👋 Hey everyone! SST is great, I've been researching it for the past few days – super impressed by the live reload feature. We're just about to embark on the serverless journey and are quickly evaluating the tooling/solutions out there. One thing though that is a bit of an open question is: how would SST work with a modular monorepo structure using lerna or yarn? I've read this handy guide and looked at the serverless-lerna-yarn-starter, but it uses

serverless.yml

everywhere and not

sst.json

. Does anyone have an example with SST? Or am I just overthinking it and it will just work fine if I just use SST in each

service

?

Welcome @Akos!

Ok, but then how can I individually deploy functions? Running

sst deploy

at the root would deploy all lambdas?

Ppl often need multiple Serverless Framework projects because each projects is 1 CloudFormation stack.

Ah, ok, got it. So in CI with lerna I'd just run

sst deploy <stack>

for the stacks / services that changed?

You don’t need lerna, you can just run

sst deploy

🧙

A bit of back story on the lerna guide, how ppl normally deploy their SLS services in CI are like

cd services/serviceA
sls deploy
cd ../..
cd services/serviceB
sls deploy
cd ../..

Ok, but how does SST understand my function's dependencies and when it only changes 1? I don't want to bundle every lambdas dependencies with every lambda.

So, to speed it up, you use lerna to figure out which ones did not change, and skip them

Ah, ok I didn't know that. Thanks!

sounds good! Let me know if you have any questions.

One final question on this structure stuff: Do you still think this "A practical approach" makes sense with SST? i.e. not having 2 repos, one for resources the other for services, but just one that contains all infra + all services. It seems like it would work quite nicely with SST? Or is the point of 2 repos to have slow-changing infra + quick changing business logic?

Oh yeah.. use 1 repo

ah cool, that simplifies things! so ephemeral envs would still be possible? we may use things like RDS and provisioning a database is quite slow to do on every PR. (as I mentioned I've yet to see seed.run in action)

Yeah if you want to conditionally provision RDS you can do something like:

if (stage === 'prod' || stage === 'dev') {
  new rds.Cluster(...)
}

Yeah, but in the case of an ephemeral env it would still need a DB so I'd want it to connect to a shared DB for quick spin-up

Yeah so in the code snippet above, you only create the RDS cluster for ‘dev’ and ‘prod’

ok, I guess I should probably try it out 🙂 I was just thinking through how I can get the DB credentials into a different stage and into a different stack's config, but I think I can sort that out with CDK.

I’m not too familiar with RDS, but you can do something like:

let dbCluster;
if (stage === 'prod' || stage === 'dev') {
  dbCluster = new rds.Cluster(...)
}
else if (stage.startsWith('pr')) {
  dbCluster = rds.fromDatabaseClusterAttributes(...)
}

Great, yeah, that's what I had in mind.

so for ephemeral env, you can reference the ‘dev’ cluster

Thanks for your answers and patience! 🙏

👍