Manage nodes in a swarm

もくじ

As part of the swarm management lifecycle, you may need to:

List nodes

To view a list of nodes in the swarm run docker node ls from a manager node:

$ docker node ls

ID                           HOSTNAME  STATUS  AVAILABILITY  MANAGER STATUS
46aqrk4e473hjbt745z53cr3t    node-5    Ready   Active        Reachable
61pi3d91s0w3b90ijw3deeb2q    node-4    Ready   Active        Reachable
a5b2m3oghd48m8eu391pefq5u    node-3    Ready   Active
e7p8btxeu3ioshyuj6lxiv6g0    node-2    Ready   Active
ehkv3bcimagdese79dn78otj5 *  node-1    Ready   Active        Leader

The AVAILABILITY column shows whether or not the scheduler can assign tasks to the node:

  • Active means that the scheduler can assign tasks to the node.
  • Pause means the scheduler doesn't assign new tasks to the node, but existing tasks remain running.
  • Drain means the scheduler doesn't assign new tasks to the node. The scheduler shuts down any existing tasks and schedules them on an available node.

The MANAGER STATUS column shows node participation in the Raft consensus:

  • No value indicates a worker node that does not participate in swarm management.
  • Leader means the node is the primary manager node that makes all swarm management and orchestration decisions for the swarm.
  • Reachable means the node is a manager node participating in the Raft consensus quorum. If the leader node becomes unavailable, the node is eligible for election as the new leader.
  • Unavailable means the node is a manager that can't communicate with other managers. If a manager node becomes unavailable, you should either join a new manager node to the swarm or promote a worker node to be a manager.

For more information on swarm administration refer to the Swarm administration guide.

Inspect an individual node

You can run docker node inspect <NODE-ID> on a manager node to view the details for an individual node. The output defaults to JSON format, but you can pass the --pretty flag to print the results in human-readable format. For example:

$ docker node inspect self --pretty

ID:                     ehkv3bcimagdese79dn78otj5
Hostname:               node-1
Joined at:              2016-06-16 22:52:44.9910662 +0000 utc
Status:
 State:                 Ready
 Availability:          Active
Manager Status:
 Address:               172.17.0.2:2377
 Raft Status:           Reachable
 Leader:                Yes
Platform:
 Operating System:      linux
 Architecture:          x86_64
Resources:
 CPUs:                  2
 Memory:                1.954 GiB
Plugins:
  Network:              overlay, host, bridge, overlay, null
  Volume:               local
Engine Version:         1.12.0-dev

Update a node

You can modify node attributes to:

Change node availability

Changing node availability lets you:

  • Drain a manager node so that it only performs swarm management tasks and is unavailable for task assignment.
  • Drain a node so you can take it down for maintenance.
  • Pause a node so it can't receive new tasks.
  • Restore unavailable or paused nodes availability status.

For example, to change a manager node to Drain availability:

$ docker node update --availability drain node-1

node-1

See list nodes for descriptions of the different availability options.

Add or remove label metadata

Node labels provide a flexible method of node organization. You can also use node labels in service constraints. Apply constraints when you create a service to limit the nodes where the scheduler assigns tasks for the service.

Run docker node update --label-add on a manager node to add label metadata to a node. The --label-add flag supports either a <key> or a <key>=<value> pair.

Pass the --label-add flag once for each node label you want to add:

$ docker node update --label-add foo --label-add bar=baz node-1

node-1

The labels you set for nodes using docker node update apply only to the node entity within the swarm. Do not confuse them with the Docker daemon labels for dockerd.

Therefore, node labels can be used to limit critical tasks to nodes that meet certain requirements. For example, schedule only on machines where special workloads should be run, such as machines that meet PCI-SS compliance.

A compromised worker could not compromise these special workloads because it cannot change node labels.

Engine labels, however, are still useful because some features that do not affect secure orchestration of containers might be better off set in a decentralized manner. For instance, an engine could have a label to indicate that it has a certain type of disk device, which may not be relevant to security directly. These labels are more easily "trusted" by the swarm orchestrator.

Refer to the docker service create CLI reference for more information about service constraints.

Promote or demote a node

You can promote a worker node to the manager role. This is useful when a manager node becomes unavailable or if you want to take a manager offline for maintenance. Similarly, you can demote a manager node to the worker role.

メモ

Regardless of your reason to promote or demote a node, you must always maintain a quorum of manager nodes in the swarm. For more information refer to the Swarm administration guide.

To promote a node or set of nodes, run docker node promote from a manager node:

$ docker node promote node-3 node-2

Node node-3 promoted to a manager in the swarm.
Node node-2 promoted to a manager in the swarm.

To demote a node or set of nodes, run docker node demote from a manager node:

$ docker node demote node-3 node-2

Manager node-3 demoted in the swarm.
Manager node-2 demoted in the swarm.

docker node promote and docker node demote are convenience commands for docker node update --role manager and docker node update --role worker respectively.

Install plugins on swarm nodes

If your swarm service relies on one or more plugins, these plugins need to be available on every node where the service could potentially be deployed. You can manually install the plugin on each node or script the installation. You can also deploy the plugin in a similar way as a global service using the Docker API, by specifying a PluginSpec instead of a ContainerSpec.

メモ

There is currently no way to deploy a plugin to a swarm using the Docker CLI or Docker Compose. In addition, it is not possible to install plugins from a private repository.

The PluginSpec is defined by the plugin developer. To add the plugin to all Docker nodes, use the service/create API, passing the PluginSpec JSON defined in the TaskTemplate.

Leave the swarm

Run the docker swarm leave command on a node to remove it from the swarm.

For example to leave the swarm on a worker node:

$ docker swarm leave

Node left the swarm.

When a node leaves the swarm, Docker Engine stops running in Swarm mode. The orchestrator no longer schedules tasks to the node.

If the node is a manager node, you receive a warning about maintaining the quorum. To override the warning, pass the --force flag. If the last manager node leaves the swarm, the swarm becomes unavailable requiring you to take disaster recovery measures.

For information about maintaining a quorum and disaster recovery, refer to the Swarm administration guide.

After a node leaves the swarm, you can run docker node rm on a manager node to remove the node from the node list.

For instance:

$ docker node rm node-2

Learn more