S3Key of bundled lambda asset changes on every run in Github Actions workflow

[fynnfluegge]

Hi 👋,

I recently moved my cdk deployment to a workflow in github actions.

But on every run, all my lambda functions are updated during cdk deploy (even there is no change), since the S3key and the asset path of the new bundled asset differs on every run with the deployed version:

I expect that only the lambda functions, where the code has changed will be updated. I bet this has something to do with the github actions runner that causes some minor tweaks on every run which leads to a different hash of the asset path and as a result a different S3key.

Has someone bypassed this issue?

Thanks in advance!

[peterwoodworth]

I bet this has something to do with the github actions runner that causes some minor tweaks on every run which leads to a different hash of the asset path and as a result a different S3key.

This was my first thought as well, but I'm not sure exactly what would be causing this - asset hashes should remain consistent across systems. Is it possible your node version is somehow switching?

I recommend testing this on a self-hosted runner and seeing if the issue persists, that could potentially unblock you

[AlessandroVol23]

Hey @fynnfluegge did you had any success with a self-hosted runner? We're still facing this issue unfortunately.
Thanks!

[Exter-dg]

Our Lambdas get updates everytime from Codepipeline. Even when there is no change in the entire pipeline

[sushantmimani]

We're facing this issue too. We are on self-hosted runners but that doesn't seem to change the outcome. Found this but currently not able to re-work this for my use case.
hupe1980/jest-cdk-snapshot#8

[fynnfluegge]

Maybe similar issue as it is in terraform with this as a workaround?
https://github.com/bboe/deterministic_zip

[sushantmimani]

My issue turned out to be that the __pycache__ directory was in .gitignore. So it was missing on the runners, but present locally when calculating the hashes. I added this to my lambda creation and tests have been passing locally and on GHA.

  self.lambda_func = _alambda.PythonFunction(
            self,
            id_,
           ....
               asset_excludes=["__pycache__", "*.pyc", ".pytest_cache"],
            ),
          .....

To identify what was causing the issue, I ran cdk deploy from my local system, downloaded the contents from S3 and compared with the contents when cdk deploy runs from GHA.