-
Notifications
You must be signed in to change notification settings - Fork 16
Metric Logging updates 1/4 #345
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
Metric Logging updates 1/4 #345
Conversation
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 Felipe!
# In multiprocessing environments, WandB service tokens can become stale and point | ||
# to dead service processes. This causes wandb.init() to hang indefinitely trying | ||
# to connect to non-existent services. Clearing forces fresh service connection. | ||
from wandb.sdk.lib.service import service_token |
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.
should this also be imported at the top?
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 dont think so. My opinion is that backends like wandb should be protected, otherwise user is required to have wandb even if they dont use it. Extrapolate that to mlflow, scuba, etc. Let me know if you disagree.
FORGE_DISABLE_METRICS
now disables spawningLocalFetcherActor
when spawning processesrecord_metrics(...)
, ifGlobalLoggingActor.init_backends
was not initialized, it would raise errors. Now it just prints a warning. This enables stand alone components to be tested without metric logging.init_backend
was called AFTER initializing the other services. Callingservice_token.clear_service_in_env()
in shared mode fixes this issue.