ss-tools/backend/src/plugins/translate/metrics.py

# #region TranslationMetrics [C:3] [TYPE Module] [SEMANTICS translate,metrics,aggregation]
# @BRIEF Aggregate translation metrics from live TranslationEvent + MetricSnapshot for per-job reporting.
# @LAYER Domain
# @RELATION DEPENDS_ON -> [TranslationEvent]
# @RELATION DEPENDS_ON -> [MetricSnapshot]
# @RELATION DEPENDS_ON -> [TranslationRun]
# @RELATION DEPENDS_ON -> [TranslationSchedule]

from datetime import datetime, timezone
from typing import Any, Dict, List, Optional, Tuple
from sqlalchemy.orm import Session
from sqlalchemy import func

from ...core.logger import logger, belief_scope
from ...models.translate import (
    TranslationEvent,
    MetricSnapshot,
    TranslationRun,
    TranslationSchedule,
)


# #region TranslationMetrics [C:3] [TYPE Class] [SEMANTICS translate,metrics]
# @BRIEF Aggregate translation metrics from live events and MetricSnapshot.
# @RELATION DEPENDS_ON -> [TranslationRun]
# @RELATION DEPENDS_ON -> [MetricSnapshot]
# @RELATION DEPENDS_ON -> [TranslationEvent]
class TranslationMetrics:

    def __init__(self, db: Session):
        self.db = db

    # #region get_job_metrics [C:2] [TYPE Function] [SEMANTICS translate,metrics,job]
    # @BRIEF Get aggregated metrics for a specific job including run counts, record stats, and duration.
    def get_job_metrics(self, job_id: str) -> Dict[str, Any]:
        with belief_scope("TranslationMetrics.get_job_metrics"):
            # Run counts from TranslationRun
            run_counts = (
                self.db.query(
                    TranslationRun.status,
                    func.count(TranslationRun.id),
                )
                .filter(TranslationRun.job_id == job_id)
                .group_by(TranslationRun.status)
                .all()
            )

            total_runs = 0
            status_counts: Dict[str, int] = {}
            for status_val, count in run_counts:
                status_counts[status_val] = count
                total_runs += count

            # Aggregate record stats from runs
            record_stats = (
                self.db.query(
                    func.coalesce(func.sum(TranslationRun.total_records), 0),
                    func.coalesce(func.sum(TranslationRun.successful_records), 0),
                    func.coalesce(func.sum(TranslationRun.failed_records), 0),
                    func.coalesce(func.sum(TranslationRun.skipped_records), 0),
                )
                .filter(TranslationRun.job_id == job_id)
                .first()
            )
            total_records = record_stats[0] if record_stats else 0
            successful_records = record_stats[1] if record_stats else 0
            failed_records = record_stats[2] if record_stats else 0
            skipped_records = record_stats[3] if record_stats else 0

            # Average duration from runs
            avg_duration = (
                self.db.query(
                    func.avg(
                        func.extract('epoch', TranslationRun.completed_at - TranslationRun.started_at) * 1000
                    )
                )
                .filter(
                    TranslationRun.job_id == job_id,
                    TranslationRun.started_at.isnot(None),
                    TranslationRun.completed_at.isnot(None),
                )
                .scalar()
            )

            # Last run
            last_run = (
                self.db.query(TranslationRun)
                .filter(TranslationRun.job_id == job_id)
                .order_by(TranslationRun.created_at.desc())
                .first()
            )

            # Next scheduled run
            next_schedule = (
                self.db.query(TranslationSchedule)
                .filter(
                    TranslationSchedule.job_id == job_id,
                    TranslationSchedule.is_active == True,
                )
                .first()
            )

            # Cumulative tokens/cost from MetricSnapshot + events
            cumulative_tokens = 0
            cumulative_cost = 0.0

            # Latest MetricSnapshot
            latest_snapshot = (
                self.db.query(MetricSnapshot)
                .filter(MetricSnapshot.job_id == job_id)
                .order_by(MetricSnapshot.snapshot_date.desc())
                .first()
            )
            if latest_snapshot:
                # MetricSnapshot stores per-snapshot aggregated tokens/cost
                # Here we sum what we stored — in practice MetricSnapshot.covers_events_before
                # indicates cutoff
                pass

            # Live events (<90 days) for token/cost
            cutoff = datetime.now(timezone.utc)
            live_events = (
                self.db.query(TranslationEvent)
                .filter(
                    TranslationEvent.job_id == job_id,
                    TranslationEvent.event_type.in_(["TRANSLATION_PHASE_COMPLETED", "RUN_COMPLETED"]),
                    TranslationEvent.created_at > cutoff,  # events newer than snapshot
                )
                .all()
            )

            return {
                "job_id": job_id,
                "total_runs": total_runs,
                "successful_runs": status_counts.get("COMPLETED", 0),
                "failed_runs": status_counts.get("FAILED", 0),
                "cancelled_runs": status_counts.get("CANCELLED", 0),
                "total_records": int(total_records),
                "successful_records": int(successful_records),
                "failed_records": int(failed_records),
                "skipped_records": int(skipped_records),
                "cumulative_tokens": cumulative_tokens,
                "cumulative_cost": cumulative_cost,
                "avg_duration_ms": int(avg_duration) if avg_duration else None,
                "last_run_at": last_run.created_at.isoformat() if last_run else None,
                "next_scheduled_run": next_schedule.last_run_at.isoformat() if next_schedule and next_schedule.last_run_at else None,
            }
    # #endregion get_job_metrics

    # #region get_all_metrics [C:2] [TYPE Function] [SEMANTICS translate,metrics,all]
    # @BRIEF Get aggregated metrics for all jobs.
    def get_all_metrics(self) -> List[Dict[str, Any]]:
        with belief_scope("TranslationMetrics.get_all_metrics"):
            job_ids = (
                self.db.query(TranslationRun.job_id)
                .distinct()
                .all()
            )
            return [self.get_job_metrics(jid[0]) for jid in job_ids if jid[0]]
    # #endregion get_all_metrics


# #endregion TranslationMetrics
# #endregion TranslationMetrics