https://pulumi.com logo
Title
w

wonderful-napkin-50018

03/29/2021, 7:32 PM
I want to define a Fargate
Cluster
in one Stack and reference it in a
FargateService
in another stack. How do I do this with awsx? The FargateService expects a whole Cluster argument, not just an ID or name. And I don't see how I can "get" an existing cluster with
awsx.ecs.Cluster
. Do I have to export the whole cluster from the first stack?
m

millions-furniture-75402

03/29/2021, 7:37 PM
You don’t have to export the whole cluster, I recall this being a little painful to figure out which resources you’ll need. Let me see if I still have my reference project I solved a similar problem with. Our cluster was declared in cloudformation and we could only use it as it, but had a service definition in a pulumi project use it.
l

little-cartoon-10569

03/29/2021, 7:38 PM
Getting ComponentResources from one stack to another is not easy. There is no built-in way to do it. Each time I've migrated an awsx resource to aws one, this is been the (or at least, a) driving factor.
💯 1
w

wonderful-napkin-50018

03/29/2021, 7:40 PM
Ah, I just found something mentioned in a Github issue: https://www.pulumi.com/docs/guides/crosswalk/aws/ecs/#using-an-existing-ecs-cluster
I will try this. No idea why I didn't find this in the docs myself.
m

millions-furniture-75402

03/29/2021, 7:44 PM
const cluster = new awsx.ecs.Cluster(`${appName}-cluster`,{
  cluster: aws.ecs.Cluster.get(`${appName}-local`, "sandbox-cluster"),
  vpc: vpc,
});
Any my service looked like:
const appService = new awsx.ecs.FargateService(`${appName}-service`, {
  assignPublicIp: false,
  cluster,
  deploymentMinimumHealthyPercent: 0,
  deploymentMaximumPercent: 100,
  desiredCount: 1,
  securityGroups: [appSecurityGroup],
  subnets: vpc.privateSubnetIds,
  taskDefinitionArgs: {
    containers: {
      grafana: {
        image: awsx.ecs.Image.fromDockerBuild(`${appName}-image`, {
          context: "./src",
          dockerfile: "./src/grafana.Dockerfile",
        }),
        logConfiguration: {
          logDriver: "awsfirelens",
          options: {
            "Name": "tail",
            "region": "us-east-1",
            "auto_create_stream": "true",
          }
        },
        portMappings: [appTargetGroup],
      }
    },
    taskRole: applicationRole,
  }
});
w

wonderful-napkin-50018

03/29/2021, 7:53 PM
Thanks, I will try it 🙂
@millions-furniture-75402 How do you import the VPC from another stack? I can't get this to work, every VPC I import is missing subnets and other info from the original VPC.
l

little-cartoon-10569

03/29/2021, 10:53 PM
You can't import awsx VPCs (the ones with subnets). You need to handle it manually, using separate aws VPCs and subnets :(
w

wonderful-napkin-50018

03/29/2021, 11:05 PM
There is no way to reuse a VPC resource across stacks?
I found
aws.ec2.getVpc()
which supposedly does what I want but I cannot pass my
vpcId
input to it because it's not a string. I am very confused.
I guess if I could get an
aws.ec2.Vpc
I could make an
awsx.ec2.Vpc
with it, right? But how...
l

little-cartoon-10569

03/29/2021, 11:34 PM
Yes, you can import an aws VPC with a VPC id (it is a string output, you use that), and you can build an unmanaged awsx stack using that.
So you can achieve it. I found it easier to refactor my projects so that everything that needs my awsx VPC is in the one project, but that's not a requirement.
The awsx VPC class provides a method fromExistingId() for the 2nd task.
The first task is even easier: just pass the Output<string> into the Vpc.get() method. Don't use getVpc(), that does something different.
getVpc() is a wrapper around the AWS SDK EC2 client's getVpc() function, which returns a GetVpcResult. It's not often used in pure Pulumi code, it's for those outlying cases where you need extra values from the SDK that Pulumi doesn't expose.
w

wonderful-napkin-50018

03/30/2021, 12:29 AM
@little-cartoon-10569 I tried using
fromExistingId()
but found that this is not really importing any of the VPC data like subnets:
const vpcId = infra.requireOutputValue('vpcId');
const vpc = awsx.ec2.Vpc.fromExistingIds('staging', { vpcId });
export const vpcSubnets = vpc.publicSubnetIds;
vpcSubnets will be empty again
Same problem with
Vpc.get()
. It's as if none of them do any actual importing and they all just create an empty object with the vpcId.
l

little-cartoon-10569

03/30/2021, 12:32 AM
Yes, that's right. The subnets are not loaded, just the VPC is.
The subnets aren't ever really loaded. Just promises to them. For your requirements, you will probably need to export the important subnet IDs from one project, and import them into another.
w

wonderful-napkin-50018

03/30/2021, 12:55 AM
Hm that's unfortunate. I really expected I could get the same VPC object with all the stuff attached by using StackReference. I will try carrying the subnets over manually then and hope that's enough.
l

little-cartoon-10569

03/30/2021, 12:58 AM
The subnets created by an awsx VPC aren't quite first class citizens even in the same stack. They're wrapped by promises. The code that creates them in the awsx package runs from within a
then
clause, rather than within the "engine" like all the other resources.
If you need to use subnets frequently, you may find that creating them yourself, and not using the awsx package, gives you more power and flexibility.
c

cool-fireman-90027

03/30/2021, 12:42 PM
Here is an example using awsx package to build the vpc in typescript and then using that same vpc with ecs. built in python.
m

millions-furniture-75402

03/30/2021, 1:14 PM
Yeah, everything that’s said here, I put my private and public subnets in configuration then use fromExisting
That said, I do it in our “base infrastructure” and export the VPC Id and subnets as outputs so the “child” stacks can recreate the VPC easier without their own configuration.