[BUG] ConnectionReset on AWS lambda - registry

[ruivitit]

Describe the bug
When running graphistry on a lambda (in this case using a Docker container) I have a ConnectionResetError on the registry step.
If I run the code locally, the issue does not occur.

To Reproduce
Code, including data, than can be run without editing:

import graphistry
graphistry.register(api=3,  personal_key_id="123123", personal_key_secret="123123")

Expected behavior
I should be able to successfully register.

Actual behavior
I get the ConnectionResetError (see screenshot attached).

Screenshots

Browser environment (please complete the following information):

• OS: Docker Lambda Environment
• Browser N/A
• Version: N/A

Graphistry GPU server environment

• Where run: AWS Docker Lambda
• If self-hosting, Graphistry Version: N/A

PyGraphistry API client environment

• Version: 0.35.4
• Python Version: 3.10

Additional context
There is connection to hub.graphistry.com since I've performed two validations:

1. On another lambda (same network configuration), I can curl the endpoint with no problems.
2. All network traffic is accepted (confirmed by looking at the VPC Flow Logs)

[lmeyerov]

Interesting!

• Did you test a curl from the same process, docker container, or host? It may be the docker container / network is locked down somehow...

• Does the same docker image succeed when run locally?

• What happens if you add a retry?

• A less likely item may be hard coding which Graphistry server to use: https://pygraphistry.readthedocs.io/en/0.34.6/server/register.html#protocol-configuration

graphistry.register(api=3, protocol='https', server='hub.graphistry.com', ...

• What happens if you use curl (bash) and then httpx (another python net client) to directly request a token? https://hub.graphistry.com/docs/api/1/rest/auth/#tokencreate

[ruivitit]

• Did you test a curl from the same process, docker container, or host? It may be the docker container / network is locked down somehow...

Yes, I've performed a curl from my machine, from a regular lambda, from a docker container.
I always get a response from hub.graphistry.com.

• Does the same docker image succeed when run locally?

Yes.

• What happens if you add a retry?

Actually didn't try that, but not sure why a retry would be needed.

• A less likely item may be hard coding which Graphistry server to use: https://pygraphistry.readthedocs.io/en/0.34.6/server/register.html#protocol-configuration
  graphistry.register(api=3, protocol='https', server='hub.graphistry.com', ...

Previous iteration of this code had the server and/or protocol specified, always the same result. I've pretty much tried every combination of parameters of the register function.

• What happens if you use curl (bash) and then httpx (another python net client) to directly request a token? https://hub.graphistry.com/docs/api/1/rest/auth/#tokencreate

I get a connection successfull.

[lmeyerov]

So the python httpx version works from the serverless, while request-based (pygraphistry default) doesn't?

It's hard to imagine what is different there, that sounds like a client env thing. For example, I think request obeys SOCKS proxy overrides, while httpx might not. It does point to a client issue vs server issue, but I'm still confused if a pygraphistry bug or your python env / server setup, like the SOCKS issue

• I wonder if some sort of telemetry can help here, like a way to capture the outbound request and compare it to a regular curl from the same box. So we can compare working vs non-working

• Another idea can be to have the docker container python process call curl via a popen/exec shell out

• We don't have an easy reproducible on our end, so been tricky to help. Maybe there is a way you can make a minimal repro, including lambda automation? Ideally using stock base images to avoid security issues etc on our end