manager: add per request timeouts #59
Merged
Conversation
This file contains 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
facebook-github-bot
added
the
CLA Signed
d4l3k
force-pushed
the
d4l3k/per_req_timeout
branch
from
2dcd290 to
731fe8c
Compare
Jackmin801
approved these changes
Jan 8, 2025
d4l3k
force-pushed
the
d4l3k/per_req_timeout
branch
3 times, most recently
from
54982e2 to
9a0cc4b
Compare
|
Updated with some unit tests and an integration test + added timeout field to |
Jackmin801
reviewed
Jan 8, 2025
torchft/manager_test.py
Outdated
Comment on lines
491
to
511
| @patch("torchft.manager.ManagerClient", autospec=True) | ||
| def test_quorum_happy_timeouts(self, client_mock: MagicMock) -> None: | ||
| manager = self._create_manager(use_async_quorum=False) | ||
| client_mock().should_commit = mock_should_commit | ||
|
|
||
| client_mock().quorum.return_value = ( | ||
| 123, # quorum_id | ||
| 1, # replica_rank | ||
| 2, # replica_world | ||
| "manager address", | ||
| f"localhost:{self.store.port}", | ||
| 1, # max_step | ||
| 1, # max_rank | ||
| 2, # max_world_size | ||
| False, # heal | ||
| ) | ||
|
|
||
| manager.start_quorum(timeout=timedelta(seconds=12)) | ||
| self.assertTrue(manager.should_commit(timeout=timedelta(seconds=23))) | ||
|
|
||
| self.assertEqual(client_mock().quorum.call_count, 1) |
There was a problem hiding this comment.
Im not sure if this actually tests that the timeout is passed correctly. Seems to just test that the _clients methods are called at all. Maybe check that _client was called with the expected timeout?
There was a problem hiding this comment.
added a check on the kwargs
torchft/manager_integ_test.py
Outdated
Comment on lines
427
to
468
| def test_quorum_timeout(self) -> None: | ||
| with ExitStack() as stack: | ||
| lighthouse = Lighthouse( | ||
| bind="[::]:0", | ||
| min_replicas=2, | ||
| ) | ||
| stack.callback(lighthouse.shutdown) | ||
|
|
||
| store = dist.TCPStore( | ||
| host_name="localhost", | ||
| port=0, | ||
| is_master=True, | ||
| wait_for_workers=False, | ||
| ) | ||
|
|
||
| pg = ProcessGroupGloo() | ||
| manager = Manager( | ||
| pg=pg, | ||
| min_replica_size=2, | ||
| load_state_dict=lambda x: None, | ||
| state_dict=lambda: None, | ||
| store_addr="localhost", | ||
| store_port=store.port, | ||
| rank=0, | ||
| world_size=2, | ||
| lighthouse_addr=lighthouse.address(), | ||
| port=19530, | ||
| use_async_quorum=False, | ||
| ) | ||
| stack.callback(manager.shutdown) | ||
|
|
||
| with self.assertRaisesRegex( | ||
| TimeoutError, | ||
| "status: Cancelled, message.*Timeout expired", | ||
| ): | ||
| manager.start_quorum(timeout=timedelta(seconds=0.01)) | ||
|
|
||
| with self.assertRaisesRegex( | ||
| TimeoutError, | ||
| "status: Cancelled, message.*Timeout expired", | ||
| ): | ||
| manager.should_commit(timeout=timedelta(seconds=0.01)) |
There was a problem hiding this comment.
This doesnt really test that methods now have method specific timeouts though. They could still timeout on the default timeout and pass. Maybe set a different timeout in Manager init from the one used by method and assert the method one was used?
There was a problem hiding this comment.
added an elapsed time check -- the default is 60s so easy to verify that we're shorter
d4l3k
force-pushed
the
d4l3k/per_req_timeout
branch
from
9a0cc4b to
03c22d9
Compare
fegin
approved these changes
Jan 9, 2025
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.
This adds per request timeouts to all the ManagerClient methods. It also forwards the timeout from the manager to the lighthouse for quorum operations.
Test plan:
updated unit tests