Netflix / metaflow

:rocket: Build and manage real-life ML, AI, and data science projects with ease!

Home Page:https://metaflow.org

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Invalid CloudWatch log group prefix

stathva opened this issue · comments

"awslogs-group": "aws/batch/job",

The default CloudWatch log group prefix above is invalid. AWS reserves log groups starting with AWS/. Perhaps change it to /aws/batch/job

According to https://docs.aws.amazon.com/batch/latest/userguide/using_awslogs.html#create_awslogs_logdriver_options
the default is aws/batch/job

Moreover, check the usage here: #1767 (comment)
We supply try/batch/custom/demo i.e. without the / in beginning and it worked as shown in #1767 (comment)

@stathva Did you have issues with running flows with what we have here i.e. aws/batch/job without the / in beginning?

@stathva Did you have issues with running flows with what we have here i.e. aws/batch/job without the / in beginning?

@madhur-ob yes, we deploy Metflow Wofklows as StepFunctions, our workflow is using the @batch decorator to define the container image and step resource requirements but it doesn't specify a log group.

The jobs failed with the following error:

"Reason": "CannotStartContainerError: Error response from daemon: failed to initialise logging driver: failed to create Cloudwatch log stream: InvalidParameterException: Log groups starting with AWS/ are reserved for AWS.",

So the problem is not missing / in general like in the try/batch/custom/demo example but using aws/* since its reserved.

Aah, I think the fix is to not include it at all I guess...let me remove it, that way, it will consider aws/batch/job by default..