fix(parsers): close stale source jobs

src/apps/parsers/frontend_compat.py

@@ -24,11 +24,13 @@ from apps.parsers.views import (
     TASKS_BY_NAME,
     build_task_kwargs,
 )
+from django.conf import settings
 from django.core.cache import cache
 from django.core.paginator import Paginator
 from django.db.models import CharField, Max, Q
 from django.db.models.functions import Cast
 from django.http import Http404, HttpResponse
+from django.utils import timezone
 from drf_yasg import openapi
 from drf_yasg.utils import swagger_auto_schema
 from rest_framework import status
@@ -44,6 +46,7 @@ SYSTEM_LOGS_TAG = "System Logs"
 ACTIVE_JOB_STATUSES = {"pending", "started", "retry"}
 SUCCESS_LOAD_STATUSES = {"success", "skipped"}
 ERROR_LOAD_STATUSES = {"failed", "failure", "error"}
+STALE_ACTIVE_MAX_AGE_MINUTES = 90
 
 PARSING_SETTINGS_CACHE_KEY = "parsers:frontend_compat:parsing_settings"
 PARSING_SETTINGS_FIELDS = {
@@ -295,6 +298,23 @@ def _serialize_active_job(job) -> dict[str, Any]:
     }
 
 
+def _stale_cutoff():
+    max_age_minutes = int(
+        getattr(
+            settings,
+            "PARSER_STALE_LOAD_MAX_AGE_MINUTES",
+            STALE_ACTIVE_MAX_AGE_MINUTES,
+        )
+    )
+    return timezone.now() - timedelta(minutes=max_age_minutes)
+
+
+def _is_stale_load(load_log: ParserLoadLog | None) -> bool:
+    if load_log is None or load_log.status != "in_progress":
+        return False
+    return load_log.updated_at < _stale_cutoff()
+
+
 def _active_tasks_for_definition(
     definition: FrontendSourceCardDefinition,
 ) -> list[dict]:
@@ -306,6 +326,7 @@ def _active_tasks_for_definition(
     queryset = BackgroundJobService.get_queryset().filter(
         task_name__in=task_names,
         status__in=ACTIVE_JOB_STATUSES,
+        updated_at__gte=_stale_cutoff(),
     )
     return [_serialize_active_job(job) for job in queryset.order_by("-created_at")[:10]]
 
@@ -330,8 +351,16 @@ def _status_for_card(
 ) -> str:
     if not definition.is_available:
         return "unavailable"
-    if active_tasks or (latest_load and latest_load.status == "in_progress"):
+    if active_tasks:
         return "in_progress"
+    if (
+        latest_load
+        and latest_load.status == "in_progress"
+        and not _is_stale_load(latest_load)
+    ):
+        return "in_progress"
+    if latest_load and latest_load.status == "in_progress":
+        return "error"
     if latest_load and latest_load.status in ERROR_LOAD_STATUSES:
         return "error"
     if last_updated_at:
@@ -416,7 +445,13 @@ def _build_source_card(definition: FrontendSourceCardDefinition) -> dict[str, An
         ),
         "last_updated_at": last_updated_at,
         "next_update_at": next_update_at,
-        "error_message": latest_load.error_message if latest_load else "",
+        "error_message": (
+            "Загрузка зависла и будет закрыта cleanup-задачей."
+            if _is_stale_load(latest_load)
+            else latest_load.error_message
+            if latest_load
+            else ""
+        ),
         "task_names": [
             PARSER_SOURCES[source_key].task_name
             for source_key in definition.source_keys

src/apps/parsers/services.py

@@ -416,7 +416,7 @@ class ParserLoadLogService(BaseService[ParserLoadLog]):
         updated = 0
         active_statuses = [JobStatus.PENDING, JobStatus.STARTED, JobStatus.RETRY]
         for log in stale_logs:
-            job = (
+            batch_job = (
                 BackgroundJob.objects.filter(
                     status__in=active_statuses,
                     meta__source=log.source,
@@ -425,6 +425,18 @@ class ParserLoadLogService(BaseService[ParserLoadLog]):
                 .order_by("-updated_at")
                 .first()
             )
+            source_job = None
+            if batch_job is None:
+                source_job = (
+                    BackgroundJob.objects.filter(
+                        status__in=active_statuses,
+                        meta__source=log.source,
+                        meta__batch_id__isnull=True,
+                    )
+                    .order_by("-updated_at")
+                    .first()
+                )
+            job = batch_job or source_job
             if job is not None and job.updated_at >= cutoff:
                 continue
 

src/apps/parsers/source_cards.py

@@ -20,6 +20,7 @@ from apps.parsers.models import (
     ParserLoadLog,
     ProcurementRecord,
 )
+from django.conf import settings
 from django.db.models import Max
 from django.http import Http404
 from django.utils import timezone
@@ -27,6 +28,7 @@ from rest_framework.exceptions import ValidationError
 
 SUCCESSFUL_LOAD_STATUSES = {"success", "skipped"}
 ACTIVE_JOB_STATUSES = [JobStatus.PENDING, JobStatus.STARTED, JobStatus.RETRY]
+STALE_ACTIVE_MAX_AGE_MINUTES = 90
 
 
 @dataclass(frozen=True)
@@ -293,6 +295,9 @@ class SourceCardService:
             latest_load=latest_load,
             last_updated_at=last_updated_at,
         )
+        error_message = latest_load.error_message if latest_load else ""
+        if cls._is_stale_load(latest_load):
+            error_message = cls._stale_load_message()
 
         return {
             "slug": definition.slug,
@@ -307,7 +312,7 @@ class SourceCardService:
             "organizations_count": organizations_count,
             "last_updated_at": last_updated_at,
             "next_update_at": cls._get_next_update_at(definition, last_updated_at),
-            "error_message": latest_load.error_message if latest_load else "",
+            "error_message": error_message,
             "task_names": list(definition.task_names),
             "refresh_requires_params": any(
                 param.required for param in definition.refresh_params
@@ -721,9 +726,11 @@ class SourceCardService:
     def _get_active_tasks(
         cls, definition: SourceCardDefinition
     ) -> list[dict[str, Any]]:
+        cutoff = cls._stale_cutoff()
         queryset = BackgroundJobService.get_queryset().filter(
             task_name__in=definition.task_names,
             status__in=ACTIVE_JOB_STATUSES,
+            updated_at__gte=cutoff,
         )
         return [
             cls._serialize_job(job) for job in queryset.order_by("-created_at")[:10]
@@ -749,14 +756,44 @@ class SourceCardService:
             return "unavailable"
         if active_tasks:
             return "in_progress"
-        if latest_load and latest_load.status == "in_progress":
+        if (
+            latest_load
+            and latest_load.status == "in_progress"
+            and not cls._is_stale_load(latest_load)
+        ):
             return "in_progress"
+        if latest_load and latest_load.status == "in_progress":
+            return "error"
         if latest_load and latest_load.status == "failed":
             return "error"
         if last_updated_at:
             return "success"
         return "idle"
 
+    @classmethod
+    def _stale_cutoff(cls):
+        max_age_minutes = int(
+            getattr(
+                settings,
+                "PARSER_STALE_LOAD_MAX_AGE_MINUTES",
+                STALE_ACTIVE_MAX_AGE_MINUTES,
+            )
+        )
+        return timezone.now() - timedelta(minutes=max_age_minutes)
+
+    @classmethod
+    def _is_stale_load(cls, latest_load: ParserLoadLog | None) -> bool:
+        if latest_load is None or latest_load.status != "in_progress":
+            return False
+        updated_at = getattr(latest_load, "updated_at", None)
+        if updated_at is None:
+            return False
+        return updated_at < cls._stale_cutoff()
+
+    @staticmethod
+    def _stale_load_message() -> str:
+        return "Загрузка зависла и будет закрыта cleanup-задачей."
+
     @classmethod
     def _get_status_label(cls, status: str) -> str:
         labels = {

src/apps/parsers/tasks.py

@@ -96,19 +96,27 @@ def _get_or_create_background_job(
 ):
     """Reuse a pre-created job or create a new one for the task."""
     job = BackgroundJobService.get_by_task_id_or_none(task_id)
+    payload = {"source": source, **(meta or {})}
+    if batch_id is not None:
+        payload["batch_id"] = batch_id
     if not job:
-        payload = {"source": source, **(meta or {})}
-        if batch_id is not None:
-            payload["batch_id"] = batch_id
         job = BackgroundJobService.create_job(
             task_id=task_id,
             task_name=task_name,
             user_id=requested_by_id,
             meta=payload,
         )
-    elif requested_by_id is not None and job.user_id is None:
-        job.user_id = requested_by_id
-        job.save(update_fields=["user_id", "updated_at"])
+    else:
+        update_fields = []
+        merged_meta = {**(job.meta or {}), **payload}
+        if merged_meta != job.meta:
+            job.meta = merged_meta
+            update_fields.append("meta")
+        if requested_by_id is not None and job.user_id is None:
+            job.user_id = requested_by_id
+            update_fields.append("user_id")
+        if update_fields:
+            job.save(update_fields=[*update_fields, "updated_at"])
     return job
 
 
@@ -1930,19 +1938,27 @@ def parse_fstec_registers(
 
 @shared_task
 def cleanup_stale_parser_loads(max_age_minutes: int | None = None) -> dict:
-    """Закрыть stale in_progress загрузки после рестартов worker/deploy."""
+    """Закрыть stale in_progress загрузки и jobs после рестартов worker/deploy."""
     if max_age_minutes is None:
         max_age_minutes = getattr(
             settings,
             "PARSER_STALE_LOAD_MAX_AGE_MINUTES",
             PARSER_STALE_LOAD_MAX_AGE_MINUTES,
         )
+    source_values = {descriptor.source for descriptor in PARSER_SOURCES.values()}
+    task_names = {descriptor.task_name for descriptor in PARSER_SOURCES.values()}
     marked_failed = ParserLoadLogService.mark_stale_in_progress_failed(
         max_age_minutes=int(max_age_minutes)
     )
+    marked_jobs_failed = BackgroundJobService.mark_stale_active_jobs_failed(
+        max_age_minutes=int(max_age_minutes),
+        task_names=task_names,
+        meta_sources=source_values,
+    )
     return {
         "status": "success",
         "marked_failed": marked_failed,
+        "marked_jobs_failed": marked_jobs_failed,
         "max_age_minutes": int(max_age_minutes),
     }