feat(BA-6600): return per-requirement selection failures with suggestions#12448
Merged
Conversation
…ions Make the batch agent selector evaluate every requirement and report each as a success or a structured failure, instead of aborting on the first unplaceable kernel. - select_agents_for_batch_requirements() now returns BatchAgentSelectionResult (selections + per-requirement failures) and no longer raises on partial failure; real scheduling re-raises an aggregated BatchAgentSelectionFailedError. - Each AgentSelectionError exposes build_suggestion() -> Suggestion (reduce resource/container, change arch, revise designated agents) so a dry-run can surface remediation without parsing messages. - NoCompatibleAgentError carries structured architecture context; InsufficientResourcesError.deficit() computes per-slot shortage. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
github-actions
Bot
added
size:XL
…ypes Move the remediation value types (Suggestion, SuggestionKind) into a dedicated selectors/types.py that depends only on common value types, so exceptions.py can return a Suggestion without importing selection logic. The selection result types (RequirementSelectionFailure, BatchAgentSelectionResult) stay in selector.py since they depend on AgentSelection/ResourceRequirements defined there. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
fregataa
force-pushed
the
feat/BA-6600-dry-run-selector-types
branch
from
67cd539 to
27f7566
Compare
Comment on lines
+76
to
+79
| def build_suggestion(self) -> Suggestion: | ||
| # The resource group has no candidate agents at all; nothing the caller | ||
| # can adjust on the request itself fixes this. | ||
| return Suggestion(kind=SuggestionKind.NONE) |
Collaborator
There was a problem hiding this comment.
Please add @override.
Comment on lines
+141
to
+143
| if any(isinstance(err, ContainerLimitExceededError) for err in self._agent_errors.values()): | ||
| # No node is short on slots; the blocker is the per-agent container limit. | ||
| return Suggestion(kind=SuggestionKind.REDUCE_CONTAINER) |
Collaborator
There was a problem hiding this comment.
Instead of handling specific errors, should we use a specific base exception?
Comment on lines
+119
to
+124
| @dataclass | ||
| class RequirementSelectionFailure: | ||
| """A requirement that could not be placed, paired with its structured reason.""" | ||
|
|
||
| resource_requirement: ResourceRequirements | ||
| error: AgentSelectionError |
Collaborator
There was a problem hiding this comment.
Would it be better to include the exception itself in the dataclass?
Comment on lines
+216
to
+217
| _required_architecture: str | ||
| _available_architectures: Sequence[str] |
Collaborator
There was a problem hiding this comment.
It appears that the _required_architecture value is saved but not used later on.
| error=e, | ||
| ) | ||
| ) | ||
| continue |
Collaborator
There was a problem hiding this comment.
If you're only dealing with resource_req anyway, it seems like you should be able to tell which resource is in short supply without having to iterate through the data. What behavior were you intending?
Member
Author
There was a problem hiding this comment.
Each candidate node has different remaining slots, so the shortage (requested − remaining) differs per node — it can't be derived from resource_req alone. And when a kernel fails mid-batch, the per-node remaining shifts as earlier kernels' placements are already accumulated, so the excess for a later kernel can differ from an earlier one. We therefore iterate the candidate nodes to pick the one needing the smallest reduction.
…or field Address review feedback: - Add @OverRide to overriding error_code/failed_predicates/build_suggestion methods in the selection exception hierarchy. - Remove NoCompatibleAgentError._required_architecture, which was stored but never read (the message uses the constructor argument directly). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ResourceRequirements is a pure value type (depends only on common types), so relocate it from selector.py to types.py and update all importers. This lets lower-layer modules (e.g. exceptions) reference it without importing selection logic. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Rename Suggestion -> RemediationHint and drop the kind discriminator: any subset of fields may be populated, plus required_container_reduction. - TrackerCompatibilityError.remediation_hint_contribution() lets each node-level error contribute its partial hint; NoAvailableAgentError merges them (no isinstance branching). - Add RequirementSelectionError base carrying the failed resource_requirement; NoAvailableAgentError / NoCompatibleAgentError extend it. - select_agents_for_batch_requirements() returns list[AgentSelection] and raises BatchAgentSelectionFailedError (public .errors) when any requirement cannot be placed, so a dry-run can catch it for per-kernel feedback. - Remove unused ArchitectureIncompatibleError. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
jopemachine
approved these changes
Jul 1, 2026
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
3 participants
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
select_agents_for_batch_requirements()evaluate every requirement and returnBatchAgentSelectionResult(successes + per-requirement failures) instead of aborting on the first unplaceable kernel. Real scheduling re-raises an aggregatedBatchAgentSelectionFailedErrorto keep its all-or-nothing semantics.build_suggestion() -> Suggestionto everyAgentSelectionErrorso a dry-run can surface structured remediation (reduce resource/container, change arch, revise designated agents) without parsing messages.NoCompatibleAgentErrorstructured architecture context and addInsufficientResourcesError.deficit()(per-slot shortage);NoAvailableAgentErrorcomputes the node-axis MIN reduction.Test plan
pants test tests/unit/manager/sokovan/scheduler::test_build_suggestion.pycovers deficit, arch/reduce/container/designated/none suggestions, node-axis MINResolves BA-6600