CODEWITHSUNDEEP
amazon-ecsaws-auto-scaling
vmax
vmax

Reputation: 41

ECS-AutoScaling tasks without delay

For a service running on ECS, we have configured autoscaling at task-definition(based on metrics such as memory usage, hit count, etc.). However, when the memory reaches the set limit, there is a delay in launching a new task, causing some hits to be missed during this time. How can we overcome this lag in new task placement, and are there any alternative approaches?

Upvotes: 0

Views: 89

Answers (2)

AxB
AxB

Reputation: 695

In parallel to what Mark has mentioned,

Also check the health check attributes of a target group of your load balancer(assuming that you are using one). A newly registered target must pass health checks before traffic can be routed to it.

Lets assume following health check settings for an alb's target group,

  • Healthy Threshold = 3
  • Interval = 60 seconds

This means in a perfect scenario, newly registered target will be ready to serve between 2-3 minutes.

Upvotes: 0

Mark B
Mark B

Reputation: 200998

How much of a delay are you seeing exactly? Between the delay in the metrics being reported, and the delay in the ECS auto-scaling action, and then the time it takes for ECS to load the docker image, and the time for the app to start and become healthy before a load balancer will send requests to it. If that is a problem, you should configure a lower threshold for your auto-scaling so that a new task is started a bit before your hard limits are reached.

Upvotes: 0

Related Questions