Run Task versus Service¶

Goal

Use Task Definition in two ways:
- Run Task for quick tests or one-off jobs.
- Service for long-running workloads, ALB, and auto scaling.

Run Task¶

Run Task behaviour

Run a Task to test a Task Definition or execute a one-off job.

Typical command:

aws ecs run-task \
  --cluster ibtisam-iq \
  --launch-type FARGATE \
  --task-definition portfolio-task \
  --network-configuration "awsvpcConfiguration={
    subnets=[subnet-abc,subnet-def],
    securityGroups=[sg-ecs],
    assignPublicIp=ENABLED
  }" \
  --count 1

Understand parameters:

--cluster: choose Cluster.
--task-definition: choose Task Definition revision.
--launch-type: choose Fargate or EC2.
--network-configuration: choose subnets and security groups.
assignPublicIp=ENABLED: assign a public IP for direct access.
--count: run a small number of Tasks.

Behaviour:

Start Tasks once.
Allow Tasks to exit or fail without replacement.
Do not attach Tasks to load balancers.
Use for testing images, validating database connectivity, running batch jobs.

Service¶

Service behaviour

Use a Service to maintain a desired number of Tasks and integrate with ALB.

Typical command:

aws ecs create-service \
  --cluster ibtisam-iq \
  --service-name portfolio-svc \
  --task-definition portfolio-task \
  --desired-count 2 \
  --launch-type FARGATE \
  --network-configuration "awsvpcConfiguration={
    subnets=[subnet-private-a,subnet-private-b],
    securityGroups=[sg-ecs],
    assignPublicIp=DISABLED
  }" \
  --load-balancers "targetGroupArn=arn:aws:elasticloadbalancing:...,containerName=portfolio,containerPort=8080"

Behaviour:

Create Tasks from the Task Definition.
Maintain desired-count by replacing failed Tasks.
Integrate with ALB or NLB.
Support auto scaling policies.

Compare to Kubernetes:

Service (ECS) ≈ Deployment + Service (LB) + HPA.
desired-count ≈ Deployment replicas.
Auto scaling policies ≈ HPA rules.

Choosing between Run Task and Service¶

Practical rules

Use Run Task for short-lived experiments and manual tests.
Use Service for long-lived workloads that must always be available.
Use Run Task as the equivalent of kubectl run or kubectl port-forward.
Use Service as the equivalent of a Deployment fronted by a Service and Ingress.

Summary:

Run Task = no self-healing, no LB, no auto scaling.
Service = self-healing, LB capable, auto scaling available.