Metric Logging updates 4/N #351
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
added 11 commits
October 8, 2025 08:38
…estamp_logging_diff2
meta-cla
bot
added
the
CLA Signed
…estamp_logging_diff3
|
could you add some more comments in the description about what this PR is enabling? |
yes! sorry. I should have marked as a draft. I am doing a 2.5/4.0 before i ask you to review it |
felipemello1
changed the title
felipemello1
changed the title
felipemello1
commented
Oct 9, 2025
felipemello1
commented
Oct 9, 2025
| self.timestamp = datetime.now(pytz.UTC).timestamp() | ||
|
|
||
|
|
||
| def get_actor_name_with_rank() -> str: |
There was a problem hiding this comment.
moved to observability/utils.py
felipemello1
commented
Oct 9, 2025
Comment on lines
+81
to
+82
| if process_name is None: | ||
| process_name = detect_actor_name_from_call_stack() |
There was a problem hiding this comment.
get name here and pass it to
local_fetcher_actor = proc.spawn(
"local_fetcher_actor", LocalFetcherActor, global_logger, process_name
)
this function is called in provisioner.py, and thats how we get the process_name for every wandb run
felipemello1
commented
Oct 9, 2025
| logger = logging.getLogger(__name__) | ||
|
|
||
|
|
||
| def detect_actor_name_from_call_stack() -> str: |
There was a problem hiding this comment.
main file to review
|
in the near future the mesh might hold a name. When this happens, we can delete the function to use the call stack and just get it from the mesh. The rest of the PR stands. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When logging per rank, we need good naming to make it easier to debug. I was previously trying to fetch this name from
monarch.actor.context, but since whenMetricCollector.init_backendsis called, it is in the context of theLocalFetcherActor, i would just get the namelocal_fetcher_actor.To solve this, i use the call stack instead and call it when the actor is spawning ->
provisioner.py->get_or_create_metric_logger->detect_actor_name_from_call_stack(). The utility then goes back until it finds aForgeActorsubclass, and get its name. This gets saved as theprocess_name, used by the metric logging backends.User can also pass a process name as input. Thats how we get the
Controllername.The process_name then goes:
LocalFetcherActor->MetricCollector->backend.init-->wandb.init(name)It does feel a bit brittle to traverse until finding a ForgeActor. But worst case scenario:
a) user can pass the process name
b) if we cannot find any, then its just UnknownActor