search

rpcme / aws-cloud-and-xilinx-workshop

MIT License
31 stars 13 forks source link

subprocess won't be killed when we need to re-deploy or update lambda function #28

Open xupsh opened 5 years ago

xupsh commented 5 years ago

https://github.com/rpcme/aws-cloud-and-xilinx-workshop/blob/da51668ec702908d0bf59b6a86d8c45afb70100b/cloud/xilinx-video-inference-handler/lambda_function.py#L59-L62

rpcme commented 5 years ago

Hello @xupsh -

Thank you for identifying this issue! We will identify a resolution and update here. It is a good idea to fix it so it doesn't set bad precedence.

Please also let us know the severity of this issue in the context of this workshop.

yunqu commented 5 years ago

This is a bug since the check_call is non-blocking. Sometimes the call is taking longer time than expected; this makes the next subprocess call error out.

Should be fixed with the new commit 795dc2cf8799ad1fb5b37de18821887785ae5a11

rpcme commented 5 years ago

@yunqu love the change-up. But even if blocking do we know for sure that the child process doesn't orphaned? Not sure if the SIGKILL propagates to the child process. It would be interesting to see if it truly does propagate -- if not, we would likely need to trap the signal and kill the child process from there.