Use environment variable per run, not per container #6480
Conversation
Currently we pass the environment vars in the init method and create the container with that. This means that we can't modify those after the container creation. We do allow this in the LocalBuildEnvironment. This is needed to run git in a container (set envs like `GIT_DIR`). And to execute the ssh agent inside, (sent envs like `SSH_AUTH_SOCKET` and `SSH_AGENT_PID`)
|
I'm not sure to understand what are the benefits of this PR and why we need this. Can you explain a little more here?
Can't these variables set at the moment the container is created instead? |
But we initialize the ssh agent (set the env vars) after the container is created. |
|
We need to do: Create container -> Run |
| @@ -452,7 +453,7 @@ def run_command_class( | |||
| kwargs['bin_path'] = env_path | |||
| if 'environment' in kwargs: | |||
| raise BuildEnvironmentError('environment can\'t be passed in via commands.') | |||
| kwargs['environment'] = self.environment | |||
| kwargs['environment'] = self.environment.copy() | |||
There was a problem hiding this comment.
Are we always calling copy so we don't accidentally modify it in memory?
Currently we pass the environment vars in the init method
and create the container with that.
This means that we can't modify those after the container creation.
We do allow this in the LocalBuildEnvironment.
This is needed to run git in a container (set envs like
GIT_DIR).And to execute the ssh agent inside, (sent envs like
SSH_AUTH_SOCKETand
SSH_AGENT_PID)This is compatible with current code, so it can be merged just now.