Hi, I was excited to discover metaflow today since...
# ask-metaflow
b
Hi, I was excited to discover metaflow today since I had been thinking it'd be great to have a 'serverless' workflow orchestration that deployed to AWS Step Functions... and to find that this exists and is quite mature! Some questions re deploying to my own AWS infra... the "admin docs" point to Outerbounds, and they point to https://github.com/outerbounds/metaflow-tools/tree/master/aws/cloudformation this sets up an RDS Postgres db is this required, and is it possible to use MySQL instead? AFAICT the website docs don't mention needing an SQL db at all the CF template says its for the Metadata Service but https://outerbounds.com/engineering/service-architecture/#metadata says it uses the Datastore and https://outerbounds.com/engineering/service-architecture/#datastore says that uses S3 ?
I guess it's this? https://github.com/Netflix/metaflow-service/blob/master/services/data/db_utils.py and it's hard-coded to Postgres by the looks of it
https://outerbounds.com/engineering/operations/introduction/ says: > Metaflow comes with an optional centralized metadata service which links to: https://outerbounds.com/engineering/service-architecture/ but that lists "Metadata" under "Required Services"
is it possible to do without the Metadata service and RDS Postgres db, and if so what do I lose by not having it? Just a nice web UI? any other features?
a
yes, you can use metaflow without the postgres db. in that case, we will dump all the metadata (index over information stored in the datastore) in your local filesystem. you will still be able to run workloads on step functions and batch but you won't have any UI and others may not be able to access your state through the client (but you would be able to)
👍 1
b
@square-wire-39606 thanks... in this case which local filesystem are we referring to? on the worker node? the scheduler? (...and here https://outerbounds.com/engineering/service-architecture/ there are maybe a "local" and a "production" scheduler?)