feat(worker): ручной пересчёт метрик через очередь, диагностика шагов и heartbeat

POST /metrics/refresh ставит задачу в sync_job (source=METRICS_JOB) и отвечает 202, пересчёт делает worker; GET /metrics/status отдаёт refreshing и consistent. metric_refresh_log хранит failed_step и step_timings. Источники с needs="tinvest_token" не попадают в расписание без токена, tinvest/moex добавлены в default_schedule. Воркер трогает heartbeat-файл для healthcheck.

backend/src/fintracker/api/routers/metrics.py

@@ -4,9 +4,9 @@ from fastapi import APIRouter, status
 from sqlalchemy import select
 
 from fintracker.api.deps import CurrentUser, SessionDep
-from fintracker.api.schemas.metrics import DataQualityRow, RefreshLogOut
-from fintracker.metrics.refresh import refresh_all
-from fintracker.models import MetricDataQuality, MetricRefreshLog
+from fintracker.api.schemas.metrics import DataQualityRow, MetricsStatusOut, RefreshLogOut
+from fintracker.api.schemas.sync import SyncJobOut
+from fintracker.models import METRICS_JOB, JobStatus, MetricDataQuality, MetricRefreshLog, SyncJob
 
 router = APIRouter(tags=["metrics"])
 
@@ -22,9 +22,10 @@ async def data_quality(session: SessionDep, _: CurrentUser) -> list[DataQualityR
 
 
 @router.get("/metrics/status", name="status")
-async def metrics_status(session: SessionDep, _: CurrentUser) -> RefreshLogOut | None:
-    """When the metric tables were last rebuilt, and whether it failed."""
-    row = (
+async def metrics_status(session: SessionDep, _: CurrentUser) -> MetricsStatusOut:
+    """When the metric tables were last rebuilt, whether they are one consistent snapshot,
+    and whether another rebuild is on its way."""
+    latest = (
         (
             await session.execute(
                 select(MetricRefreshLog).order_by(MetricRefreshLog.started_at.desc()).limit(1)
@@ -33,13 +34,54 @@ async def metrics_status(session: SessionDep, _: CurrentUser) -> RefreshLogOut |
         .scalars()
         .first()
     )
-    if row is None:
-        return None
-    return RefreshLogOut.model_validate(row, from_attributes=True)
+    last_finished = (
+        (
+            await session.execute(
+                select(MetricRefreshLog)
+                .where(MetricRefreshLog.finished_at.is_not(None))
+                .order_by(MetricRefreshLog.started_at.desc())
+                .limit(1)
+            )
+        )
+        .scalars()
+        .first()
+    )
+    queued = (
+        await session.execute(
+            select(SyncJob.id)
+            .where(
+                SyncJob.source == METRICS_JOB,
+                SyncJob.status.in_([JobStatus.queued, JobStatus.running]),
+            )
+            .limit(1)
+        )
+    ).first()
+    return MetricsStatusOut(
+        last_refresh=RefreshLogOut.model_validate(latest, from_attributes=True) if latest else None,
+        consistent=last_finished is None or last_finished.error is None,
+        refreshing=queued is not None or (latest is not None and latest.finished_at is None),
+    )
 
 
 @router.post("/metrics/refresh", name="refresh", status_code=status.HTTP_202_ACCEPTED)
-async def metrics_refresh(session: SessionDep, _: CurrentUser) -> RefreshLogOut:
-    """Rebuild every metric_* table inline (seconds at personal volumes)."""
-    entry = await refresh_all(session, trigger="manual")
-    return RefreshLogOut.model_validate(entry, from_attributes=True)
+async def metrics_refresh(session: SessionDep, _: CurrentUser) -> SyncJobOut:
+    """Queue a rebuild of every metric_* table; the worker runs it. Poll `/metrics/status`
+    until `refreshing` is false.
+
+    A request while one is already waiting shares it. One that arrives while a rebuild is
+    RUNNING queues another, because the running one may have read the data before the change
+    that prompted this call."""
+    waiting = (
+        await session.execute(
+            select(SyncJob)
+            .where(SyncJob.source == METRICS_JOB, SyncJob.status == JobStatus.queued)
+            .limit(1)
+        )
+    ).scalar_one_or_none()
+    if waiting is not None:
+        return SyncJobOut.model_validate(waiting, from_attributes=True)
+    job = SyncJob(source=METRICS_JOB, status=JobStatus.queued)
+    session.add(job)
+    await session.commit()
+    await session.refresh(job)
+    return SyncJobOut.model_validate(job, from_attributes=True)

backend/src/fintracker/api/schemas/metrics.py

@@ -23,3 +23,17 @@ class RefreshLogOut(BaseModel):
     finished_at: datetime | None
     trigger: str
     error: str | None
+    failed_step: str | None
+    """The step that raised; earlier steps committed, later ones did not run."""
+    step_timings: dict[str, float] | None
+    """Seconds per step that completed, in run order."""
+
+
+class MetricsStatusOut(BaseModel):
+    last_refresh: RefreshLogOut | None
+    """Latest run, finished or not; null before the first one."""
+    consistent: bool
+    """False when the latest FINISHED run failed part-way: some metric tables are from it and
+    the rest from an earlier run, so numbers on different screens may disagree."""
+    refreshing: bool
+    """A rebuild is queued or running, so `last_refresh` is about to be superseded."""