-
Notifications
You must be signed in to change notification settings - Fork 166
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
Job exit when all nodecheck failed #1323
base: master
Are you sure you want to change the base?
Job exit when all nodecheck failed #1323
Conversation
and if yes, stop the job. it happens in workers of AllReduce_Strategy jobs
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #1323 +/- ##
==========================================
+ Coverage 80.46% 80.48% +0.02%
==========================================
Files 222 222
Lines 20740 20760 +20
==========================================
+ Hits 16688 16709 +21
+ Misses 4052 4051 -1 ☔ View full report in Codecov by Sentry. |
0, | ||
msg, | ||
level=TrainingExceptionLevel.RDZV_ERROR, | ||
) |
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.
process_error(message=msg, level=...)
@@ -118,6 +118,7 @@ class JobExitReason(object): | |||
RDZV_TIMEOUT_ERROR = "RdzvTimeout" | |||
PENDING_TIMEOUT = "PendingTimeout" | |||
UNCOMPLETED_TIMEOUT = "UncompletedTimeout" | |||
RDZV_ALL_FAILED = "RdzvAllFailed" |
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.
Change name to 'NodeCheckFailed'. This is actually a training failure, not a rdzv failure.
@@ -243,7 +243,42 @@ def is_all_reduce_type_job(self): | |||
== DistributionStrategy.ALLREDUCE | |||
) | |||
|
|||
def is_all_workers_node_check_failed(self): |
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.
Better move this function into WorkerManager
. And work for 'AllReduceType' training only.
return all( | ||
[ | ||
node.is_node_check_failed() | ||
for _, node in self._job_nodes[NodeType.WORKER].items() |
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.
for node in self._job_nodes[NodeType.WORKER].values()
What changes were proposed in this pull request?
exit job if all workers can't pass node check
Why are the changes needed?
no need to re-check or try to fault tolerant this error
Does this PR introduce any user-facing change?
No
How was this patch tested?
UT