From 74b592fad55624001bc5e006c9a47d67dc80743f Mon Sep 17 00:00:00 2001 From: qian Date: Sun, 13 Sep 2026 22:20:16 +0800 Subject: [PATCH] =?UTF-8?q?feat(db):=E7=89=88=E6=9C=AC=E5=8F=98=E6=9B=B4?= =?UTF-8?q?=E6=96=87=E4=BB=B6?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- migrations/000021_v0_2_20_version.sql | 272 ++++++++++++++++++++++++++ 1 file changed, 272 insertions(+) create mode 100644 migrations/000021_v0_2_20_version.sql diff --git a/migrations/000021_v0_2_20_version.sql b/migrations/000021_v0_2_20_version.sql new file mode 100644 index 0000000..2052fbc --- /dev/null +++ b/migrations/000021_v0_2_20_version.sql @@ -0,0 +1,272 @@ +-- +goose Up +-- 定价存储 JSON 化 + 特殊计费 + 视频任务软删除 + 租户标签备注/钱包累计消费 + 模型厂商（由原 000021~000027 合并）。 +-- 设计文档：docs/模型定价存储JSON化与按秒计费设计.md、docs/租户标签备注与累计消费-执行规划.md +-- +-- mdl_pricing 一步收敛为「每模型一行 + pricing JSONB 唯一真相」： +-- * 旧价格列/多行档位回填进 JSON 后删除，model_id 加唯一约束； +-- * 新增 official_pricing 官方参考定价（管理端折扣计算的基准参照，非计费依据）； +-- * billing_mode 值域新增 special（pricing JSONB 顶层 scheme 键声明计费方案，如 custom:minimax-material）； +-- * 配套 bil_usage_logs.duration_seconds 落按秒计费任务时长，不塞 token 字段凑合。 +-- 另含四笔独立变更：tsk_model_tasks 软删除、tnt_tenants 标签/备注、bil_wallets 累计消费、mdl_models 厂商字段。 +-- ⚠️ 部署顺序强约束：必须先部署识别 special 的新代码，再执行本迁移—— +-- 旧代码读到 special 会按未配价 fail-closed 拒绝该模型的同步路径请求，且旧代码依赖的标量价格列在本迁移中被删除。 + +-- ── 一、tsk_model_tasks：OpenAI Videos DELETE 端点软删除 ── +-- 官方语义为「永久删除已完成/失败视频及其资产」，网关侧保留计费与审计记录， +-- 采用软删除（deleted=true 后客户端查询/下载均按 404 处理）。 +-- +goose StatementBegin +ALTER TABLE tsk_model_tasks + ADD COLUMN deleted BOOLEAN NOT NULL DEFAULT FALSE, + ADD COLUMN deleted_at TIMESTAMPTZ NULL; +-- +goose StatementEnd + +COMMENT ON COLUMN tsk_model_tasks.deleted IS '软删除标记：true 表示已被客户端删除，查询/下载按不存在处理（计费与审计记录保留）'; +COMMENT ON COLUMN tsk_model_tasks.deleted_at IS '软删除时间'; + +-- ── 二、mdl_pricing：收敛为每模型一行，pricing JSONB 唯一真相 ── +-- 保留列：model_id（加 UNIQUE）、billing_mode（可索引筛选）、pricing、official_pricing、 +-- price_note / discount_label / price_change_note。 + +-- 1. 锚点行去重（历史脏数据：个别模型存在多条 min_tokens=0 行）。 +-- 保留每组 id 最大的一条——与旧读取语义一致（GetModelPriceAt 全行扫描循环覆盖，后行胜出）； +-- 去重先于回填，避免被删锚点行作为 min_tokens=0 档位混进聚合出的 tiers 数组、污染首档价格。 +-- +goose StatementBegin +DELETE FROM mdl_pricing p + USING mdl_pricing p2 + WHERE p.model_id = p2.model_id + AND p.min_tokens = 0 AND p2.min_tokens = 0 + AND p.id < p2.id; +-- +goose StatementEnd + +-- 2. 新增 pricing（唯一计费真相）与 official_pricing（官方参考定价）两列 +-- +goose StatementBegin +ALTER TABLE mdl_pricing + ADD COLUMN IF NOT EXISTS pricing JSONB, + ADD COLUMN IF NOT EXISTS official_pricing JSONB; +-- +goose StatementEnd + +COMMENT ON COLUMN mdl_pricing.pricing IS '计费详情（JSONB，按 billing_mode 单模式存储）：token={input_price,output_price,cache_read_price,cache_creation_price}；tiered={tiers:[{min_tokens,max_tokens,input_price,output_price}],cache_read_price,cache_creation_price}；per_request={price}；per_second={unit:"second",prices:{分辨率规格:每秒单价,"*":兜底价}}；顶层可含 time_segments（时段定价）与 scheme（特殊计费方案）'; + +COMMENT ON COLUMN mdl_pricing.official_pricing IS '官方参考定价 JSONB（结构与 pricing 一致 + billing_mode 键，本位币；仅作定价折扣计算基准与几折展示，非计费依据）'; + +-- 3. 存量回填：仅锚点行（min_tokens=0）写 pricing 列 +-- token / per_request：锚点行标量列构造 JSON +-- tiered：全部档位行聚合成 tiers 数组（锚点行承载 cache 价） +-- time_segments：并入 pricing 顶层 +-- +goose StatementBegin +WITH tiers AS ( + SELECT model_id, + jsonb_agg(jsonb_build_object( + 'min_tokens', min_tokens, 'max_tokens', max_tokens, + 'input_price', input_price, 'output_price', output_price + ) ORDER BY min_tokens) AS tier_arr + FROM mdl_pricing GROUP BY model_id +) +UPDATE mdl_pricing p SET pricing = + CASE p.billing_mode + WHEN 'per_request' THEN jsonb_build_object('price', COALESCE(p.per_request_price, 0)) + WHEN 'tiered' THEN jsonb_build_object( + 'tiers', t.tier_arr, + 'cache_read_price', p.cache_read_price, + 'cache_creation_price', p.cache_creation_price) + ELSE jsonb_build_object( + 'input_price', p.input_price, 'output_price', p.output_price, + 'cache_read_price', p.cache_read_price, + 'cache_creation_price', p.cache_creation_price) + END + || CASE WHEN p.time_segments IS NOT NULL + THEN jsonb_build_object('time_segments', p.time_segments) ELSE '{}'::jsonb END +FROM tiers t +WHERE t.model_id = p.model_id AND p.min_tokens = 0 AND p.pricing IS NULL; +-- +goose StatementEnd + +-- 4. 删除阶梯档位行（档位已聚合进锚点行 pricing->'tiers'） +-- +goose StatementBegin +DELETE FROM mdl_pricing WHERE min_tokens <> 0; +-- +goose StatementEnd + +-- 5. 删除旧价格列与时段列（时段已并入 pricing 顶层） +-- +goose StatementBegin +ALTER TABLE mdl_pricing + DROP COLUMN IF EXISTS min_tokens, + DROP COLUMN IF EXISTS max_tokens, + DROP COLUMN IF EXISTS input_price, + DROP COLUMN IF EXISTS output_price, + DROP COLUMN IF EXISTS per_request_price, + DROP COLUMN IF EXISTS cache_read_price, + DROP COLUMN IF EXISTS cache_creation_price, + DROP COLUMN IF EXISTS time_segments; +-- +goose StatementEnd + +-- 6. 每模型一行唯一约束 +-- +goose StatementBegin +ALTER TABLE mdl_pricing ADD CONSTRAINT uk_mdl_pricing_model UNIQUE (model_id); +-- +goose StatementEnd + +-- 7. billing_mode 值域新增 special：pricing 顶层声明了 scheme 键的存量方案模型（如 MiniMax-H3 的 +-- custom:minimax-material 素材计费）从 per_second 改标 special，使用量日志/账单展示「特殊计费」 +-- 而非「按秒计费」。计费金额不受影响（计费分发只看 pricing JSONB 的 scheme 键，billing_mode 是 +-- 展示参考字段）。WHERE 带 billing_mode='per_second' 条件，可重复执行（幂等）。 +UPDATE mdl_pricing + SET billing_mode = 'special' + WHERE billing_mode = 'per_second' + AND pricing->>'scheme' IS NOT NULL + AND pricing->>'scheme' <> ''; + +COMMENT ON COLUMN mdl_pricing.billing_mode IS '计费模式：token=按量；per_request=按次；tiered=阶梯；per_second=按秒（视频）；special=特殊计费（pricing JSONB 顶层 scheme 声明计费方案，如 custom:minimax-material）'; + +-- ── 三、bil_usage_logs：按秒计费任务时长 ── +-- +goose StatementBegin +ALTER TABLE bil_usage_logs ADD COLUMN IF NOT EXISTS duration_seconds INT; +-- +goose StatementEnd + +COMMENT ON COLUMN bil_usage_logs.duration_seconds IS '按秒计费任务的视频时长（秒），来自任务提交时的 spec.duration；非时长类任务为 NULL'; + +-- ── 四、tnt_tenants 标签备注 + bil_wallets 累计消费（方案 A：Redis 权威 + DB 物化副本）── +-- +goose StatementBegin +ALTER TABLE tnt_tenants + ADD COLUMN tags JSONB NOT NULL DEFAULT '[]'::jsonb, + ADD COLUMN remark VARCHAR(1000) NOT NULL DEFAULT ''; +-- +goose StatementEnd + +COMMENT ON COLUMN tnt_tenants.tags IS '租户标签（JSONB 字符串数组，管理后台运营标注，不暴露给租户控制台）'; +COMMENT ON COLUMN tnt_tenants.remark IS '管理员备注（运营备注，不暴露给租户控制台）'; + +-- +goose StatementBegin +ALTER TABLE bil_wallets + ADD COLUMN total_consumed NUMERIC(20,10) NOT NULL DEFAULT 0; +-- +goose StatementEnd + +COMMENT ON COLUMN bil_wallets.total_consumed IS '累计消费总额（本位币；Redis 钱包 hash total_consumed 的物化副本，随结算事件递增/补偿递减）'; + +-- 存量回填：从流水账重放历史消费（consume 流水 amount 为负值），幂等可重跑 +UPDATE bil_wallets w +SET total_consumed = COALESCE(( + SELECT SUM(-t.amount) FROM bil_transactions t + WHERE t.tenant_id = w.tenant_id AND t.type = 'consume' +), 0); + +-- ── 五、mdl_models：研发厂商字段 ── +-- 注意与 chn_channels.type（渠道供应商/托管方）是两个维度：研发厂商指模型的开发公司， +-- 例如 Bedrock 上的 Claude 厂商仍是 anthropic，硅基流动上的 Qwen 厂商仍是 alibaba。 +-- +goose StatementBegin +ALTER TABLE mdl_models + ADD COLUMN vendor VARCHAR(30) NOT NULL DEFAULT ''; +-- +goose StatementEnd + +COMMENT ON COLUMN mdl_models.vendor IS '研发厂商枚举：openai/anthropic/google/xai/mistral/cohere/meta/alibaba/bytedance/deepseek/zhipu/moonshot/minimax/baidu/tencent/xunfei/kuaishou/midjourney/suno，空串=未分类'; + +-- 存量回填：按 model_id 前缀推断厂商（仅回填 vendor 为空的行，幂等可重跑；推断不准可在管理后台修正） +UPDATE mdl_models SET vendor = 'openai' WHERE vendor = '' AND (model_id LIKE 'gpt-%' OR model_id LIKE 'chatgpt-%' OR model_id LIKE 'o1%' OR model_id LIKE 'o3%' OR model_id LIKE 'o4%' OR model_id LIKE 'codex-%' OR model_id LIKE 'sora-%' OR model_id LIKE 'dall-e-%' OR model_id LIKE 'gpt-image%' OR model_id LIKE 'whisper-%' OR model_id LIKE 'tts-%' OR model_id LIKE 'text-embedding-%'); +UPDATE mdl_models SET vendor = 'anthropic' WHERE vendor = '' AND model_id LIKE 'claude-%'; +UPDATE mdl_models SET vendor = 'google' WHERE vendor = '' AND (model_id LIKE 'gemini-%' OR model_id LIKE 'veo-%' OR model_id LIKE 'imagen-%'); +UPDATE mdl_models SET vendor = 'xai' WHERE vendor = '' AND model_id LIKE 'grok-%'; +UPDATE mdl_models SET vendor = 'mistral' WHERE vendor = '' AND (model_id LIKE 'mistral-%' OR model_id LIKE 'codestral-%' OR model_id LIKE 'pixtral-%' OR model_id LIKE 'magistral-%'); +UPDATE mdl_models SET vendor = 'cohere' WHERE vendor = '' AND (model_id LIKE 'command-%' OR model_id LIKE 'embed-%' OR model_id LIKE 'rerank-%'); +UPDATE mdl_models SET vendor = 'meta' WHERE vendor = '' AND model_id LIKE 'llama-%'; +UPDATE mdl_models SET vendor = 'alibaba' WHERE vendor = '' AND (model_id LIKE 'qwen%' OR model_id LIKE 'qwq%' OR model_id LIKE 'qvq%'); +UPDATE mdl_models SET vendor = 'bytedance' WHERE vendor = '' AND (model_id LIKE 'doubao-%' OR model_id LIKE 'seedream-%' OR model_id LIKE 'seedance-%' OR model_id LIKE 'jimeng-%'); +UPDATE mdl_models SET vendor = 'deepseek' WHERE vendor = '' AND model_id LIKE 'deepseek-%'; +UPDATE mdl_models SET vendor = 'zhipu' WHERE vendor = '' AND model_id LIKE 'glm-%'; +UPDATE mdl_models SET vendor = 'moonshot' WHERE vendor = '' AND (model_id LIKE 'kimi-%' OR model_id LIKE 'moonshot-%'); +UPDATE mdl_models SET vendor = 'minimax' WHERE vendor = '' AND (model_id LIKE 'minimax-%' OR model_id LIKE 'abab-%'); +UPDATE mdl_models SET vendor = 'baidu' WHERE vendor = '' AND (model_id LIKE 'ernie-%' OR model_id LIKE 'wenxin-%'); +UPDATE mdl_models SET vendor = 'tencent' WHERE vendor = '' AND model_id LIKE 'hunyuan-%'; +UPDATE mdl_models SET vendor = 'xunfei' WHERE vendor = '' AND model_id LIKE 'spark-%'; +UPDATE mdl_models SET vendor = 'kuaishou' WHERE vendor = '' AND model_id LIKE 'kling-%'; +UPDATE mdl_models SET vendor = 'midjourney' WHERE vendor = '' AND (model_id LIKE 'mj_%' OR model_id LIKE 'midjourney-%'); +UPDATE mdl_models SET vendor = 'suno' WHERE vendor = '' AND model_id LIKE 'suno-%'; + +-- +goose Down +-- 逆序回滚：厂商 → 租户/钱包 → 用量时长 → 定价旧结构重建 → 视频软删除列。 + +-- ── 五、mdl_models：移除厂商字段 ── +-- +goose StatementBegin +ALTER TABLE mdl_models + DROP COLUMN IF EXISTS vendor; +-- +goose StatementEnd + +-- ── 四、tnt_tenants / bil_wallets ── +-- +goose StatementBegin +ALTER TABLE tnt_tenants + DROP COLUMN IF EXISTS tags, + DROP COLUMN IF EXISTS remark; +ALTER TABLE bil_wallets + DROP COLUMN IF EXISTS total_consumed; +-- +goose StatementEnd + +-- ── 三、bil_usage_logs ── +-- +goose StatementBegin +ALTER TABLE bil_usage_logs DROP COLUMN IF EXISTS duration_seconds; +-- +goose StatementEnd + +-- ── 二、mdl_pricing：从 pricing JSONB 无损反向重建旧结构 ── +-- JSON 自迁移起为唯一被写的真相，覆盖其后的全部改价。 +-- special 反向还原：scheme 模型收回 per_second（旧代码完全兼容形态）； +-- 迁移后通过方案编辑器新保存的模型 scheme 非空，同样被覆盖回 per_second。 +UPDATE mdl_pricing + SET billing_mode = 'per_second' + WHERE billing_mode = 'special' + AND pricing->>'scheme' IS NOT NULL + AND pricing->>'scheme' <> ''; + +COMMENT ON COLUMN mdl_pricing.billing_mode IS '计费模式：token=按量；per_request=按次；tiered=阶梯；per_second=按秒（视频）'; + +-- +goose StatementBegin +ALTER TABLE mdl_pricing DROP CONSTRAINT IF EXISTS uk_mdl_pricing_model; +ALTER TABLE mdl_pricing + ADD COLUMN IF NOT EXISTS min_tokens BIGINT DEFAULT 0 NOT NULL, + ADD COLUMN IF NOT EXISTS max_tokens BIGINT, + ADD COLUMN IF NOT EXISTS input_price NUMERIC(20,10) DEFAULT 0 NOT NULL, + ADD COLUMN IF NOT EXISTS output_price NUMERIC(20,10) DEFAULT 0 NOT NULL, + ADD COLUMN IF NOT EXISTS per_request_price NUMERIC(20,10), + ADD COLUMN IF NOT EXISTS cache_read_price NUMERIC(20,10) DEFAULT 0 NOT NULL, + ADD COLUMN IF NOT EXISTS cache_creation_price NUMERIC(20,10) DEFAULT 0 NOT NULL, + ADD COLUMN IF NOT EXISTS time_segments JSONB; +-- +goose StatementEnd + +-- 锚点行标量列从 JSON 回写（tiered 取首档价，与旧行为一致） +-- +goose StatementBegin +UPDATE mdl_pricing SET + input_price = COALESCE((pricing->>'input_price')::numeric, (pricing->'tiers'->0->>'input_price')::numeric, 0), + output_price = COALESCE((pricing->>'output_price')::numeric, (pricing->'tiers'->0->>'output_price')::numeric, 0), + per_request_price = (pricing->>'price')::numeric, + cache_read_price = COALESCE((pricing->>'cache_read_price')::numeric, 0), + cache_creation_price = COALESCE((pricing->>'cache_creation_price')::numeric, 0), + time_segments = pricing->'time_segments' +WHERE pricing IS NOT NULL; +-- +goose StatementEnd + +-- tiered 档位行从 JSON tiers 数组展开重建（跳过首档——首档即锚点行 min_tokens=0） +-- +goose StatementBegin +INSERT INTO mdl_pricing (model_id, billing_mode, min_tokens, max_tokens, input_price, output_price, created_at, updated_at) +SELECT p.model_id, + p.billing_mode, + (t.tier->>'min_tokens')::bigint, + NULLIF(t.tier->>'max_tokens', '')::bigint, + COALESCE((t.tier->>'input_price')::numeric, 0), + COALESCE((t.tier->>'output_price')::numeric, 0), + now(), + now() +FROM mdl_pricing p + CROSS JOIN LATERAL jsonb_array_elements(COALESCE(p.pricing->'tiers', '[]'::jsonb)) + WITH ORDINALITY AS t(tier, ord) +WHERE p.billing_mode = 'tiered' + AND p.pricing IS NOT NULL + AND (t.tier->>'min_tokens')::bigint <> 0 + AND t.ord > 1; +-- +goose StatementEnd + +-- JSON 列最后删除（上方旧结构重建均需读取 pricing） +-- +goose StatementBegin +ALTER TABLE mdl_pricing + DROP COLUMN IF EXISTS pricing, + DROP COLUMN IF EXISTS official_pricing; +-- +goose StatementEnd + +-- ── 一、tsk_model_tasks ── +-- +goose StatementBegin +ALTER TABLE tsk_model_tasks + DROP COLUMN IF EXISTS deleted, + DROP COLUMN IF EXISTS deleted_at; +-- +goose StatementEnd