There are a few options for creating conda environments in OpenSARLab.

Each option come with benefits and drawbacks.

Create Conda Environments, Register Their Kernels, and Run Any Setup Scripts in the Docker Image/s

Users don't have to create conda environments, which saves them time
Users don't need to know much about conda; they can just start running notebooks

Users cannot install additional packages into their conda environments
Changes to environments on the docker image involve rebuilding the container and CodePipelines
Large environments may overrun the 20GB root volume mounted on each EC2 instance, requiring that larger, more expensive root volumes be used.

Users don't have to create conda environments, which saves them time
Users don't need to know much about conda at all; they can just start running notebooks
The environments are stored in $HOME/.local, so users have permissions to install, update, remove, and debug packages
- Environments are synced, not copied, so changes made by users will persist across server restarts

Increases the time it takes to start an OpenSARLab server
- Syncing environments from the docker image to $HOME/.local, registering their kernels, and running any needed setup scripts all happens at server startup
Large environments may overrun the 20GB root volume mounted on each EC2 instance, requiring that larger, more expensive root volumes be used.
By storing the environment on both user volumes and EC2 node volumes, you effectively double pay for that storage.

Docker images remain small, avoiding potential storage overruns on the EC2 nodes' 20GB volumes.
Server start ups do not require copying or syncing environments, and so require less time.
Users have full control over their conda environments and their changes will persist across server restarts.

Users have to create their own conda environments
- This requires some knowledge of conda and takes time.
- Note: There is an ASF notebook repo to aid users in building their own environments.