Docker Swarm allows managing Docker clusters remotely. The key components are swarm managers, swarm nodes, and a scheduler. Swarm managers oversee nodes in the cluster using Docker APIs. The scheduler uses strategies and filters to determine where to place containers on nodes. Discovery services help register and discover nodes in the cluster.

1. Docker Swarm
Author : Rajdeep Dua, VMware
Twitter : @rajdeepdua
Last Updated : March 25 2015

2. Overview
• Allows to manage Docker Cluster Remotely
• Key Components
– Swarm Manager
– Swarm Node
– Scheduler
– Discovery

3. Architecture Diagram

4. Swarm Manager
• Manages the Nodes in the Cluster
• Uses Docker APIs to access the Docker Daemon
running on each node
• Nodes are added to the Swarm manager by
callback from Discovery Service’s fetch() function
• Elements of a Swarm Cluster Manager
– Event Handler
– Map of Nodes
– Store
– Options

5. Swarm Node
• Runtime Instance representing a node in the Cluster.
• Talks to the actual Host using Docker Go Client
• Created from a DiscoveryEntry fetched from a Discovery Service
• Elements of Swarm Node
– Id
– IP Address (of remote host)
– Map of Containers
– Map of Images
– Health state of the node
– Total CPUs
– Used CPUs
– Total Memory
– Used Memory

6. Scheduler
• Responsible for scheduling a container on a
Node
• Pluggable architecture – Bring your own
scheduler
• Elements of a Scheduler
– Placement Strategy Instance
– Array of Filters

7. Scheduler .. contd
• The `Docker Swarm` scheduler comes with multiple strategies.
• These strategies are used to rank nodes using a scores
computed by the strategy. `Docker Swarm` currently supports
2 strategies:
– BinPacking Strategy
– Random Strategy
Usage : You can choose the strategy you want to use with the `--
strategy` flag of `swarm manage`

8. Scheduler .. contd
BinPacking strategy
• The BinPacking strategy will rank the nodes using their CPU and RAM
available and will return the node the most packed already.
• This avoid fragmentation, it will leave room for bigger containers on
unused machines.
Random strategy
• The Random strategy, as it's name says, chooses a random node, it's used
mainly for debug

9. Scheduler Filters
• Scheduler uses the following filters for container
placement on a node
– Affinity Filter
– Constraint Filter : When creating a container, the user
can select a subset of nodes that should be
considered for scheduling by specifying one or more
sets of matching key/value pairs.
– Port Filter : Select node where public port required by
the container is not already used
– Health Filter : Select only healthy nodes from the
cluster

10. Swarm Store
• Stores the state of the Cluster.
• Currently implemented as a JSON file
• State is loaded in memory when the cluster starts
• Lifecycle events of the store
– Get state for a key
– Store the state of a container
– Load all the data stored
– Replace the state of the key with a new state
– Delete the state

11. Discovery Service
• Helps Swarm Manager discover nodes
• Three main functions
– Register : Registers a new node
– Watch : Callback method for Swarm Manager
when a new Node is added to the Discovery
Service
– Fetch : Fetch the List of Entries

12. Discovery Services
• Token Based
• etcd based
• Zookeeper based
• File Based
• Consul Based
• Bring your own?

13. Docker Swarm Architecture - Exploded

14. Token Based Discovery Service
• Based on REST End Point hosted on Docker-
hub
• URL Queried for entries is
DISCOVERY_URL = "https://discovery-stage.hub.docker.com/v1"
HTTP GET https://discovery-stage.hub.docker.com/v1/clusters/<token-id>

15. Discovery Fetch Entries

16. Swarm CLI
• Allows Setting up and Managing the Cluster
create a cluster
$ docker run --rm swarm create
6856663cdefdec325839a4b7e1de38e8
On each nodes, start the swarm agent
$ docker run -d swarm join --addr=<node_ip:2375> token://<cluster_id>

17. Swarm CLI
Start the manager on any machine / Laptop to manage the
cluster
$ docker run -d -p <swarm_port>:2375 swarm manage token://<cluster_id>
Use the regular docker cli
$ docker -H tcp://<swarm_ip:swarm_port> info
$ docker -H tcp://<swarm_ip:swarm_port> run ...
$ docker -H tcp://<swarm_ip:swarm_port> ps
$ docker -H tcp://<swarm_ip:swarm_port> logs ...
…
List nodes in your cluster
$ docker run --rm swarm list token://<cluster_id>
<node_ip:2375>

18. Swarm APIs
• Compatible with Docker Remote API with
some differences
• Containers
GET /containers/(id)
POST /containers/create
DELETE /containers/(id)
POST /container/(id)/start
additional parameters passed in the POST body

19. Swarm APIs…
Images
GET /images/json
POST /images/create
DELETE /images/(id)
GET /images/search

20. Summary
• Docker Swarm allows you to use native Docker
CLI commands and APIs to manage containers
across a cluster of nodes
• It has a basic scheduler which can be
enhanced
• Pluggable Discovery mechanism