Docker Swarm · Lesson 9 of 10 · 3 min read
Troubleshooting and choosing an orchestrator
Common problems
| Symptom | Check |
|---|---|
Service shows 0/3 replicas |
docker service ps --no-trunc svc — image pull errors (private registry without --with-registry-auth), "no suitable node" (constraints/resources), port conflicts, crash loops |
| Tasks keep restarting | Container exits — docker service logs, healthcheck failing, restart policy |
| Containers on different nodes can't talk | Ports 7946 TCP/UDP and 4789/UDP blocked by security groups/firewalls; MTU issues with VXLAN (overlay adds ~50 bytes) |
| Published port works on some nodes only | Ingress network or IPVS issue on those nodes; host-mode publishing only listens where tasks run |
| Brief errors during deploys | No healthcheck, stop-first order, no graceful shutdown handling |
docker service commands fail: "rpc error ... the swarm does not have a leader" |
Quorum lost — restore managers or --force-new-cluster |
A node shows Down |
Swarm reschedules its tasks elsewhere automatically; investigate the node (Docker daemon, network) |
| Uneven task distribution after adding nodes | No automatic rebalancing — docker service update --force |
Swarm vs Kubernetes vs ECS
| Swarm | Kubernetes (EKS) | ECS | |
|---|---|---|---|
| Setup | Minutes; built into Docker | Complex (managed services help) | AWS-managed, simple |
| Config format | Compose | Manifests, Helm, Kustomize | Task definitions |
| Ecosystem | Small | Huge (operators, CRDs, service meshes, GitOps) | AWS-integrated |
| Autoscaling | None built in | HPA, VPA, Cluster Autoscaler, Karpenter | Service and cluster autoscaling |
| Multi-tenancy, RBAC, policy | Minimal | Rich (RBAC, NetworkPolicy, admission control) | IAM-based |
| Fit | Small teams, small clusters, simple apps | Platforms, many teams, complex needs | AWS-only shops wanting simplicity |
An honest senior answer: Swarm is pleasant for small, simple deployments, but for a product company's platform you'd choose Kubernetes for its ecosystem, extensibility, autoscaling and managed offerings like EKS — or ECS if the team wants AWS-native simplicity.
Enlarge diagram
Swipe sideways to see all of it, or tap to open it full screen.