-
Notifications
You must be signed in to change notification settings - Fork 874
Fixing task ID replacement for MNP jobs on AWS Batch #2574
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: master
Are you sure you want to change the base?
Conversation
self._task_id.replace("control-", "") | ||
+ "-node-$AWS_BATCH_JOB_NODE_INDEX", | ||
) | ||
# Fix: Only replace task ID in specific arguments, not in environment variables |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
thanks for the fix! @saikonen any suggestions on approaches that might allow us to not lean on string substitution (quite finicky)!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Haven't touched this implementation in a bit so maybe there is a reason that the task-id patterns are kept in execute()
but at first glance this seems (in the original implementation) like quite a late phase to be doing anything with the command.
Could a better place be when the cmd is not yet joined into a string, when we can operate on options separately? f.ex. batch_cli.py#251
and step_functions.py#925
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm not 100% sure of the flow but I think that is possible if the flag --task-id
is the only thing that needs to be changed. Looking at the command, I'm not sure if there is something else that requires us to modify the task ID (ex. MF_PATHSPEC
). If task-id
is the only thing, then I can definitely make that change.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've updated the code. I think it is difficult to fully rely on batch_cli.py#251
because we don't have access to $AWS_BATCH_JOB_NODE_INDEX
as this would only be available at runtime for a given worker MNP node, I believe. But I've added more placeholders in batch_cli.py#251
that should make the regex more reliable. I'm less familiar with the step_functions
file as I'm not using that right now.
Is it possible to get approval on this soon?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
…place with node-index
If we aren't using the Metaflow metadata service provider, Metaflow defaults to generating task IDs locally. But these task IDs are just simple integers based on how many tasks/steps there are and are sequentially incremented based on
new_task_id
inmetaflow/plugins/metadata_providers/local.py
. This presents a problem when we're doing AWS Batch MNP, since currently we try and mass replace based on the task ID in the secondary command. If this is a simple integer, this will replace many erroneous places.For example, if the task ID is "3", there could be many instances of "3" in the secondary command that then have many replacements with "-node-$AWS_BATCH_JOB_NODE_INDEX" when really we just want to replace the actual task ID.
Here, I've identified two places - the input task ID via
--task-id
and the task ID inMF_PATHSPEC
, that should be the only two places in the command that have the actual task ID in them that need replacing. It is better to have more specific regexes this way.