aws
[ECS] [feature]: Amazon ECS enhanced capacity management
Community Note
- Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request
- Please do not leave "+1" or "me too" comments, they generate extra noise for issue followers and do not help prioritize the request
- If you are interested in working on this issue or have submitted a pull request, please leave a comment
Tell us about your request
Amazon ECS will deliver a new capacity management solution that combines the simplicity of AWS Fargate’s operational model with the breadth of capabilities supported by Amazon EC2 instances. Customers will be able to use the same simple interface as Fargate by only needing to request vCPU and memory for their tasks, with ECS automatically provisioning, patching, scaling, and cost-optimizing fully-managed compute. At the same time, customers who need more control, will have the ability to select specific EC2 instance types and sizes. Customers can also enable privileged Linux capabilities, and run their desired security and monitoring agents as Daemons. Salient features of enhanced capacity management:
- Run workloads requiring specific compute capabilities, such as accelerated compute (GPUs), CPU instruction sets, network-optimized or burstable instances, while retaining the operational benefits of AWS-managed compute infrastructure and automatic security patching
- Leverage breadth of EC2 billing and purchase options for fully-managed compute, including Reserved Instances, Spot Instances, Compute and Instance Savings Plans, On-demand Capacity Reservations, and Capacity Blocks for ML
- Rapid and cost-effective autoscaling with smart task placement and rebalancing to ensure your applications run on highly available, performant, and cost-efficient compute
- Improved security posture with EC2 instances preconfigured with AWS security best practices and automatically updated with the latest security patches. Use event windows for more control over instance patching and updates
- Run managed Daemon sets, and optionally configure privileged Linux capabilities, including CAP_NET_ADMIN, CAP_SYS_ADMIN, CAP_BPF, etc., to run your desired monitoring, observability, and security solutions
Popular issues that will be addressed by this capability
AWS Fargate GPU Support: When is GPU support coming to fargate? #88 [Fargate] [request]: compute optimized options #1030 [Fargate] [request]: Burstable CPU #163 [Fargate] [request]: offer high-performance network options #715
[Fargate] [request]: Allow privileged mode #1000 [Fargate] [request]: Provide the ability to use ebpf on fargate instances #1027
[Rebalancing] Smarter allocation of ECS resources #105 ECS tasks re-balancing on autoscaling #42
[ECS] Full support for Capacity Providers in CloudFormation #631
Which service(s) is this request for? Amazon ECS, AWS Fargate
Amazon ECS team is excited to build this feature, and we look forward to your feedback.
Awesome to see all these features being worked on! As an avid Fargate user, it would also be great if the issue https://github.com/aws/containers-roadmap/issues/938 could get some love. If you are using EC2 with ECS, you can do something like (CloudFormation)
LinuxParameters:
Tmpfs:
- ContainerPath: /tmp
MountOptions:
- rw
Size: 200
that is not possible with Fargate. You cannot specify specific mounting options:
MountPoints:
- SourceVolume: tmp
ContainerPath: /tmp
In the above example the /tmp folder will be mounted by root, unless the docker image has this as a volume with another user configured.
Enhancing this feature would make Fargate so much more useful.
We’re excited to announce the launch of Amazon ECS Managed Instances, a new fully managed compute option designed to eliminate infrastructure management overhead while giving you access to the full capabilities of Amazon EC2. By offloading infrastructure operations to AWS, ECS Managed Instances helps you quickly launch and scale your workloads, while enhancing performance and reducing your total cost of ownership. ECS Managed Instances is designed to give customers the best of both worlds - the simplicity of AWS Fargate, combined with the powerful flexibility of Amazon EC2. To learn more, visit the What's New post, feature page, documentation, and AWS News launch blog.
ECS Managed Instances delivers support for several highly requested ECS/Fargate features, including the following linked issues, and we have a rich roadmap planned. As always, we appreciate your engagement and look forward to your feedback.
AWS Fargate GPU Support: When is GPU support coming to fargate? #88 [Fargate] [request]: compute optimized options #1030 [Fargate] [request]: Burstable CPU #163 [Fargate] [request]: offer high-performance network options #715
[Fargate] [request]: Allow privileged mode #1000 [Fargate] [request]: Provide the ability to use ebpf on fargate instances #1027
[Rebalancing] Smarter allocation of ECS resources #105 ECS tasks re-balancing on autoscaling #42
[ECS] Full support for Capacity Providers in CloudFormation #631
It is great to see new products added to ECS, but i do hope that the original intent of all those Issues (-> Fargate) is not being forgotten. This may be great as a workaround, but for customers who really do not care about instances this does not solve the original feature request for Fargate containers. So i do hope that the referenced Fargate issues are not closed, since it still does not support those features.
Is Fargate a mess (interally at AWS) or why does it seem to be impossible to implement those highly requested features to Fargate?
@AbhishekNautiyal In relation to the above, I did a quick check of the documentation and could not find a reference to eBPF probes as highlighted in https://github.com/aws/containers-roadmap/issues/1027
Could you kindly point to where these will be an option (thinking from a security standpoint where we wish to monitor systemcalls)
Lets put it simply. GCP Cloud Run has ephemeral container/serverless hosting AND with GPU as an option: https://cloud.google.com/run/pricing?hl=en
This was what we were expecting with Fargate
@AbhishekNautiyal In relation to the above, I did a quick check of the documentation and could not find a reference to eBPF probes as highlighted in #1027
Could you kindly point to where these will be an option (thinking from a security standpoint where we wish to monitor systemcalls)
Hi, This is covered in documentation here. You can specify Linux parameters as part of your container definitions (documented here) - just make sure that your requiresCompatibilities in the ECS task definition is set to either EC2 or MANAGED_INSTANCES (because FARGATE does not support elevated privileges). Let us know if you face any issues.
How exactly does Managed Instances address rebalancing concerns raised in https://github.com/aws/containers-roadmap/issues/42? Is it possible to achieve spread across AZ with optimal binpack in each zone?
