Placement in Swarm

Managing service placement in Docker Swarm ensures that each workload runs on the best-suited node. By default, Swarm distributes tasks evenly across any node with available resources. In real-world clusters, however, you often have nodes optimized for different workloads:

And services such as:

Without explicit placement constraints, Swarm might schedule a CPU-heavy task on a memory-optimized node or vice versa, causing contention and inefficiency.

1. Labeling Nodes

Assign key-value labels to each node to reflect its resource profile:

docker node update --label-add type=CPU-optimized    worker1
docker node update --label-add type=memory-optimized worker2
docker node update --label-add type=GP               worker3

docker node inspect --format '{{ .Description.Hostname }}: {{ .Spec.Labels }}' worker1

Below is a diagram illustrating the labeled nodes and associated workloads:

2. Applying Placement Constraints

Use the --constraint flag with docker service create to bind services to nodes based on labels or built-in properties.

2.1 Match a Label

Schedule CPU-intensive and memory-intensive services on their respective optimized nodes:

docker service create \
  --name batch-processing \
  --constraint 'node.labels.type==CPU-optimized' \
  batch-processing

docker service create \
  --name realtime-analytics \
  --constraint 'node.labels.type==memory-optimized' \
  realtime-analytics

2.2 Exclude a Label

Prevent the web service from running on memory-optimized nodes:

docker service create \
  --name web \
  --constraint 'node.labels.type!=memory-optimized' \
  web

This ensures web is placed on either the CPU-optimized or general-purpose node.

2.3 Using Built-in Node Properties

You can also constrain by node role. For example, to run a service only on worker nodes (excluding managers):

docker service create \
  --name worker-only-service \
  --constraint 'node.role==worker' \
  alpine:latest ping docker.com

3. Summary of Constraints

By combining node labels with placement constraints, you guarantee that each workload in your Swarm cluster runs on the most suitable hardware, improving both performance and resource utilization.

Links and References

• Docker Swarm Overview
• Service Create reference
• Docker Engine CLI