLE="your_pre_reserved_role/" # Set this only if you started the service with a pre-reserved-role
dcos node ssh --master-proxy --leader \
    "docker run mesosphere/janitor /janitor.py -r ${PRE_RESERVED_ROLE}${SERVICE_NAME}-role -p ${SERVICE_NAME}-principal -z dcos-service-${SERVICE_NAME}"

Cleanup Portworx framework

If you are trying to cleanup Portworx framework, you will have to perform additional steps to cleanup the remnants from slave nodes. Run the commands below on all the private agents where Portworx was running.

Stop Portworx service

sudo systemctl stop portworx
sudo docker rm -f portworx.service

Remove Portworx service

sudo rm -f /etc/systemd/system/portworx.service
sudo rm -f /etc/systemd/system/dcos.target.wants/portworx.service
sudo rm -f /etc/systemd/system/multi-user.target.wants/portworx.service
sudo systemctl daemon-reload

Wipe Portworx drives and config

Note: If you are going to re-install Portworx, you should wipe out the filesystem from the disks so that they can be picked up by Portworx in the next install. This can be done by running the following pxctl command:
# Use with care since this will wipe data from all the disks given to Portworx
sudo /opt/pwx/bin/pxctl service node-wipe --all

If you are running Portworx version < 1.3, run the following commands instead of node-wipe:

sudo wipefs -a /dev/sda123 # Replace with your disk names
sudo chattr -i /etc/pwx/.private.json
sudo rm -rf /etc/pwx
sudo umount /opt/pwx/oci
sudo rm -rf /opt/pwx

Remove Portworx kernel module

sudo rmmod -f px

Cleanup slaves with script

Alternatively, if you have the DC/OS CLI installed, then you can execute the above steps on all the nodes by running the following script:

ips=(`dcos node --json | jq -r '.[] | select(.type == "agent") | .id'`)
for ip in "${ips[@]}"
do
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo systemctl stop portworx'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo docker rm -f portworx.service'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo rm -f /etc/systemd/system/portworx.service'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo rm -f /etc/systemd/system/dcos.target.wants/portworx.service'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo rm -f /etc/systemd/system/multi-user.target.wants/portworx.service'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo systemctl daemon-reload'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo /opt/pwx/bin/pxctl service node-wipe --all'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo chattr -i /etc/pwx/.private.json'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo rm -rf /etc/pwx'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo umount /opt/pwx/oci'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo rm -rf /opt/pwx'
        dcos node ssh --mesos-id=${ip} --master-proxy 'sudo rmmod -f px'
done

Remove Portworx metadata from Zookeeper

Note: Only if you were running Portworx with internal kvdb, you will have to cleanup Portworx metadata from DC/OS Zookeeper.

In the Exhibitor UI, the metadata should be present under the Zookeeper node /pwx/<portworx_cluster_id>. In the below example it is /pwx/portworx-dcos.

Exhibitor Portworx metadata

Select the /pwx/<portworx_cluster_id> node and click on Modify button at the bottom of the page. In the Modify Node dialog box select the Type as Delete.

Exhibitor Portworx delete metadata



Last edited: Thursday, Aug 8, 2019