fa2c7d7525
GET /api/analyses?documentId=X retournait toutes les analyses (le query param était ignoré). Le frontend prenait alors la première COMPLETED de la réponse, qui pouvait être celle d'un autre doc → bboxes d'un autre doc projetées sur l'image en cours. Backend - analysis_repo: nouvelle méthode find_by_document(document_id, limit, offset) - analysis_service: expose find_by_document - api/analyses: GET /api/analyses accepte ?documentId=... (alias Pydantic pour respecter la règle ruff N803). Si présent, filtre via la nouvelle méthode; sinon comportement inchangé. - test: test_list_analyses_filtered_by_document vérifie le routing vers find_by_document quand le query param est fourni. Frontend (defensive) - DocInspectTab / DocAskTab: filtre client-side analyses.find(a => a.documentId === requestedId && a.status === 'COMPLETED') pour rester safe même si un backend antérieur ignore le param.
526 lines
20 KiB
Python
526 lines
20 KiB
Python
"""Analysis service — async document parsing orchestration.
|
|
|
|
Uses injected ports (converter, chunker, repositories) so the service is
|
|
decoupled from infrastructure implementations.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
import functools
|
|
import json
|
|
import logging
|
|
import math
|
|
from dataclasses import asdict, dataclass
|
|
from typing import TYPE_CHECKING
|
|
|
|
import pypdfium2 as pdfium
|
|
|
|
from domain.exceptions import InvalidLifecycleTransitionError
|
|
from domain.models import AnalysisJob, AnalysisStatus
|
|
from domain.services import classify_error, merge_results
|
|
from domain.value_objects import (
|
|
ChunkingOptions,
|
|
ChunkResult,
|
|
ConversionOptions,
|
|
ConversionResult,
|
|
DocumentLifecycleState,
|
|
)
|
|
|
|
if TYPE_CHECKING:
|
|
from domain.ports import (
|
|
AnalysisRepository,
|
|
DocumentChunker,
|
|
DocumentConverter,
|
|
DocumentRepository,
|
|
)
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
def _chunk_to_dict(c: ChunkResult) -> dict:
|
|
"""Serialize ChunkResult to a camelCase dict matching the frontend API contract."""
|
|
return {
|
|
"text": c.text,
|
|
"headings": c.headings,
|
|
"sourcePage": c.source_page,
|
|
"tokenCount": c.token_count,
|
|
"bboxes": [{"page": b.page, "bbox": b.bbox} for b in c.bboxes],
|
|
"docItems": [{"selfRef": d.self_ref, "label": d.label} for d in c.doc_items],
|
|
}
|
|
|
|
|
|
# Maximum number of concurrent analysis jobs to prevent resource exhaustion.
|
|
_DEFAULT_MAX_CONCURRENT = 3
|
|
|
|
|
|
def _count_pdf_pages(file_path: str) -> int:
|
|
"""Count pages in a PDF. Returns 0 if the file is not a valid PDF."""
|
|
try:
|
|
pdf = pdfium.PdfDocument(file_path)
|
|
count = len(pdf)
|
|
pdf.close()
|
|
return count
|
|
except Exception:
|
|
logger.debug("Cannot open %s as PDF, batching disabled", file_path)
|
|
return 0
|
|
|
|
|
|
@dataclass
|
|
class AnalysisConfig:
|
|
"""Configuration values needed by AnalysisService, extracted from settings."""
|
|
|
|
default_table_mode: str = "accurate"
|
|
batch_page_size: int = 0
|
|
neo4j_required: bool = False # if True, ingestion fails when Neo4j write fails
|
|
|
|
|
|
class AnalysisService:
|
|
"""Orchestrates document analysis using injected ports."""
|
|
|
|
def __init__(
|
|
self,
|
|
converter: DocumentConverter,
|
|
analysis_repo: AnalysisRepository,
|
|
document_repo: DocumentRepository,
|
|
chunker: DocumentChunker | None = None,
|
|
conversion_timeout: int = 600,
|
|
max_concurrent: int = _DEFAULT_MAX_CONCURRENT,
|
|
config: AnalysisConfig | None = None,
|
|
neo4j_driver=None,
|
|
):
|
|
self._converter = converter
|
|
self._chunker = chunker
|
|
self._analysis_repo = analysis_repo
|
|
self._document_repo = document_repo
|
|
self._conversion_timeout = conversion_timeout
|
|
self._semaphore = asyncio.Semaphore(max_concurrent)
|
|
self._running_tasks: dict[str, asyncio.Task] = {}
|
|
self._background_tasks: set[asyncio.Task] = set()
|
|
self._config = config or AnalysisConfig()
|
|
self._neo4j = neo4j_driver
|
|
# Duck-typed callable injected at startup. Wired in main.py to
|
|
# `ChunkService.promote_from_analysis_if_empty` so the canonical
|
|
# chunkset (#205) is populated on the first successful analysis,
|
|
# making the Doc workspace tab functional immediately (#256).
|
|
# Optional: when None, analysis behaviour is unchanged.
|
|
self._chunk_promoter = None
|
|
|
|
def set_chunk_promoter(self, chunk_service) -> None:
|
|
"""Inject the canonical-chunk promoter (post-construction wiring).
|
|
|
|
Kept loosely typed to avoid an import cycle between
|
|
`analysis_service` and `chunk_service`. The contract is a
|
|
coroutine `(document_id: str, chunks_json: str) -> int`.
|
|
"""
|
|
self._chunk_promoter = chunk_service
|
|
|
|
async def create(
|
|
self,
|
|
document_id: str,
|
|
*,
|
|
pipeline_options: dict | None = None,
|
|
chunking_options: dict | None = None,
|
|
) -> AnalysisJob:
|
|
"""Create a new analysis job and launch background processing."""
|
|
doc = await self._document_repo.find_by_id(document_id)
|
|
if not doc:
|
|
raise ValueError(f"Document not found: {document_id}")
|
|
|
|
job = AnalysisJob(document_id=document_id)
|
|
job.document_filename = doc.filename
|
|
await self._analysis_repo.insert(job)
|
|
|
|
task = asyncio.create_task(
|
|
self._run_analysis(
|
|
job.id,
|
|
doc.storage_path,
|
|
doc.filename,
|
|
pipeline_options,
|
|
chunking_options,
|
|
)
|
|
)
|
|
self._running_tasks[job.id] = task
|
|
task.add_done_callback(functools.partial(self._on_task_done, job_id=job.id))
|
|
|
|
return job
|
|
|
|
async def find_all(self) -> list[AnalysisJob]:
|
|
"""Return all analysis jobs, newest first."""
|
|
return await self._analysis_repo.find_all()
|
|
|
|
async def find_by_document(self, document_id: str) -> list[AnalysisJob]:
|
|
"""Return analysis jobs for a given document, newest first."""
|
|
return await self._analysis_repo.find_by_document(document_id)
|
|
|
|
async def find_by_id(self, job_id: str) -> AnalysisJob | None:
|
|
"""Find an analysis job by ID, or return None."""
|
|
return await self._analysis_repo.find_by_id(job_id)
|
|
|
|
async def delete(self, job_id: str) -> bool:
|
|
"""Delete an analysis job, cancelling any running task. Returns True if it existed."""
|
|
task = self._running_tasks.pop(job_id, None)
|
|
if task and not task.done():
|
|
task.cancel()
|
|
logger.info("Cancelled running task for job %s", job_id)
|
|
return await self._analysis_repo.delete(job_id)
|
|
|
|
async def rechunk(self, job_id: str, chunking_options: dict) -> list[ChunkResult]:
|
|
"""Re-chunk an existing completed analysis with new options."""
|
|
job = await self._analysis_repo.find_by_id(job_id)
|
|
if not job:
|
|
raise ValueError(f"Analysis not found: {job_id}")
|
|
if job.status != AnalysisStatus.COMPLETED:
|
|
raise ValueError(f"Analysis is not completed: {job_id}")
|
|
if not job.document_json:
|
|
raise ValueError(f"No document data available for re-chunking: {job_id}")
|
|
if not self._chunker:
|
|
raise ValueError("Chunking is not available")
|
|
|
|
options = ChunkingOptions(**chunking_options)
|
|
chunks = await self._chunker.chunk(job.document_json, options)
|
|
|
|
chunks_json = json.dumps([_chunk_to_dict(c) for c in chunks])
|
|
await self._analysis_repo.update_chunks(job_id, chunks_json)
|
|
|
|
# Re-chunk drives the document into Chunked (idempotent if already
|
|
# Chunked; #204 will mark per-store links Stale separately).
|
|
await self._transition_document(job.document_id, DocumentLifecycleState.CHUNKED)
|
|
|
|
return chunks
|
|
|
|
async def update_chunk_text(self, job_id: str, chunk_index: int, text: str) -> list[dict]:
|
|
"""Update the text of a single chunk by index. Returns the full updated chunks list."""
|
|
job = await self._analysis_repo.find_by_id(job_id)
|
|
if not job:
|
|
raise ValueError(f"Analysis not found: {job_id}")
|
|
if job.status != AnalysisStatus.COMPLETED:
|
|
raise ValueError(f"Analysis is not completed: {job_id}")
|
|
if not job.chunks_json:
|
|
raise ValueError(f"No chunks available: {job_id}")
|
|
|
|
chunks = json.loads(job.chunks_json)
|
|
if chunk_index < 0 or chunk_index >= len(chunks):
|
|
raise ValueError(f"Chunk index out of range: {chunk_index}")
|
|
|
|
chunks[chunk_index]["text"] = text
|
|
chunks[chunk_index]["modified"] = True
|
|
|
|
chunks_json = json.dumps(chunks)
|
|
await self._analysis_repo.update_chunks(job_id, chunks_json)
|
|
|
|
return chunks
|
|
|
|
async def delete_chunk(self, job_id: str, chunk_index: int) -> list[dict]:
|
|
"""Soft-delete a chunk by index. Returns the full updated chunks list."""
|
|
job = await self._analysis_repo.find_by_id(job_id)
|
|
if not job:
|
|
raise ValueError(f"Analysis not found: {job_id}")
|
|
if job.status != AnalysisStatus.COMPLETED:
|
|
raise ValueError(f"Analysis is not completed: {job_id}")
|
|
if not job.chunks_json:
|
|
raise ValueError(f"No chunks available: {job_id}")
|
|
|
|
chunks = json.loads(job.chunks_json)
|
|
if chunk_index < 0 or chunk_index >= len(chunks):
|
|
raise ValueError(f"Chunk index out of range: {chunk_index}")
|
|
|
|
chunks[chunk_index]["deleted"] = True
|
|
|
|
chunks_json = json.dumps(chunks)
|
|
await self._analysis_repo.update_chunks(job_id, chunks_json)
|
|
|
|
return chunks
|
|
|
|
async def _run_batched_conversion(
|
|
self,
|
|
job_id: str,
|
|
file_path: str,
|
|
options: ConversionOptions,
|
|
total_pages: int,
|
|
batch_size: int,
|
|
) -> ConversionResult | None:
|
|
"""Convert a document in batches using page_range.
|
|
|
|
Returns None if the job was deleted mid-batch (caller should abort).
|
|
Raises on batch failure (fail-fast: entire job fails).
|
|
"""
|
|
num_batches = math.ceil(total_pages / batch_size)
|
|
await self._analysis_repo.update_progress(job_id, 0, total_pages)
|
|
logger.info(
|
|
"Batched conversion: %d pages in %d batches of %d for job %s",
|
|
total_pages,
|
|
num_batches,
|
|
batch_size,
|
|
job_id,
|
|
)
|
|
|
|
results: list[ConversionResult] = []
|
|
for batch_idx in range(num_batches):
|
|
start = batch_idx * batch_size + 1
|
|
end = min(start + batch_size - 1, total_pages)
|
|
|
|
if not await self._analysis_repo.find_by_id(job_id):
|
|
logger.info(
|
|
"Job %s deleted during batch %d/%d, aborting",
|
|
job_id,
|
|
batch_idx + 1,
|
|
num_batches,
|
|
)
|
|
return None
|
|
|
|
try:
|
|
batch_result = await asyncio.wait_for(
|
|
self._converter.convert(file_path, options, page_range=(start, end)),
|
|
timeout=self._conversion_timeout,
|
|
)
|
|
except Exception as exc:
|
|
raise RuntimeError(
|
|
f"Batch {batch_idx + 1}/{num_batches} (pages {start}-{end}) failed: {exc}"
|
|
) from exc
|
|
|
|
results.append(batch_result)
|
|
await self._analysis_repo.update_progress(job_id, end, total_pages)
|
|
logger.info(
|
|
"Batch %d/%d done (pages %d-%d) for job %s",
|
|
batch_idx + 1,
|
|
num_batches,
|
|
start,
|
|
end,
|
|
job_id,
|
|
)
|
|
|
|
return merge_results(results)
|
|
|
|
def _on_task_done(self, task: asyncio.Task, *, job_id: str) -> None:
|
|
"""Cleanup running tasks and handle failures."""
|
|
self._running_tasks.pop(job_id, None)
|
|
if task.cancelled():
|
|
logger.warning("Analysis task was cancelled: %s", job_id)
|
|
self._schedule_mark_failed(job_id, "Task was cancelled")
|
|
return
|
|
exc = task.exception()
|
|
if exc:
|
|
logger.error("Unhandled exception in analysis task %s: %s", job_id, exc, exc_info=True)
|
|
self._schedule_mark_failed(job_id, classify_error(exc))
|
|
|
|
def _schedule_mark_failed(self, job_id: str, error: str) -> None:
|
|
"""Schedule _mark_failed as a tracked background task."""
|
|
t = asyncio.ensure_future(self._mark_failed(job_id, error))
|
|
self._background_tasks.add(t)
|
|
t.add_done_callback(self._background_tasks.discard)
|
|
|
|
async def _mark_failed(self, job_id: str, error: str) -> None:
|
|
"""Safely mark a job as failed, handling DB errors gracefully."""
|
|
try:
|
|
job = await self._analysis_repo.find_by_id(job_id)
|
|
if job:
|
|
job.mark_failed(error)
|
|
await self._analysis_repo.update_status(job)
|
|
await self._transition_document(job.document_id, DocumentLifecycleState.FAILED)
|
|
except OSError:
|
|
logger.exception("Database I/O error marking job %s as failed", job_id)
|
|
except Exception:
|
|
logger.exception("Unexpected error marking job %s as failed", job_id)
|
|
|
|
async def _transition_document(
|
|
self,
|
|
document_id: str,
|
|
target: DocumentLifecycleState,
|
|
) -> None:
|
|
"""Drive a Document lifecycle transition (#202).
|
|
|
|
Idempotent on the target — if the document is already in the
|
|
requested state, no write happens. Invalid transitions are
|
|
logged at WARNING and swallowed so a lifecycle hiccup does not
|
|
crash an otherwise-successful pipeline run.
|
|
"""
|
|
doc = await self._document_repo.find_by_id(document_id)
|
|
if doc is None:
|
|
return
|
|
if doc.lifecycle_state == target:
|
|
return
|
|
try:
|
|
event = doc.transition_to(target)
|
|
except InvalidLifecycleTransitionError:
|
|
logger.warning(
|
|
"Skipped invalid lifecycle transition for doc %s: %s -> %s",
|
|
document_id,
|
|
doc.lifecycle_state.value,
|
|
target.value,
|
|
)
|
|
return
|
|
await self._document_repo.update_lifecycle(document_id, event.current, event.at)
|
|
logger.info(
|
|
"lifecycle_changed doc_id=%s from=%s to=%s",
|
|
document_id,
|
|
event.previous.value,
|
|
event.current.value,
|
|
)
|
|
|
|
async def _run_analysis(
|
|
self,
|
|
job_id: str,
|
|
file_path: str,
|
|
filename: str,
|
|
pipeline_options: dict | None = None,
|
|
chunking_options: dict | None = None,
|
|
) -> None:
|
|
"""Background task: run conversion and optionally chunk.
|
|
|
|
Acquires the concurrency semaphore to limit parallel conversions
|
|
and prevent CPU/memory exhaustion on modest hardware.
|
|
"""
|
|
async with self._semaphore:
|
|
await self._run_analysis_inner(
|
|
job_id, file_path, filename, pipeline_options, chunking_options
|
|
)
|
|
|
|
def _build_conversion_options(self, pipeline_options: dict | None) -> ConversionOptions:
|
|
"""Build ConversionOptions, applying default table_mode if not specified."""
|
|
opts_dict = pipeline_options or {}
|
|
if "table_mode" not in opts_dict:
|
|
opts_dict = {**opts_dict, "table_mode": self._config.default_table_mode}
|
|
return ConversionOptions(**opts_dict)
|
|
|
|
async def _run_conversion(
|
|
self,
|
|
job_id: str,
|
|
file_path: str,
|
|
options: ConversionOptions,
|
|
) -> ConversionResult | None:
|
|
"""Run batched or single conversion. Returns None if the job was deleted mid-batch.
|
|
|
|
Batching is only used for local mode — it limits memory usage when
|
|
Docling runs in-process. In remote mode the Serve instance manages
|
|
its own resources, and batching would discard document_json (needed
|
|
for chunking).
|
|
"""
|
|
total_pages = _count_pdf_pages(file_path)
|
|
batch_size = self._config.batch_page_size
|
|
if batch_size > 0 and total_pages > batch_size and self._converter.supports_page_batching:
|
|
return await self._run_batched_conversion(
|
|
job_id, file_path, options, total_pages, batch_size
|
|
)
|
|
return await asyncio.wait_for(
|
|
self._converter.convert(file_path, options),
|
|
timeout=self._conversion_timeout,
|
|
)
|
|
|
|
async def _finalize_analysis(
|
|
self,
|
|
job_id: str,
|
|
result: ConversionResult,
|
|
chunking_options: dict | None,
|
|
) -> None:
|
|
"""Serialize results, optionally chunk, mark job completed, update page count."""
|
|
pages_json = json.dumps([asdict(p) for p in result.pages])
|
|
|
|
chunks_json = None
|
|
if chunking_options and self._chunker and result.document_json:
|
|
chunk_opts = ChunkingOptions(**chunking_options)
|
|
chunks = await self._chunker.chunk(result.document_json, chunk_opts)
|
|
chunks_json = json.dumps([_chunk_to_dict(c) for c in chunks])
|
|
logger.info("Chunking produced %d chunks for job %s", len(chunks), job_id)
|
|
|
|
# Re-read the job so we don't lose progress_current/progress_total
|
|
# written to the DB during batched conversion.
|
|
job = await self._analysis_repo.find_by_id(job_id)
|
|
if not job:
|
|
return
|
|
job.mark_completed(
|
|
markdown=result.content_markdown,
|
|
html=result.content_html,
|
|
pages_json=pages_json,
|
|
document_json=result.document_json,
|
|
chunks_json=chunks_json,
|
|
)
|
|
await self._analysis_repo.update_status(job)
|
|
|
|
if result.page_count:
|
|
await self._document_repo.update_page_count(job.document_id, result.page_count)
|
|
|
|
# Promote chunks into the canonical doc-centric chunkset on the
|
|
# first analysis (#256). Idempotent: a doc that already has
|
|
# canonical chunks is left untouched, so subsequent ad-hoc
|
|
# analyses (Studio / OCR Debug) never overwrite hand-edited state.
|
|
if chunks_json is not None and self._chunk_promoter is not None:
|
|
try:
|
|
await self._chunk_promoter.promote_from_analysis_if_empty(
|
|
job.document_id, chunks_json
|
|
)
|
|
except Exception:
|
|
# Promotion is a best-effort hook — never fail an analysis
|
|
# because the canonical promotion hit a snag.
|
|
logger.exception("Canonical chunk promotion failed for doc %s", job.document_id)
|
|
|
|
# Drive the document lifecycle (#202): chunks present → Chunked,
|
|
# otherwise → Parsed.
|
|
target_state = (
|
|
DocumentLifecycleState.CHUNKED
|
|
if chunks_json is not None
|
|
else DocumentLifecycleState.PARSED
|
|
)
|
|
await self._transition_document(job.document_id, target_state)
|
|
|
|
await self._write_tree_to_neo4j(job, result.document_json)
|
|
|
|
logger.info("Analysis completed: %s (%d pages)", job_id, result.page_count)
|
|
|
|
async def _write_tree_to_neo4j(self, job, document_json: str | None) -> None:
|
|
"""Mirror the DoclingDocument tree into Neo4j if configured.
|
|
|
|
Silent no-op when Neo4j isn't wired in. Logs but does not fail the
|
|
analysis when the write fails, unless `config.neo4j_required` is set.
|
|
"""
|
|
if self._neo4j is None or not document_json:
|
|
return
|
|
try:
|
|
from infra.neo4j import write_document
|
|
|
|
await write_document(
|
|
self._neo4j,
|
|
doc_id=job.document_id,
|
|
filename=job.document_filename or job.document_id,
|
|
document_json=document_json,
|
|
)
|
|
except Exception:
|
|
logger.exception("Neo4j TreeWriter failed for doc %s", job.document_id)
|
|
if self._config.neo4j_required:
|
|
raise
|
|
|
|
async def _run_analysis_inner(
|
|
self,
|
|
job_id: str,
|
|
file_path: str,
|
|
filename: str,
|
|
pipeline_options: dict | None = None,
|
|
chunking_options: dict | None = None,
|
|
) -> None:
|
|
"""Inner analysis logic — called under the concurrency semaphore."""
|
|
try:
|
|
job = await self._analysis_repo.find_by_id(job_id)
|
|
if not job:
|
|
logger.error("Analysis job %s not found", job_id)
|
|
return
|
|
|
|
job.mark_running()
|
|
await self._analysis_repo.update_status(job)
|
|
logger.info("Analysis started: %s (file: %s)", job_id, filename)
|
|
|
|
options = self._build_conversion_options(pipeline_options)
|
|
result = await self._run_conversion(job_id, file_path, options)
|
|
if result is None:
|
|
return # job was deleted mid-batch
|
|
|
|
await self._finalize_analysis(job_id, result, chunking_options)
|
|
|
|
except TimeoutError:
|
|
logger.error("Analysis timed out after %ds: %s", self._conversion_timeout, job_id)
|
|
await self._mark_failed(
|
|
job_id, f"Conversion timed out after {self._conversion_timeout}s"
|
|
)
|
|
|
|
except Exception as e:
|
|
logger.exception("Analysis failed: %s", job_id)
|
|
await self._mark_failed(job_id, classify_error(e))
|