retry quorum #228
Open
retry quorum #228
+126
−27
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
This was referenced Jul 10, 2025
facebook-github-bot
added
the
CLA Signed
tushar00jain
force-pushed
the
pr228
branch
5 times, most recently
from
2210b6c to
5c14c9e
Compare
d4l3k
approved these changes
Jul 12, 2025
src/manager.rs
Outdated
| timeout: Duration, | ||
| lighthouse_request: LighthouseQuorumRequest, | ||
| ) -> Result<tonic::Response<LighthouseQuorumResponse>, Status> { | ||
| let mut client = self.lighthouse_client.clone(); |
There was a problem hiding this comment.
Will this reconnect if an error occured or do we need to construct a new lighthouse_client in that case?
There was a problem hiding this comment.
updated, now we create a new client each time the request fails
torchft/manager.py
Outdated
| @@ -271,6 +276,7 @@ def __init__( | |||
| world_size=group_world_size, | |||
| heartbeat_interval=heartbeat_interval, | |||
| connect_timeout=connect_timeout, | |||
| quorum_retries=int(os.environ.get(QUORUM_RETRIES_ENV, "0")), | |||
There was a problem hiding this comment.
may be good to also add a config variable for this
There was a problem hiding this comment.
the env var is the config? or you mean somewhere else?
There was a problem hiding this comment.
i.e. able to specify it via an arg to Manager init
tushar00jain
force-pushed
the
pr228
branch
3 times, most recently
from
9ae71da to
cc1b895
Compare
tushar00jain
force-pushed
the
pr228
branch
7 times, most recently
from
ecd4720 to
0931e7e
Compare
Summary: - we currently don't retry quorum requests from the manager to lighthouse - if lighthouse crashes, this can result in all replicas crashing - so add retries configurable through env var - remove holding state lock when making network calls in manager - the manager tries reconnecting to lighthouse if a response from lighthouse fails up to configured number of retries - there's still some unhandled cases - manager doesn't broadcast the result to all ranks if there's a failure in `_run_quorum`, resulting in a hang - if a rank gets error from quorum, it still crashes (the handling will be more complicated if ranks are on multiple hosts and they can independently reconnect)
d4l3k
approved these changes
Jul 15, 2025
| "Failed to send heartbeat to lighthouse: {}", | ||
| e.to_string() | ||
| ); | ||
| let _ = self.create_lighthouse_client().await; |
There was a problem hiding this comment.
Should we be checking status for this?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Summary:
_run_quorum, resulting in a hang