3 ms·

That's very interesting, thank you for typing it up! > Express based project Ahh, so you use a monolith in a lambda. I should have gone this route. I was conc

That's very interesting, thank you for typing it up! > Express based project Ahh, so you use a monolith in a lambda. I should have gone this route. I was concerned about package size (due to lambda limits) and so went the "one function per endpoint" route instead. That seemed to be what AWS was saying you should do but it really screws you in a number of ways, size is the only place where it really helps. Provisioned lambdas, for example, are so much better for a monolith. Did you not have size issues putting all the code in 1 package? That was a real limit I hit when I was (naively) packaging all my code for every endpoint instead of just the handler and the code it needed. I'd be nervous of pushing up against the limit. I guess the escape hatch is moving to a docker container on lambda since that gives you 10GB. > The only thing I kind of dislike is decommissioning of older nodejs runtimes so they aren't available at all for new Lambdas anymore. Yep, I've been burned by this and forced to update. One more question, how do you handle local development? Do you just run Express directly? That's one reason I've leaned into the serverless frameworks out there. My function-per-endpoint setup doesn't have any default scaffolding, running via lambda is the only way to run the code (or one of the project that tries to replicate lambda locally). SST has been the best on this front since you deploy your dev environment like normal (spin up real queues, etc) but the lambdas are just proxying the request to the code running locally on your computer. This has resulted in the least "gotchas" when it comes time to deploy (though arch differences are always a blast).

leptons 1y ago
I abandoned Express when I moved to Lambda. Express was just handing the routing, which I didn't need anymore with Lambda, API Gateway handles that now. My front-end also calls most Lambdas directly using Cognito for auth. The only time I have problems with Lambdas growing too large is when I try to include a really big payload that includes something like Chrome for headless screenshots, or FFMPEG, or other large binary programs. But for normal Lambda functions I don't have any size issues. Recently when trying to get headless Chrome working in Lambda, I had to boot up EC2 running AWS Linux, and create a .zip file there with library files that were missing in Lambda, then .zip up everything and upload to S3 and then import the .zip into Lambda. After this, I can modify my .js files locally and have my toolchain upload the changes to Lambda to update the Lambda. But this time the payload was too large for importing into a Lambda, and too big for a Lambda layer. The solution was saving the .zip file with Chrome in it to an S3 bucket, and when the Lambda machine starts up, it first downloads the .zip file to /tmp, unzips it, and now the Lambda has the Chrome binaries in the /tmp folder. This bypasses the size limits associated with Lambda. It's rare that I need to do this, and it's not fun, but there are always going to be some edge cases like this with Lambda and getting large architecture specific binaries running in Lambda. I actually put many different functions into a single Lambda, and I call the Lambda with a specific action parameter to perform a specific function. Local development... when I create a Lambda function I may add some functionality so I can also run it on the command line if I'm still early in developing a complex function. But most of the time I just use my self-built toolchain which watches my local code for changes, then creates/updates the Lambda in AWS, which takes about 2 to 5 seconds typically. Then I test the code as it runs inside AWS. If I didn't have internet access, then it's impossible to develop in this way, but that's almost never the case.
- joshstrange 1y ago
  Thank you so much for the reply and all the details! It's great to hear how other people are tackling these issues.

leptons 1y ago

I abandoned Express when I moved to Lambda. Express was just handing the routing, which I didn't need anymore with Lambda, API Gateway handles that now. My front-end also calls most Lambdas directly using Cognito for auth. The only time I have problems with Lambdas growing too large is when I try to include a really big payload that includes something like Chrome for headless screenshots, or FFMPEG, or other large binary programs. But for normal Lambda functions I don't have any size issues. Recently when trying to get headless Chrome working in Lambda, I had to boot up EC2 running AWS Linux, and create a .zip file there with library files that were missing in Lambda, then .zip up everything and upload to S3 and then import the .zip into Lambda. After this, I can modify my .js files locally and have my toolchain upload the changes to Lambda to update the Lambda. But this time the payload was too large for importing into a Lambda, and too big for a Lambda layer. The solution was saving the .zip file with Chrome in it to an S3 bucket, and when the Lambda machine starts up, it first downloads the .zip file to /tmp, unzips it, and now the Lambda has the Chrome binaries in the /tmp folder. This bypasses the size limits associated with Lambda. It's rare that I need to do this, and it's not fun, but there are always going to be some edge cases like this with Lambda and getting large architecture specific binaries running in Lambda. I actually put many different functions into a single Lambda, and I call the Lambda with a specific action parameter to perform a specific function. Local development... when I create a Lambda function I may add some functionality so I can also run it on the command line if I'm still early in developing a complex function. But most of the time I just use my self-built toolchain which watches my local code for changes, then creates/updates the Lambda in AWS, which takes about 2 to 5 seconds typically. Then I test the code as it runs inside AWS. If I didn't have internet access, then it's impossible to develop in this way, but that's almost never the case.

joshstrange 1y ago
Thank you so much for the reply and all the details! It's great to hear how other people are tackling these issues.