There are many use cases for needing multiple persistent disks to be mounted to a BOSH deployed VM. We are working on a DC/OS BOSH Release which requires multiple disks to be mounted to the mesos-agent servers, one for /var/vcap/store and one as a raw disk for Docker.

Before we dive into how this is done there are a couple of important bits:

  • The disks are mounted and a symlink is provided
  • The disks start as unformatted
  • None of the disks will be automatically be mounted to /var/vcap/store

Fear not, we will work around these last two issues.

Let's start!

BOSH Manifest Changes

There are a couple of items required in a manifest for a job to leverage multiple disks:

  • A consumes: key with the list of mappings to persistent disks.
  • A persistent_disks array mapping of the consumed disks list to disk pool names
  • A disk_pools or disk_type array of disk sizes and names

Here is an example stub manifest defining two disks to be mounted, one called portworx (used as raw disk) and another called store (to be mounted to /var/vcap/store):

- instances: 1
  - name: mesos-agent
    release: dcos
  - consumes:
        from: portworx      # <=== Maps to the `persistent_disks` name
        from: store         # <=== Maps to the `persistent_disks` name
    name: portworx
    release: dcos
  name: mesos-agent
  - name: dcos_z1
  - name: store
    type: 10GB               # <== Maps to `disk_pool` or `disk_types` name
  - name: portworx
    type: 100GB              # <== Maps to `disk_pool` or `disk_types` name
  resource_pool: default

The disk pools are defined as one of the two options below, notice that these map to the persistent_disks in the mesos_agent instance group above.

# Older BOSH v1 style disk pools:
- disk_size: 10240
  name: 10GB
- disk_size: 100240
  name: 100GB

# Or for newer BOSH v2 disk descriptions:
- disk_size: 10240
  name: 10GB
- disk_size: 100240
  name: 100GB

At this point if the deployment manifest was deployed, two disks would be mounted on the mesos-agent vm in /var/vcap/instance/disks as:

lrwxrwxrwx 1 root root 9 Sep 29 14:12 portworx -> /dev/xvdg  
lrwxrwxrwx 1 root root 9 Sep 29 14:12 store -> /dev/xvdf  

But there is nothing mounted to /var/vcap/store and /dev/xvdf is not formatted yet. You can do this manually, however we modified the BOSH release with a pre-start script to do this work automatically for us as seen in the next section.

BOSH Release Changes

Any BOSH Job which you would like to leverage multiple disks needs to format the disks. An example pre-start script shows off the basics of checking if the filesystem exists and then symlinking a disk to /var/vcap/store:


STORE=/var/vcap/instance/disks/<%= link("store").p("name") %>  
if ! e2fsck -n $STORE    #Checks if the filesystem exists  
  mkfs.ext4 -F $STORE   #...if not make the filesystem

#  `/var/vcap/store` is not defined for you, you have to do it yourself:
if ! mountpoint -q $STORE_DIR  
  mkdir -p $STORE_DIR
  mount -t ext4 $STORE $STORE_DIR

If we deploy the manifest with this modified release you will have a 10GB ext4 formatted drive symlinked to /var/vcap/store.

In the examples above the portworx disk was ignored because it is being presented in raw format to Docker to manage. If you wanted to format and use this disk as well we could have added the following to the pre-start script:

PORTWORX=/var/vcap/instance/disks/<%= link("portworx").p("name") %>  
if ! e2fsck -n $PORTWORX  
  mkfs.ext4 -F $PORTWORX
if ! mountpoint -q $PORTWORX_DIR  
  mkdir -p $PORTWORX_DIR
  mount -t ext4 $PORTWORX $PORTWORX_DIR

This results in the second disk being with 100GB available at /var/vcap/portworx formatted as ext4.

Retrieving the Device

In a BOSH release if you need the mapping of the device, you can readlink:

DISKPATH=$(readlink /var/vcap/instance/disks/<%= link("portworx").p("name") %>)  

If you echoed DISKPATH you would get the value /dev/xvdg.

Final Thoughts

The disks are not automatically deleted if the deployment is deleted, there is a 5 day grace period before the orphaned disks are cleaned up by BOSH. Verify your BOSH settings before relying on this feature.

Instead of embedding the prestart script in the DC/OS BOSH Release it may be more beneficial to create a separate multidisk-boshrelease and colocate it on the instance groups so it is easier to reuse. Look for an update on this!

Everyone deserves nice things, multiple disks via BOSH is a nice thing, be sure to share your tips on the greatness of BOSH!