Hey, I’m deploying my flows on different AWS accounts. When I want to inspect my flows, running:

from metaflow import Metaflow
print(Metaflow().flows())

I want to control from which account I’m pulling the results. Which environmental variable should I set, to ensure showing flows from desired AWS account?

@steep-traffic-55869 You would need to have a Metaflow Configuration file per account in this case. For example, you could maintain two

config.json

files under

~/.metaflowconfig

. One for the default account and another for the other account, such as

config_prod.json

. You can switch to the other one with the postfix

prod

using the environment variable

METAFLOW_USER

where you set it to

METAFLOW_USER=prod python script.py

. Another thing to make the AssumeRole between the accounts easier would be to use aws-vault with the aforementioned method. How to run in the default account:

aws-vault exec default_account -- python script.py

How to run it in the prod account:

aws-vault exec prod_account -- METAFLOW_USER=prod python script.py

if deploying to multiple accounts, do you also have multiple Metaflow configs on the machine, one for each target account?

METAFLOW_PROFILE

controls which config is picked for the library, then for authenticating against AWS resources you'd also have to set

AWS_PROFILE

to a profile that has suitable credentials (or alternatively the aws-vault approach from Yudhiesh). One thing to note is that the config seems to only be read once during module loading, so doing something like

os.environ["METAFLOW_PROFILE"]="otherprofile"

does not change anything with already imported libraries based on a quick test.