Skip to content

SKILL.md

Latest commit

 

History

History
122 lines (101 loc) · 6.22 KB

SKILL.md

File metadata and controls

122 lines (101 loc) · 6.22 KB
name docker-swarm
description Docker Swarm mode orchestration - cluster management, service deployment, scaling, node administration, and rolling updates
standards-version 1.6.3

Docker Swarm Orchestration

Manage Docker Swarm clusters, deploy and scale services, administer nodes, and perform rolling updates and rollbacks.

Workflow Diagram

flowchart TD
    A[Start] --> B{Swarm initialized?}
    B -->|No| C[Initialize cluster]
    C -->|docker_swarmInit| D[Add nodes]
    D -->|docker_swarmJoin| E[Cluster ready]
    B -->|Yes| E
    E --> F[Deploy service]
    F -->|docker_serviceCreate| G[Monitor tasks]
    G -->|docker_servicePs| H{Healthy?}
    H -->|No| I[Check logs]
    I -->|docker_serviceLogs| J[Fix and update]
    J -->|docker_serviceUpdate| G
    H -->|Yes| K{Scale needed?}
    K -->|Yes| L[Scale replicas]
    L -->|docker_serviceScale| G
    K -->|No| M{Update needed?}
    M -->|Yes| N[Rolling update]
    N -->|docker_serviceUpdate| O{Update OK?}
    O -->|No| P[Rollback]
    P -->|docker_serviceRollback| G
    O -->|Yes| G
Loading

Trigger

Activate when the user:

  • Asks about Docker Swarm mode or cluster orchestration
  • Wants to initialize, join, or manage a Swarm cluster
  • Needs to create, update, scale, or rollback Swarm services
  • Asks about node management, promotion, demotion, or draining
  • Mentions "swarm init", "service create", "node ls", or "rolling update"

Required Inputs

  • Task type: cluster setup, service management, node administration, or troubleshooting
  • Cluster details (for init/join): advertise address, join token, remote manager address
  • Service details (for create/update): image, replicas, ports, resource limits

Workflow

  1. Assess cluster state - Use docker_nodeLs to check existing nodes and docker_serviceLs to list running services.
  2. Initialize or join - Use docker_swarmInit to bootstrap a new cluster or docker_swarmJoin to add nodes.
  3. Deploy services - Use docker_serviceCreate with appropriate replicas, ports, networks, and resource limits.
  4. Monitor and scale - Use docker_servicePs to check task distribution, docker_serviceLogs for debugging, and docker_serviceScale to adjust replicas.
  5. Update or rollback - Use docker_serviceUpdate for rolling updates and docker_serviceRollback to revert if issues arise.
  6. Manage nodes - Use docker_nodeUpdate to drain nodes for maintenance, docker_nodePromote/docker_nodeDemote for role changes.

Key References

  • Docker CLI: docker swarm init, docker swarm join, docker service create, docker service update, docker service scale, docker node ls, docker node update
  • Swarm port: 2377/tcp (cluster management), 7946/tcp+udp (node communication), 4789/udp (overlay network)
  • Service modes: replicated (specified number of tasks) vs global (one task per node)

Example Interaction

User: "Set up a 3-replica nginx service with a rolling update strategy"

Assistant: First checks if this node is a Swarm manager.

  • Calls docker_nodeLs to verify cluster state
  • If not in swarm mode, calls docker_swarmInit to initialize
  • Calls docker_serviceCreate with name, image, replicas=3, ports, and resource limits
  • Calls docker_servicePs to verify all 3 tasks are running
  • Explains how to update with docker_serviceUpdate --image nginx:new --update-parallelism 1 --update-delay 10s

MCP Usage

Tool When to Use
docker_swarmInit Initializing a new Swarm cluster
docker_swarmJoin Adding worker or manager nodes to the cluster
docker_swarmLeave Removing the current node from the Swarm
docker_swarmJoinToken Retrieving or rotating join tokens
docker_swarmUpdate Updating Swarm-wide settings (task history, cert expiry, autolock)
docker_swarmUnlock Unlocking a locked Swarm manager after restart
docker_swarmUnlockKey Retrieving or rotating the Swarm unlock key
docker_swarmCa Viewing or rotating the cluster root CA certificate
docker_serviceCreate Deploying a new replicated or global service
docker_serviceUpdate Rolling updates to image, replicas, env, or resources
docker_serviceRm Removing services from the Swarm
docker_serviceLs Listing all services and their status
docker_serviceInspect Viewing detailed service configuration
docker_serviceLogs Retrieving service or task logs for debugging
docker_servicePs Checking task distribution across nodes
docker_serviceScale Scaling services up or down
docker_serviceRollback Reverting a service to its previous version
docker_nodeLs Listing all nodes in the cluster
docker_nodeInspect Viewing detailed node information
docker_nodePs Listing tasks running on a specific node
docker_nodeRm Removing nodes from the cluster
docker_nodeUpdate Changing node availability (active/drain) or role
docker_nodePromote Promoting workers to managers for HA
docker_nodeDemote Demoting managers to workers

Common Pitfalls

  1. Single manager - A single-manager Swarm has no fault tolerance. Promote at least 2 additional nodes to manager (odd numbers: 3, 5, 7).
  2. Draining without replacement - Draining a node moves all tasks to other nodes. Ensure sufficient capacity before draining.
  3. Missing resource limits - Services without CPU/memory limits can starve other services on the same node.
  4. Port conflicts - Published ports in ingress mode are cluster-wide. Two services cannot publish the same host port.
  5. Overlay network encryption - Overlay networks are unencrypted by default. Use --opt encrypted for sensitive traffic between nodes.
  6. Token exposure - Join tokens grant cluster access. Rotate them regularly with docker_swarmJoinToken and never commit them to repos.
  7. Autolock disabled - Without autolock, a stolen disk image of a manager node exposes all cluster secrets. Enable with docker_swarmUpdate.

See Also

  • docker-security skill - for general container security hardening
  • docker-networking skill - for overlay network configuration
  • docker-resource-management skill - for CPU/memory limits and monitoring
  • swarm-security rule - automated checks for Swarm security issues