feat(us1): add dataset review orchestration automatic review slice

This commit is contained in:
2026-03-17 10:57:49 +03:00
parent e916cb1f17
commit 023bacde39
24 changed files with 4870 additions and 131 deletions

View File

@@ -0,0 +1,386 @@
# [DEF:DatasetReviewOrchestrator:Module]
# @COMPLEXITY: 5
# @SEMANTICS: dataset_review, orchestration, session_lifecycle, intake, recovery
# @PURPOSE: Coordinate dataset review session startup and lifecycle-safe intake recovery for one authenticated user.
# @LAYER: Domain
# @RELATION: [DEPENDS_ON] ->[DatasetReviewSessionRepository]
# @RELATION: [DEPENDS_ON] ->[SemanticSourceResolver]
# @RELATION: [DEPENDS_ON] ->[ClarificationEngine]
# @RELATION: [DEPENDS_ON] ->[SupersetContextExtractor]
# @RELATION: [DEPENDS_ON] ->[SupersetCompilationAdapter]
# @RELATION: [DEPENDS_ON] ->[TaskManager]
# @PRE: session mutations must execute inside a persisted session boundary scoped to one authenticated user.
# @POST: state transitions are persisted atomically and emit observable progress for long-running steps.
# @SIDE_EFFECT: creates task records, updates session aggregates, triggers upstream Superset calls, persists audit artifacts.
# @DATA_CONTRACT: Input[SessionCommand] -> Output[DatasetReviewSession | CompiledPreview | DatasetRunContext]
# @INVARIANT: Launch is blocked unless a current session has no open blocking findings, all launch-sensitive mappings are approved, and a non-stale Superset-generated compiled preview matches the current input fingerprint.
from __future__ import annotations
# [DEF:DatasetReviewOrchestrator.imports:Block]
from dataclasses import dataclass, field
from typing import Any, Dict, List, Optional
from src.core.config_manager import ConfigManager
from src.core.logger import belief_scope, logger
from src.core.task_manager import TaskManager
from src.core.utils.superset_context_extractor import (
SupersetContextExtractor,
SupersetParsedContext,
)
from src.models.auth import User
from src.models.dataset_review import (
BusinessSummarySource,
ConfidenceState,
DatasetProfile,
DatasetReviewSession,
FindingArea,
FindingSeverity,
RecommendedAction,
ReadinessState,
ResolutionState,
SessionPhase,
SessionStatus,
ValidationFinding,
)
from src.services.dataset_review.repositories.session_repository import (
DatasetReviewSessionRepository,
)
from src.services.dataset_review.semantic_resolver import SemanticSourceResolver
# [/DEF:DatasetReviewOrchestrator.imports:Block]
# [DEF:StartSessionCommand:Class]
# @COMPLEXITY: 2
# @PURPOSE: Typed input contract for starting a dataset review session.
@dataclass
class StartSessionCommand:
user: User
environment_id: str
source_kind: str
source_input: str
# [/DEF:StartSessionCommand:Class]
# [DEF:StartSessionResult:Class]
# @COMPLEXITY: 2
# @PURPOSE: Session-start result carrying the persisted session and intake recovery metadata.
@dataclass
class StartSessionResult:
session: DatasetReviewSession
parsed_context: Optional[SupersetParsedContext] = None
findings: List[ValidationFinding] = field(default_factory=list)
# [/DEF:StartSessionResult:Class]
# [DEF:DatasetReviewOrchestrator:Class]
# @COMPLEXITY: 5
# @PURPOSE: Coordinate safe session startup while preserving cross-user isolation and explicit partial recovery.
# @RELATION: [DEPENDS_ON] ->[DatasetReviewSessionRepository]
# @RELATION: [DEPENDS_ON] ->[SupersetContextExtractor]
# @RELATION: [DEPENDS_ON] ->[TaskManager]
# @RELATION: [DEPENDS_ON] ->[SessionRepo]
# @RELATION: [DEPENDS_ON] ->[ConfigManager]
# @PRE: constructor dependencies are valid and tied to the current request/task scope.
# @POST: orchestrator instance can execute session-scoped mutations for one authenticated user.
# @SIDE_EFFECT: downstream operations may persist session/profile/finding state and enqueue background tasks.
# @DATA_CONTRACT: Input[StartSessionCommand] -> Output[StartSessionResult]
# @INVARIANT: session ownership is preserved on every mutation and recovery remains explicit when partial.
class DatasetReviewOrchestrator:
# [DEF:DatasetReviewOrchestrator.__init__:Function]
# @COMPLEXITY: 3
# @PURPOSE: Bind repository, config, and task dependencies required by the orchestration boundary.
# @RELATION: [DEPENDS_ON] ->[SessionRepo]
# @RELATION: [DEPENDS_ON] ->[ConfigManager]
def __init__(
self,
repository: DatasetReviewSessionRepository,
config_manager: ConfigManager,
task_manager: Optional[TaskManager] = None,
semantic_resolver: Optional[SemanticSourceResolver] = None,
) -> None:
self.repository = repository
self.config_manager = config_manager
self.task_manager = task_manager
self.semantic_resolver = semantic_resolver or SemanticSourceResolver()
# [/DEF:DatasetReviewOrchestrator.__init__:Function]
# [DEF:DatasetReviewOrchestrator.start_session:Function]
# @COMPLEXITY: 5
# @PURPOSE: Initialize a new session from a Superset link or dataset selection and trigger context recovery.
# @RELATION: [DEPENDS_ON] ->[SessionRepo]
# @RELATION: [CALLS] ->[SupersetContextExtractor.parse_superset_link]
# @RELATION: [CALLS] ->[create_task]
# @PRE: source input is non-empty and environment is accessible.
# @POST: session exists in persisted storage with intake/recovery state and task linkage when async work is required.
# @SIDE_EFFECT: persists session and may enqueue recovery task.
# @DATA_CONTRACT: Input[StartSessionCommand] -> Output[StartSessionResult]
# @INVARIANT: no cross-user session leakage occurs; session and follow-up task remain owned by the authenticated user.
def start_session(self, command: StartSessionCommand) -> StartSessionResult:
with belief_scope("DatasetReviewOrchestrator.start_session"):
normalized_source_kind = str(command.source_kind or "").strip()
normalized_source_input = str(command.source_input or "").strip()
normalized_environment_id = str(command.environment_id or "").strip()
if not normalized_source_input:
logger.explore("Blocked dataset review session start due to empty source input")
raise ValueError("source_input must be non-empty")
if normalized_source_kind not in {"superset_link", "dataset_selection"}:
logger.explore(
"Blocked dataset review session start due to unsupported source kind",
extra={"source_kind": normalized_source_kind},
)
raise ValueError("source_kind must be 'superset_link' or 'dataset_selection'")
environment = self.config_manager.get_environment(normalized_environment_id)
if environment is None:
logger.explore(
"Blocked dataset review session start because environment was not found",
extra={"environment_id": normalized_environment_id},
)
raise ValueError("Environment not found")
logger.reason(
"Starting dataset review session",
extra={
"user_id": command.user.id,
"environment_id": normalized_environment_id,
"source_kind": normalized_source_kind,
},
)
parsed_context: Optional[SupersetParsedContext] = None
findings: List[ValidationFinding] = []
dataset_ref = normalized_source_input
dataset_id: Optional[int] = None
dashboard_id: Optional[int] = None
readiness_state = ReadinessState.IMPORTING
recommended_action = RecommendedAction.REVIEW_DOCUMENTATION
current_phase = SessionPhase.RECOVERY
if normalized_source_kind == "superset_link":
extractor = SupersetContextExtractor(environment)
parsed_context = extractor.parse_superset_link(normalized_source_input)
dataset_ref = parsed_context.dataset_ref
dataset_id = parsed_context.dataset_id
dashboard_id = parsed_context.dashboard_id
if parsed_context.partial_recovery:
readiness_state = ReadinessState.RECOVERY_REQUIRED
recommended_action = RecommendedAction.REVIEW_DOCUMENTATION
findings.extend(self._build_partial_recovery_findings(parsed_context))
else:
readiness_state = ReadinessState.REVIEW_READY
else:
dataset_ref, dataset_id = self._parse_dataset_selection(normalized_source_input)
readiness_state = ReadinessState.REVIEW_READY
current_phase = SessionPhase.REVIEW
session = DatasetReviewSession(
user_id=command.user.id,
environment_id=normalized_environment_id,
source_kind=normalized_source_kind,
source_input=normalized_source_input,
dataset_ref=dataset_ref,
dataset_id=dataset_id,
dashboard_id=dashboard_id,
readiness_state=readiness_state,
recommended_action=recommended_action,
status=SessionStatus.ACTIVE,
current_phase=current_phase,
)
persisted_session = self.repository.create_session(session)
profile = self._build_initial_profile(
session_id=persisted_session.session_id,
parsed_context=parsed_context,
dataset_ref=dataset_ref,
)
persisted_session = self.repository.save_profile_and_findings(
persisted_session.session_id,
command.user.id,
profile,
findings,
)
active_task_id = self._enqueue_recovery_task(
command=command,
session=persisted_session,
parsed_context=parsed_context,
)
if active_task_id:
persisted_session.active_task_id = active_task_id
self.repository.db.commit()
self.repository.db.refresh(persisted_session)
logger.reason(
"Linked recovery task to started dataset review session",
extra={"session_id": persisted_session.session_id, "task_id": active_task_id},
)
logger.reflect(
"Dataset review session start completed",
extra={
"session_id": persisted_session.session_id,
"dataset_ref": persisted_session.dataset_ref,
"dataset_id": persisted_session.dataset_id,
"dashboard_id": persisted_session.dashboard_id,
"readiness_state": persisted_session.readiness_state.value,
"active_task_id": persisted_session.active_task_id,
"finding_count": len(findings),
},
)
return StartSessionResult(
session=persisted_session,
parsed_context=parsed_context,
findings=findings,
)
# [/DEF:DatasetReviewOrchestrator.start_session:Function]
# [DEF:DatasetReviewOrchestrator._parse_dataset_selection:Function]
# @COMPLEXITY: 3
# @PURPOSE: Normalize dataset-selection payload into canonical session references.
# @RELATION: [DEPENDS_ON] ->[DatasetReviewSession]
def _parse_dataset_selection(self, source_input: str) -> tuple[str, Optional[int]]:
normalized = str(source_input or "").strip()
if not normalized:
raise ValueError("dataset selection input must be non-empty")
if normalized.isdigit():
dataset_id = int(normalized)
return f"dataset:{dataset_id}", dataset_id
if normalized.startswith("dataset:"):
suffix = normalized.split(":", 1)[1].strip()
if suffix.isdigit():
return normalized, int(suffix)
return normalized, None
return normalized, None
# [/DEF:DatasetReviewOrchestrator._parse_dataset_selection:Function]
# [DEF:DatasetReviewOrchestrator._build_initial_profile:Function]
# @COMPLEXITY: 3
# @PURPOSE: Create the first profile snapshot so exports and detail views remain usable immediately after intake.
# @RELATION: [DEPENDS_ON] ->[DatasetProfile]
def _build_initial_profile(
self,
session_id: str,
parsed_context: Optional[SupersetParsedContext],
dataset_ref: str,
) -> DatasetProfile:
dataset_name = dataset_ref.split(".")[-1] if dataset_ref else "Unresolved dataset"
business_summary = (
f"Review session initialized for {dataset_ref}."
if dataset_ref
else "Review session initialized with unresolved dataset context."
)
confidence_state = (
ConfidenceState.MIXED
if parsed_context and parsed_context.partial_recovery
else ConfidenceState.MOSTLY_CONFIRMED
)
return DatasetProfile(
session_id=session_id,
dataset_name=dataset_name or "Unresolved dataset",
schema_name=dataset_ref.split(".")[0] if "." in dataset_ref else None,
business_summary=business_summary,
business_summary_source=BusinessSummarySource.IMPORTED,
description="Initial review profile created from source intake.",
dataset_type="unknown",
is_sqllab_view=False,
completeness_score=0.25,
confidence_state=confidence_state,
has_blocking_findings=False,
has_warning_findings=bool(parsed_context and parsed_context.partial_recovery),
manual_summary_locked=False,
)
# [/DEF:DatasetReviewOrchestrator._build_initial_profile:Function]
# [DEF:DatasetReviewOrchestrator._build_partial_recovery_findings:Function]
# @COMPLEXITY: 4
# @PURPOSE: Project partial Superset intake recovery into explicit findings without blocking session usability.
# @RELATION: [DEPENDS_ON] ->[ValidationFinding]
# @PRE: parsed_context.partial_recovery is true.
# @POST: returns warning-level findings that preserve usable but incomplete state.
# @SIDE_EFFECT: none beyond structured finding creation.
# @DATA_CONTRACT: Input[SupersetParsedContext] -> Output[List[ValidationFinding]]
def _build_partial_recovery_findings(
self,
parsed_context: SupersetParsedContext,
) -> List[ValidationFinding]:
findings: List[ValidationFinding] = []
for unresolved_ref in parsed_context.unresolved_references:
findings.append(
ValidationFinding(
area=FindingArea.SOURCE_INTAKE,
severity=FindingSeverity.WARNING,
code="PARTIAL_SUPERSET_RECOVERY",
title="Superset context recovered partially",
message=(
"Session remains usable, but some Superset context requires review: "
f"{unresolved_ref.replace('_', ' ')}."
),
resolution_state=ResolutionState.OPEN,
caused_by_ref=unresolved_ref,
)
)
return findings
# [/DEF:DatasetReviewOrchestrator._build_partial_recovery_findings:Function]
# [DEF:DatasetReviewOrchestrator._enqueue_recovery_task:Function]
# @COMPLEXITY: 4
# @PURPOSE: Link session start to observable async recovery when task infrastructure is available.
# @RELATION: [CALLS] ->[create_task]
# @PRE: session is already persisted.
# @POST: returns task identifier when a task could be enqueued, otherwise None.
# @SIDE_EFFECT: may create one background task for progressive recovery.
# @DATA_CONTRACT: Input[StartSessionCommand,DatasetReviewSession,SupersetParsedContext|None] -> Output[task_id:str|None]
def _enqueue_recovery_task(
self,
command: StartSessionCommand,
session: DatasetReviewSession,
parsed_context: Optional[SupersetParsedContext],
) -> Optional[str]:
if self.task_manager is None:
logger.reason(
"Dataset review session started without task manager; continuing synchronously",
extra={"session_id": session.session_id},
)
return None
task_params: Dict[str, Any] = {
"session_id": session.session_id,
"user_id": command.user.id,
"environment_id": session.environment_id,
"source_kind": session.source_kind,
"source_input": session.source_input,
"dataset_ref": session.dataset_ref,
"dataset_id": session.dataset_id,
"dashboard_id": session.dashboard_id,
"partial_recovery": bool(parsed_context and parsed_context.partial_recovery),
}
create_task = getattr(self.task_manager, "create_task", None)
if create_task is None:
logger.explore("Task manager has no create_task method; skipping recovery enqueue")
return None
try:
task_object = create_task(
plugin_id="dataset-review-recovery",
params=task_params,
)
except TypeError:
logger.explore(
"Recovery task enqueue skipped because task manager create_task contract is incompatible",
extra={"session_id": session.session_id},
)
return None
task_id = getattr(task_object, "id", None)
return str(task_id) if task_id else None
# [/DEF:DatasetReviewOrchestrator._enqueue_recovery_task:Function]
# [/DEF:DatasetReviewOrchestrator:Class]
# [/DEF:DatasetReviewOrchestrator:Module]