Draft
Conversation
…e reliability - Replace individual UPDATE queries with bulk UPDATE for job state changes - Add retry logic with configurable max attempts for database operations - Implement consecutive empty fetch tracking to prevent infinite loops - Add proper error handling and recovery for database session failures - Track enqueued jobs to prevent duplicate processing - Add configurable TUNA_MAX_EMPTY_FETCHES environment variable - Improve logging for better observability of enqueue process This optimization significantly reduces database round-trips when updating multiple job states and makes the enqueue process more resilient to transient failures.
…ocker build stage
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Motivation
Draft PR. Some changes to make life easier for gathering data for training our 3D heuristics.
Does not need to be merged at the moment.
Some small changes to the docker-file. This should probably be moved somewhere else as to not dirty this one.
The main changes are in the way the distributor and workers interact when collecting and running jobs.
We run the distributor and worker on isolated nodes where the only communication happens implicitly through the database.
For that reason the distributors should be a bit less greedy in grabbing all jobs from the database, but only a single batch.
Only when the batch is nearing completion, it will query the database again to fetch a new batch of jobs.
This makes it easier to run multiple nodes in parallel without requiring a single central distributor and direct communication between the nodes.
Technical Details
miopen_lib.y:
extract_job_id_from_context()- A new method to extract job IDs from MIOpen celery task contextsmituna_interface.py
Job Tracking: Added instance variables for tracking claimed and completed jobs:
self.claimed_job_ids = set()self.completed_job_ids = set()self.progress_factor = 0.25(for determining when to grab more jobs, default check after 75% is finished)Polling Optimization:
TUNA_POLL_INTERVAL(default 5 seconds)Job Completion Tracking:
parse_result()to track completed jobsextract_job_id_from_context()that subclasses must implementTest Plan
Test Result
Submission Checklist