I'm currently using AWS ECS Fargate to run my Express Node TypeScript web application and I've noticed an issue with how slowly my service scales. My setup has 1 vCPU and runs 2 tasks. I've configured a scaling alarm that should trigger when CPU utilization exceeds 40%, but despite having data points above this threshold, it takes about 3 minutes for the alarm to actually change to the alarm state when there's a spike in traffic. Can anyone explain why this delay happens? Is there anything I can do to speed up the response time?
1 Answer
The delay you're experiencing is pretty common with ECS and Fargate. ECS publishes CPU and memory metrics every minute, but CloudWatch can have a lag of 1 to 2 minutes before showing those datapoints. To address this, consider a few options:
1. If your metrics support it, set your alarm period to 30 seconds instead of 60.
2. Implement Step Scaling to allow for smaller scale-outs when usage increases.
3. Use Target Tracking Scaling; it keeps utilization around your target without you needing to fiddle with thresholds.
4. Enable Container Insights for more granular, quicker data, but keep in mind this may increase your CloudWatch costs.
5. You could also pre-warm tasks if you anticipate traffic spikes at certain times, such as before a scheduled rush.
Thanks for the tips! I do have Container Insights enabled, but I'm not sure if I set it up correctly for my alarm. I was using the default settings with a 60-second period. Should I switch the alarm to utilize the Container Insights metrics and configure it for a 30-second period?