[Question] How to manage vm/cluster resource for non ecs tasks?

[jespada]

Hi Humans, thanks for open sourcing this repo :).
I would like to know how do you handle non ecs tasks/containers regarding resource available, in particular the "standalone" docker container ecs-logs that is lunched when the vm start, do you trust the foot print will be low enough that you can dismiss it, like ecs-agent, or you impose some kind of constrains(--memory, --cpu..)?
AFAIU the ECS scheduler wont take those into account when scheduling work .

[achille-roussel]

Indeed, we just don't set limits on ecs-agent or ecs-logs. These don't use too much resources and we don't run at 100% resource usages so there's always some room for OS-related processes.

There are other things that ECS doesn't have control over on our infrastructure, like the datadog agent or lower-level components like journald so it's a good idea to leave a bit of room on the hosts for these to execute. Usually we don't go over 90% reservation on the hosts.