Running usd_from_gltf in an AWS Lambda - docker

Question

I'm trying to run Google's usd_from_gltf utility inside AWS Lambda, using a custom Docker image. The setup seems to be working locally but when executing the same Lambda in AWS, it fails for certain input files.
Minimal app
https://github.com/petrbroz/glb-to-usdz-test
This is a minimalistic AWS SAM app with a Lambda function called GlbToUsdzFunction that downloads a Glb file from specified URL and converts it to Usdz. The Lambda function uses a custom Docker image (https://github.com/leon/docker-gltf-to-udsz), and Python's subprocess to run the usd_from_gltf tool to handle the conversion.
Sample file URLs
https://petrbroz.s3.us-west-1.amazonaws.com/glb-to-usdz-issues/snowmobile.glb
https://petrbroz.s3.us-west-1.amazonaws.com/glb-to-usdz-issues/wall-e.glb
When running locally
The Lambda function succeeds for both snowmobile.glb and wall-e.glb. Here's an example output for the former:
$ sam build
$ echo "{ \"url\": \"https://petrbroz.s3.us-west-1.amazonaws.com/glb-to-usdz-issues/snowmobile.glb\" }" | sam local invoke "GlbToUsdzFunction" --event -
Reading invoke payload from stdin (you can also pass it from file with --event)
Invoking Container created from glbtousdzfunction:glb-to-usdz-lambda
Building image.................
Skip pulling image and use local one: glbtousdzfunction:rapid-1.46.0-x86_64.
START RequestId: 720b6b49-e36c-4429-96fb-9e0e5c02c09b Version: $LATEST
Downloading file
Converting file
Warning: extensionsUsed: Extension is in extensionsUsed but not actually referenced: KHR_texture_transform [GLTF_WARN_EXTENSION_UNREFERENCED]
END RequestId: 720b6b49-e36c-4429-96fb-9e0e5c02c09b
REPORT RequestId: 720b6b49-e36c-4429-96fb-9e0e5c02c09b Init Duration: 0.22 ms Duration: 19997.59 ms Billed Duration: 19998 ms Memory Size: 1024 MB Max Memory Used: 1024 MB
{"status": "success"}
When running in AWS
The Lambda function succeeds for snowmobile.glb but fails for wall-e.glb. Here's the output for the latter:
START RequestId: b1bdc496-ec12-430e-a641-2574af354d60 Version: $LATEST
Downloading file
Converting file
ERROR: USD: Insufficient permissions to write to destination directory '/var/tmp' (Replace) [UFG_ERROR_USD]
ERROR: USD: Failed to map '/var/tmp/output.usdc': No such file or directory (AddFile) [UFG_ERROR_USD]
Warning: USD: Failed to add temporary layer at '/var/tmp/output.usdc' to the package at path 'output.usdz'. (_CreateNewUsdzPackage) [UFG_WARN_USD]
ERROR: Cannot write USD: "/tmp/output.usdz" [UFG_ERROR_IO_WRITE_USD]
Command '['usd_from_gltf', '/tmp/input.glb', '/tmp/output.usdz']' returned non-zero exit status 255.
END RequestId: b1bdc496-ec12-430e-a641-2574af354d60
REPORT RequestId: b1bdc496-ec12-430e-a641-2574af354d60 Duration: 2039.96 ms Billed Duration: 5166 ms Memory Size: 1024 MB Max Memory Used: 101 MB Init Duration: 3125.71 ms
Has anyone run into this? Am I doing something wrong here, or is this perhaps a bug on the AWS side, or on the usd_from_gltf side?

Answer 1

Some USD conversions cause the library to write intermediate files and it looks like it is built to use /var/tmp for this purpose. Since Lambdas can only write to /tmp, the workaround we came up with is to link /var/tmp to /tmp:
in the glb-to-usdz Dockerfile, add a line like RUN rm -rf /var/tmp && ln -s /tmp /var/tmp
This allows your second example to succeed.