Repository navigation
feat(execute): show worker failures across Core and fix admission - #66
Merged
Merged
Conversation
Exact worker selection compared two otherwise identical installation inventories but obtained different DRYML versions from an unrestricted metadata search of source-tree paths. Use the selected installation roots for both record fields and preserve a detectable digest change when the installed version changes. Verified the source-path shadowing regression, exact worker selector integration, and the representative good-enough gate as part of the integrated change.
TensorFlow treats a non-list set_visible_devices argument as one device; the adapter supplied its internal tuple and rejected an otherwise available selected GPU before Keras import completed. Pass a list without changing allocation or visibility checks. Verified fake TensorFlow API behavior for zero, one, and two selected GPUs and the representative runtime and good-enough suites. Real GPU qualification remains tracked separately.
Transport bounded, default-on raw tracebacks through the generic worker v5 channel and the version-2 Core outcome while retaining typed failure reasons, exact publication evidence, and cleanup precedence. The selected backend supports raw_traceback=False for secret-bearing work; the owner explicitly accepted this narrow unredacted diagnostic risk. Cover setup/invocation failures, malformed and oversized diagnostics, opt-out, Ray unit transport, and real subprocess/Core regressions in the representative policy. The integrated local good-enough suite passed 3,442 tests with 30 skips; real Ray/GPU qualification remains opt-in.
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
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
RemoteExecutionError.remote_tracebackandCoreExecutionError.remote_traceback. Keepraw_traceback=Falsefor type/category-only errors, preserve completed publication evidence and cleanup precedence, and bump the private worker protocol to v5 and Core outcome to v2.tf.config.set_visible_devices, fixing the reproduced tuple-as-one-device error during Keras import.Review and verification
raw_traceback=Falsefor secret-bearing work. The acceptance and scope are recorded in issue Report bounded raw Execute and Core worker tracebacks by default #62 and the parent workspace threat-model instructions.good-enough: 3,442 passed, 30 skipped in 19m58s on the exact source tree; no exhaustive suite was requested or run.Closes #62. Separate follow-ups: #63 (generator-cardinality training flexibility), #64 (incomplete cleanup after failed work), and #65 (real TensorFlow GPU visibility qualification). The full traceback in the notebook identified the independent, documented
BasicTrainingsafe-point/validation and finite-cardinality constraints; this PR does not relax either.