Just before it is wheels up and hopping over the Atlantic Ocean to the US for AWS re:invent. I wanted to do a quick blog containing my wish list. So here it is:
More features in SageMaker Serverless
SageMaker Serverless went GA in April 2022, having been announced at last years re:invent. However, it has the following limitations:
GPUs, AWS marketplace model packages, private Docker registries, Multi-Model Endpoints, VPC configuration, network isolation, data capture, multiple production variants, Model Monitor, and inference pipelines.
These for me are pretty much showstoppers for any situation I wish to use SageMaker Serverless, so I revert back to Lambda and do things the long way. OK off course I know GPU is still an issue with Lambda but please AWS can have these features for SageMaker Serverless.
Managed MLFlow
I am seeing growing demand for MLflow (https://mlflow.org/) and I am seeing a lot of people looking at Databricks as commercial offering for MLflow. Alternatively, some popele are implementing something like Managing your Machine Learning lifecycle with MLflow. Therefore, I think this was on my wish list last year, but I really hope AWS announce a Managed MLFlow Service. I know version 2.X is too new but at least 1.X would be great start.
Public CodeCommit
I used CodeCommit every day and it offers basic GIT functionality for private repositories. However, I wish to share anything with the public then I need to use my GIT Hub account as a second remote and remember to push too both. It would be awesome for AWS to offer public repositories that I can push my open-source projects too.
3rd party integration in StepFunctions
This year I have got into Apache Airflow and I wrote a blog Running Thousands of Models a month with Apache AirFlow. For me the main reason you would choose Apache Airflow over StepFunctions is the type of integrations you require. If you’re mainly using a lots of Open Source or Commercial offerings, either SaaS or PaaS, then your best using Apache Airflow. However, if you’re going mainly AWS Cloud Native then your use StepFunctions. Maybe with the odd Lambda for any non-AWS integrations. It would be awesome for AWS offer away, like for the boto3 SDK, to offer a way to use 3rd Party SDKs
StepFunctions Local Dev
I know I am not the only that fines this tedious but testing step functions locally with local integrations is hard work. It also not easily to integration test a complete serverless app. If you really do have time, you can build something with testcontainers and docker compose to spin up various AWS lock offerings:
- https://hub.docker.com/r/amazon/aws-stepfunctions-local
- https://docs.aws.amazon.com/serverless-application-model/latest/developerguide/sam-cli-command-reference-sam-local-start-lambda.html
- https://github.com/awslabs/amazon-ecs-local-container-endpoints
- https://hub.docker.com/r/amazon/dynamodb-local
Please AWS I would love a complete local serverless development environment for the main serverless services.
Cross account param store replication
Lastly my final wish is replication of parameters in Parameter store between accounts. I
have recently built some MLOPS pipelines where the S3 buckets and Dynamo DB tables
exist in one account and they are used by other accounts. I want the name of the buckets and tables, that are created dynamically by CloudFormation to be shared via Parameter store with other accounts. This would mean other Cloud Formation in the other accounts could just reference the Parameter
I hope you like the list and you want to meet at re:invent, contact me via PeerTalk
Top comments (0)