add: ECS example template #3915
Conversation
There was a problem hiding this comment.
Thanks for doing this! Tested it out and left some comments which seem to scale up/down the service based on stop and add support for 2 workspaces.
I don't know too much about ECS, but my basic understanding is:
- You create a cluster, typically with AWS instances (VMs)
- You deploy "services" which are containers that run across the VMs
Does this use case aim to provision 1 cluster per workspace, or provision workspaces on an existing cluster, similar to Kubernetes? After the proposed changes are added, this template somehow:
- provisions a cluster when workspace 1 is created + a service + autoscaling group
- provisions services, autoscaling groups on the existing cluster created by workspace 1
It's kinda nice, but doesn't seem like the expected behavior given the Terraform code. I expected it would provision 1 cluster per developer, which seems overkill.
If we don't want this template to create a cluster (at least 1 VM) per developer, we might want to recommend that the Coder admin provisions the cluster (and perhaps the autoscaling group, capacity provider) outside of Coder and then just uses 1 aws_ecs_service per workspace, similar to our K8s examples.
The cluster can still be provisioned with Terraform, just via a single terraform apply, outside of Coder not as a part of each workspace.
|
|
||
| variable "ecs-cluster" { | ||
| description = "Input the ECS cluster ARN to host the workspace" | ||
| default = "" |
There was a problem hiding this comment.
| default = "" | |
| sensitive = true |
We're currently using this to indicate whether this is a developer-level variable or template-wide variable. I assume the intended behavior is that the admin picks which cluster workspaces are deploying to?
I also suggest not setting a default value since it will skip it on coder templates create
Some docs: https://coder.com/docs/coder-oss/latest/templates#parameters
Co-authored-by: Ben Potter <ben@coder.com>
this PR adds an example template for creating a workspace as an Elastic Container Service (ECS) hosted container.
I'm submitting this as a draft because of a current issue with stop/start. the workspace builds fine initially, but does not shut down the resources once stopped. I tinkered with adding the
count = data.coder_workspace.me.start_countargument to the Task Definition resource, but am receiving an odd error when building the template:I haven't been able to debug this error, and would like some extra eyes to assist.