Hello,
We have an issue with ArcGIS Notebook (10.9.1), the solution was to remove the tensorflow-gpu library from the container, after that it can be used normally.
Some ideas.
- It is convenient that the documentation specifies the requirement of having a CPU, hypervisor and OS that use AVX (I didn't see oficial information published by ESRI).
o In Notebook you can use !more /proc/cpuinfo | grep flags to check; if avx is supported on the response parameters you have a string (...aes xsave avx f16c rdrand ).
- Give the option to have a "normal" container (without tensorflow-gpu) and another one with the functionality.
- Give the alternative of using a command at the moment of loading the container through the Notebook Server to not load that functionality (e.g. --tensorflow-gpu no, --add tensorflow-gpu etc).
- At installation verify that the CPU has support for AVX, with a message, such as hard drive space, RAM memory etc.
- When loading the container image, indicate that AVX is not supported and disable that option.
- Provide additional information on how to find out if the CPU has AVX support, and how to enable it in case it does and it is not identified by the VM.
Enviroment.
QAS, Red Hat Enterprise Linux Server 8 Update 4, ArcGIS Notebook Server 10.9.1, VM created on Hyper-V, 4 cores and 16 GB RAM, it's a federated environment using AGE 10.9.1; it's using a CA AD certificate. The AGE servers are using Windows Server 2019 Standard.
Issue.
Notebooks looks fine, docker images are configured, the account was created, web adaptor configured etc.
It's possible use the basic stuff like addition, subtraction, division, printing messages, numpy etc.
But if I try to use the arcgis (GIS, ArcPy etc) elements the kernel crashes.
Error message.
The kernel appears to have died. It will restart automatically.
The issue is related to.
- CPU (AVX) support.
- Tensorflow (tensorflow-gpu) (deep learning).
Regards.