积分冻结释放

video-gen-api/app/api/v1/generation.py

@@ -15,7 +15,6 @@ from app.dependencies import get_db, get_current_user
 from app.models.user import User
 from app.models.project import Project
 from app.models.generation_record import GenerationRecord
-from app.models.system_config import SystemConfig
 from app.schemas.generation import (
     OptimizeParams,
     GenerationRecordOut,
@@ -27,7 +26,6 @@ from app.services.generation.pipeline.db_lock_service import (
     DatabaseRowLockBusy,
     execute_with_lock_timeout,
 )
-from app.services.credits import deduct_credits, add_credits, calc_text_credits
 from app.services.llm import optimize_prompt
 from app.services.video_url import validate_and_get_record_id, get_video_stream_url
 from app.services.private_portrait.reference_resolver import batch_resolve_private_portrait_reference_display_urls, resolve_private_portrait_reference_display_urls
@@ -45,14 +43,27 @@ from app.enums.generation_status import (
     GenerationType,
 )
 from app.enums.common import LogEventStatusEnum
+from app.enums.credit_record import (
+    CreditRecordBillingScene,
+    CreditRecordChargeKind,
+    CreditRecordSourceModule,
+)
+from app.enums.llm_billing import LlmBillingConfigKey
+from app.services.llm_billing import (
+    LlmBillingContext,
+    log_provider_failure,
+    log_provider_start,
+    log_provider_success,
+    release_on_failure,
+    settle_success,
+    start_hold,
+)
 from app.enums.generation_record import (
     GenerationRecordConfigSourceEnum,
     GenerationRecordEventTypeEnum,
 )
 from app.services.generation.billing_service import (
-    CHARGE_TEXT_PROMPT,
     OWNER_GENERATION_RECORD,
-    build_credit_biz_key,
     charge_generation_media_for_record,
     get_next_credit_attempt_no,
 )
@@ -76,7 +87,6 @@ from app.services.generation.media_reference_service import (
     calculate_media_reference_usage,
     validate_media_reference_usage_for_engine,
 )
-from app.services.credit_record_meta_service import build_generation_record_prompt_meta
 from app.enums.audio_reference import (
     AUDIO_ALLOWED_EXTENSIONS,
     AUDIO_ALLOWED_MIME_TYPES,
@@ -506,29 +516,26 @@ async def optimize(
     )
 
 
-    hold_credits = 5
-    hold_result = await db.execute(
-        select(SystemConfig).where(SystemConfig.key == "optimize_hold_credits").limit(1)
-    )
-    hold_row = hold_result.scalar_one_or_none()
-    if hold_row and hold_row.value:
-        try:
-            hold_credits = max(0, int(hold_row.value))
-        except (ValueError, TypeError):
-            hold_credits = 5
-
-    hold_scope = req.idempotency_key or generate_id()
-    hold_biz_key = f"optimize_hold:{hold_scope}"
-    hold_refund_biz_key = f"optimize_hold_refund:{hold_scope}"
-    await deduct_credits(
-        db,
-        user_id_snapshot,
-        hold_credits,
-        "AI创作预扣积分",
-        biz_key=hold_biz_key,
+    record_id_value = generate_id()
+    prompt_attempt_no = 1
+    llm_billing_context = LlmBillingContext(
+        user_id=user_id_snapshot,
+        owner_type=OWNER_GENERATION_RECORD,
+        owner_id=record_id_value,
+        attempt_no=prompt_attempt_no,
+        charge_kind=CreditRecordChargeKind.TEXT_PROMPT.value,
+        billing_scene=CreditRecordBillingScene.GENERATION_RECORD_TEXT_PROMPT_OPTIMIZE.value,
+        source_module=CreditRecordSourceModule.GENERATION_RECORD.value,
+        related_id=record_id_value,
+        hold_config_key=LlmBillingConfigKey.HOLD_GENERATION_RECORD_PROMPT.value,
+        description_prefix="AI创作提示词优化",
+        trace_id=f"generation-optimize:{record_id_value}",
+        request_id=req.idempotency_key,
     )
+    await start_hold(db, llm_billing_context)
     await db.commit()
 
+    log_provider_start(llm_billing_context, detail={"gen_type": req.gen_type.value})
     try:
         optimized, token_usage = await optimize_prompt(
             db,
@@ -544,55 +551,63 @@ async def optimize(
             log_module="generation_record",
             log_step="prompt_optimize",
             log_project_id=req.project_id,
+            log_owner_type=OWNER_GENERATION_RECORD,
+            log_owner_id=record_id_value,
+            generation_attempt_no=prompt_attempt_no,
         )
+        log_provider_success(llm_billing_context, usage=token_usage)
     except Exception as exc:
         from app.services.error_codes import extract_error_message
 
         await db.rollback()
-        await add_credits(
-            db,
-            user_id_snapshot,
-            hold_credits,
-            "AI创作预扣积分退还",
-            record_type="refund",
-            biz_key=hold_refund_biz_key,
-            refund_for_biz_key=hold_biz_key,
-        )
+        log_provider_failure(llm_billing_context, error=str(exc))
+        await release_on_failure(db, llm_billing_context, error=str(exc))
         await db.commit()
         raise HTTPException(
             status_code=502,
             detail=f"AI模型调用失败: {extract_error_message(exc, '提示词')}",
         ) from exc
 
-    try:
-        text_credits = await calc_text_credits(
-            db,
-            int(token_usage.get("input_tokens", 0) or 0),
-            int(token_usage.get("output_tokens", 0) or 0),
-        )
-
-        record = GenerationRecord(
-            id=generate_id(),
-            user_id=user_id_snapshot,
-            project_id=req.project_id,
-            original_prompt=req.prompt,
-            optimized_prompt=optimized,
-            gen_type=req.gen_type.value,
-            duration=req.duration if req.gen_type == GenerationType.video else None,
-            aspect_ratio=req.aspect_ratio if req.gen_type == GenerationType.video else None,
-            resolution=req.resolution if req.gen_type == GenerationType.video else None,
-            image_size=req.image_size if req.gen_type == GenerationType.image else None,
-            image_proportion=req.image_proportion if req.gen_type == GenerationType.image else None,
-            image_px=req.image_px if req.gen_type == GenerationType.image else None,
-            status="prompt_optimized",
-            pipeline_stage=None,
-            credits_cost=0,
-            text_credits_cost=round(text_credits, 2),
-            text_tokens_used=int(token_usage.get("total_tokens", 0) or 0),
-            media_references=json.dumps(req.references, ensure_ascii=False) if req.references else None,
-            include_media_references=bool(req.include_media_references),
-            idempotency_key=req.idempotency_key,
+    async def _persist_optimized_result() -> str:
+        existing_result = await db.execute(
+            select(GenerationRecord)
+            .where(GenerationRecord.id == record_id_value)
+            .with_for_update()
+            .limit(1)
         )
+        record = existing_result.scalar_one_or_none()
+        if record is None:
+            record = GenerationRecord(
+                id=record_id_value,
+                user_id=user_id_snapshot,
+                project_id=req.project_id,
+                original_prompt=req.prompt,
+                optimized_prompt=optimized,
+                gen_type=req.gen_type.value,
+                duration=req.duration if req.gen_type == GenerationType.video else None,
+                aspect_ratio=req.aspect_ratio if req.gen_type == GenerationType.video else None,
+                resolution=req.resolution if req.gen_type == GenerationType.video else None,
+                image_size=req.image_size if req.gen_type == GenerationType.image else None,
+                image_proportion=req.image_proportion if req.gen_type == GenerationType.image else None,
+                image_px=req.image_px if req.gen_type == GenerationType.image else None,
+                status="prompt_optimized",
+                pipeline_stage=None,
+                credits_cost=0,
+                text_credits_cost=0,
+                text_tokens_used=int(token_usage.get("total_tokens", 0) or 0),
+                media_references=json.dumps(req.references, ensure_ascii=False) if req.references else None,
+                include_media_references=bool(req.include_media_references),
+                idempotency_key=req.idempotency_key,
+            )
+            db.add(record)
+        else:
+            # commit 结果不确定或本地持久化重试时，复用同一主键和同一账务 attempt。
+            record.optimized_prompt = optimized
+            record.status = "prompt_optimized"
+            record.pipeline_stage = None
+            record.error_message = None
+            record.text_credits_cost = 0
+            record.text_tokens_used = int(token_usage.get("total_tokens", 0) or 0)
 
         if req.gen_type == GenerationType.video:
             from app.services.video_upscale.snapshot_service import build_video_upscale_snapshot
@@ -619,61 +634,41 @@ async def optimize(
             engine=engine_snapshot_source,
             source=GenerationRecordConfigSourceEnum.PROMPT_OPTIMIZE,
         )
-        db.add(record)
         await db.flush()
-
-        prompt_attempt_no = 1
-        prompt_biz_key = build_credit_biz_key(
-            owner_type=OWNER_GENERATION_RECORD,
-            owner_id=record.id,
-            attempt_no=prompt_attempt_no,
-            charge_kind=CHARGE_TEXT_PROMPT,
-            action="charge",
-        )
-        prompt_meta = await build_generation_record_prompt_meta(
+        billing = await settle_success(
             db,
-            record_id=record.id,
-            attempt_no=prompt_attempt_no,
-            charge_kind=CHARGE_TEXT_PROMPT,
+            llm_billing_context,
             usage=token_usage,
+            description=f"提示词优化 - {project_name_snapshot}",
         )
-        # Release the hold and charge the exact prompt usage in one transaction.
-        await add_credits(
-            db,
-            user_id_snapshot,
-            hold_credits,
-            f"AI创作预扣积分退还 - {project_name_snapshot}",
-            related_id=record.id,
-            record_type="refund",
-            biz_key=hold_refund_biz_key,
-            refund_for_biz_key=hold_biz_key,
-        )
-        await deduct_credits(
-            db,
-            user_id_snapshot,
-            text_credits,
-            f"提示词优化 - {project_name_snapshot}",
-            related_id=record.id,
-            biz_key=prompt_biz_key,
-            record_meta=prompt_meta,
+        charge_item = next(
+            (item for item in billing.items if item.biz_key == llm_billing_context.charge_biz_key),
+            None,
         )
+        if charge_item:
+            record.text_credits_cost = round(charge_item.amount, 2)
         record_id_snapshot = str(record.id)
         await db.commit()
-    except Exception:
+        return record_id_snapshot
+
+    try:
+        record_id_snapshot = await _persist_optimized_result()
+    except Exception as first_exc:
         await db.rollback()
-        # Any local pricing/snapshot/persistence failure after the provider call
-        # must release the committed hold. The refund key is idempotent.
-        await add_credits(
-            db,
-            user_id_snapshot,
-            hold_credits,
-            "AI创作预扣积分退还",
-            record_type="refund",
-            biz_key=hold_refund_biz_key,
-            refund_for_biz_key=hold_biz_key,
+        logger.exception(
+            "prompt optimize local persistence/settlement failed after provider success; retry once: record_id=%s",
+            record_id_value,
         )
-        await db.commit()
-        raise
+        try:
+            record_id_snapshot = await _persist_optimized_result()
+        except Exception:
+            await db.rollback()
+            logger.exception(
+                "prompt optimize idempotent persistence retry failed; active HOLD retained for repair: record_id=%s",
+                record_id_value,
+            )
+            raise first_exc
+
 
     refreshed = await db.execute(
         select(GenerationRecord, Project.name)