variablesResolutionMode: 20210326 failing to resolve variable with multiple defaults

I have been working on upgrading my project from v1 to v2, and the variable resolution broke on a certain condition. When there are multiple defaults, if the second one fails to resolve, the whole resolution will fail.

serverless.yml

service: aws-node-project

frameworkVersion: '2'
variablesResolutionMode: 20210326

provider:
  name: aws
  runtime: nodejs12.x
  stage: ${opt:stage, env:STAGE, env:AWS_STAGE, "local"}
  lambdaHashingVersion: 20201221

functions:
  hello:
    handler: handler.hello

npx serverless package output

❯ npx serverless package

 Serverless Error ----------------------------------------

  Cannot resolve serverless.yml: "provider.stage" property is not accessible (configured behind variables which cannot be resolved at this stage)

  Get Support --------------------------------------------
     Docs:          docs.serverless.com
     Bugs:          github.com/serverless/serverless/issues
     Issues:        forum.serverless.com

  Your Environment Information ---------------------------
     Operating System:          linux
     Node Version:              14.17.1
     Framework Version:         2.54.0 (local)
     Plugin Version:            5.4.3
     SDK Version:               4.2.6
     Components Version:        3.15.0

Running STAGE=foo npx serverless package will work correctly since then the first default will resolve.

Installed version

Framework Core: 2.54.0 (local)
Plugin: 5.4.3
SDK: 4.2.6
Components: 3.15.0

[perrin4869]

I see! Yeah I’ve been using provider.stage since like forever, didn’t notice sls:stage until recently 😃 Indeed changing it to provider.defaultStage would clarify the intended use of the option. Still, there are practical difficulties not being able to use the env STAGE as a default for stage.

Usually I do not run serverless directly, but rather define a script inside package.json, like "deploy": "serverless deploy --verbose". This would mean changing the scripts to be rather "deploy": "serverless deploy --verbose --stage $STAGE". The difficulty here is, what if I run this locally, and I do not have $STAGE defined? In this case I want to default to "local" or "development" or something, which is what I was achieving by doing stage: ${opt:stage, env:STAGE, "local"}.

Maybe stage could be a special case, where it does rely on an environment variable when setting it the first time?

Edit: Of course I can always work around this issue to use serverless as is intended, but I think there is practicality in allowing the stage to depend on the environment and would prefer it to stay like that.

[medikoo]

The staging pipeline will have STAGE=staging and the production pipeline will have STAGE=production. I do need to reevaluate the usage of AWS_STAGING variable.

Do you rely on one CI configuration for different pipelines? If such you may also solve it by keeping setting the STAGE, but by propagating it to sls via --stage=$STAGE flag.

Note that provider.stage setting in a configuratioin is technically a setting for default stage. It’s a last resort fallback if there’s no --stage flag passed (the default is normally dev, but if you wish to change it, it’s when provider.stage should be configured). We also have proposal of renaming this to provider.defaultStage -> https://github.com/serverless/serverless/issues/8684 so it’s not confusing, but we’re also considering totally removing support for this property (as it’s hard to find a good use case for dev not being a default stage)

Additionally if you want to rely on resolved stage in configuration you should simply refer to sls:stage (until it was introduced, developers quite often ensured that actual stage was resolved on provider.stage, and then referring to that in variables, but it was just dirty workaround for fact that there wasn’t anything like sls:stage available. It is now)