Docker interaction with Portworx
Docker interaction with Portworx
Portworx implements the Docker Volume Plugin Specification.
The plugin API allows creation, instantiation, and lifecycle management of Portworx volumes. This allows direct use by Docker, Docker swarm, and DCOS via dvdi.
Docker scans the plugin directory (
/run/docker/plugins) on startup and whenever a user or a container requests a plugin by name.
When the Portworx container is run, a unix domain socket
pxd.sock is exported under
/var/run/docker/plugins directory. Portworx volumes are shown as owned by volume driver
Portworx volumes are created by specifying volume driver as
Here is an example of how to create a 10GB volume with replication factor set to 3:
docker volume create --driver pxd \ --opt size=10G \ --opt repl = 3 \ --name my_portworx_vol
Docker looks in its cache before sending the request to create to Portworx. For this reason, Portworx, Inc. recommends to not mix-and-match create and delete operations with pxctl and docker. If a volume with the same name is created again, it is a No-op.
Use of options in docker volume create
You can include any desired volume options with the
volume create command:
The following table lists what options you can include:
|fs||Filesystem to be laid out: xfs|ext4||fs: “ext4”|
|repl||Replication factor for the volume: 1|2|3||repl: “3”|
|sharedv4||Flag to create a globally shared namespace volume which can be used by multiple pods over NFS with POSIX compliant semantics||sharedv4: “true”|
|priority_io||IO Priority: low|medium|high||priority_io: “high”|
|io_profile||Overrides the I/O algorithm Portworx uses for a volume. For more information about IO profiles, see the IO profiles section of the documentation.||io_profile: “db”|
|group||The group a volume should belong too. Portworx will restrict replication sets of volumes of the same group on different nodes. If the force group option ‘fg’ is set to true, the volume group rule will be strictly enforced. By default, it’s not strictly enforced.||group: “volgroup1”|
|fg||This option enforces volume group policy. If a volume belonging to a group cannot find nodes for it’s replication sets which don’t have other volumes of same group, the volume creation will fail.||fg: “true”|
|label||List of comma-separated name=value pairs to apply to the Portworx volume||label: “name=mypxvol”|
|nodes||Comma-separated Portworx Node ID’s to use for replication sets of the volume||nodes: “minion1,minion2”|
|aggregation_level||Specifies the number of replication sets the volume can be aggregated from||aggregation_level: “2”|
|sticky||Flag to create sticky volumes that cannot be deleted until the flag is disabled||sticky: “true”|
|journal||Flag to indicate if you want to use journal device for the volume’s data. This will use the journal device that you used when installing Portworx. This is useful to absorb frequent syncs from short bursty workloads. Default: false||journal: “true”|
|placement_strategy||Flag to refer the name of the
|secure||Flag to create an encrypted volume. For details about how you can create encrypted volumes, see the Create encrypted PVCs page.||secure: “true”|
|Flag to create scheduled snapshots with Stork. For example:
Note: This example references two schedules:For details about how you can create scheduled snapshots with Stork, see the Scheduled snapshots page.
|sharedv4_svc_type||Flag to indicate the mechanism Kubernetes will use for locating your
Specify replica nodes
Multiple nodes through docker volume create is supported from 184.108.40.206.
Use the nodes option to specify the nodes you wish the replicas to reside on.
Some valid examples of this are:
It is important to note that the number of nodes should equal the repl option otherwise Portworx will pick a node for the remaining requested replica’s.
Scheduled snapshots are only available in Portworx 1.3 and higher.
Use the _snapschedule option to specify the snapshot schedule.
Following are the accepted formats:
snaps-to-keep is optional. Periodic, Daily, Weekly and Monthly keep last 5, 7, 5 and 12 snapshots by default respectively.
Some examples of snapshots schedules are:
There is no explicit Snapshot operation via Docker plugin API. However, this can be achieved via the create operation. Specifying a
parent operation will create a snapshot.
The following command creates the volume
snap_of_my_portworx_vol by taking a snapshot of
docker volume create --driver pxd \ --opt parent=my_portworx_vol \ --name snap_of_my_portworx_vol
The snapshot can then be used as a regular Portworx volume.
Mount operation mounts the Portworx volume in the propagated mount location. If the device is un-attached,
Mount will implicitly perform an attach as well. Mounts are reference counted and are idempotent. The same volume can be mounted at muliple locations on the same node. The same device can be mounted at the same location multiple times.
The docker plugin API does not have an Attach call. The Attach call is called internally via Mount on the first mount call for the volume.
Portworx exports virtual block devices in the host namespace. This is done via the Portworx container running on the system and does not rely on an external protocol such as iSCSI or NBD. Portworx virtual block devices only exist in host kernel memory. Two interesting consequences of this architecture are: 1) volumes can be unmounted from dead/disconnected nodes 2) IOs on porworx can survive a Portworx restart.
Portworx volume can be attached to any participating node in the cluster, although it can be attached to only one node at any given point in time. The node where the Portworx volume is attached is deemed the transaction coordinator and all I/O access to the volume is arbitrated by that node.
Attach is idempotent - multiple attach calls of a volume on the same node will return success. Attach on a node will return a failure, if the device is attached on a different node.
The following command will instantiate a virtual block device in the host namespace and mount it under propagated mount location. The mounted volume is then bind mounted under /data in the busybox container.
docker run -it -v my_portworx_vol:/data busybox c
Running it again will create a second instance of busybox, another bind mount and the Portworx volume reference count will be at 2. Both containers need to exit for the Portworx volume to be unmounted (and detached).
Umount operation unmounts the Portworx volume from the propagated mount location. If this is the last surviving mount on a volume, then the volume is detached as well. Once succesfully unmounted the volume can be mounted on any other node in the system.
The docker plugin API does not have an Detach call. The Detach call is called internally via Unmount on the last unmount call for the volume.
Detach operation involves unexporting the virtual block device from the host namespace. Similar to attach, this is again accomplished via the Portworx container and does not require any external protocol. Detach is idempotent, multiple calls to detach on the same device will return success. Detach is not allowed if the device is mounted on the system.
Remove will delete the underlying Portworx volume and all associated data. The operation will fail if the volume is mounted.
The following command will remove the volume
docker volume rm my_portworx_vol
The Portworx volume driver identifies itself as a
global driver. Portworx operations can be executed on any node in the cluster. Portworx volumes can be used and managed from any node in the cluster.