align hierarchy outputs with DWS naming

This commit is contained in:
2026-08-20 19:01:46 +08:00
parent faddf38c9b
commit 33e60d02e3
6 changed files with 380 additions and 487 deletions
+38 -22
View File
@@ -2,23 +2,40 @@
Databricks 数仓 SQL 重构仓库。
## 数仓分层
仓库采用三层模型,数据依赖保持单向流动:
```text
DWD -> DWS -> DM
```
- `DWD`:清洗后的原子明细和基础主数据。
- `DWS`:引用 DWD 构建的公共维度、宽表和可复用事实表。
- `DM`:面向具体分析主题、指标和报表的数据集市。
## 首批 review 范围
本批次重构原 `CHPA/01` 中两个层级维表脚本:
本批次重构原 `CHPA/01` 中两个层级维表脚本:
- `sql/chpa/01_dwd/01_dwd_ims_atc_hierarchy.sql`
- `sql/chpa/01_dwd/02_dwd_ims_nfc_hierarchy.sql`
- `sql/chpa/02_dws/01_dws_ext_td_ims_atc_hierarchy.sql`
- `sql/chpa/02_dws/02_dws_ext_td_ims_nfc_hierarchy.sql`
工作区 `CHPA/` 下的原脚本保持不变,便于逐项对照。
两个脚本读取 DWD,并写入新的 DWS 维度表:
## 兼容性决策
- `dws.dws_ext_td_ims_atc_hierarchy`
- `dws.dws_ext_td_ims_nfc_hierarchy`
首批代码继续写入现有物理表:
工作区 `CHPA/` 下的原脚本保持不变。旧表 `dwd.dwd_ims_atc_hierarchy``dwd.dwd_ims_nfc_hierarchy` 仅作为本批新旧结果校验基线。
- `dwd.dwd_ims_atc_hierarchy`
- `dwd.dwd_ims_nfc_hierarchy`
## 表命名
这样不会立即影响 pack property 和 DWS 等现有下游任务。推荐的新表名及迁移规则记录在 `docs/sql_refactoring_standard.md`;是否执行物理改名,待本批 review 后决定。
- DWS 维度表:`dws_ext_td_<业务实体>`
- DWS 事实表:`dws_ext_tf_<业务实体>`
- DM 维度表:`dm_ext_td_<业务实体>`
- DM 事实表:`dm_ext_tf_<业务实体>`
完整规范及首批表名映射见 `docs/sql_refactoring_standard.md`
## 目录结构
@@ -29,26 +46,25 @@ RE/
| `-- sql_refactoring_standard.md
|-- sql/
| `-- chpa/
| `-- 01_dwd/
| |-- 01_dwd_ims_atc_hierarchy.sql
| `-- 02_dwd_ims_nfc_hierarchy.sql
| `-- 02_dws/
| |-- 01_dws_ext_td_ims_atc_hierarchy.sql
| `-- 02_dws_ext_td_ims_nfc_hierarchy.sql
`-- validation/
`-- chpa/
`-- 01_dwd/
`-- 02_dws/
`-- validate_hierarchy_refactor.sql
```
## 执行顺序
1. 在覆盖目标表前运行 `validation/chpa/01_dwd/validate_hierarchy_refactor.sql`
2. 核对两个 `DESCRIBE TABLE` 结果;兼容性检查必须全部返回 `passed = true`
3. review 重复路径和未匹配层级指标,确认它们符合现有业务口径
4. 通过后,按文件名前缀顺序执行 `sql/chpa/01_dwd/` 中的重构脚本
两个重构脚本都要求现有 DWD 源表、目标表已经创建。
1. 确认两个 DWS 目标表已按脚本中的显式字段契约创建
2. 按文件名前缀顺序执行 `sql/chpa/02_dws/` 中的脚本
3. 运行 `validation/chpa/02_dws/validate_hierarchy_refactor.sql`,直接比较新 DWS 输出与旧 DWD 基线
4. 兼容性检查必须全部返回 `passed = true`;同时 review 重复路径和未匹配层级指标
5. 下游迁移完成前保留旧 DWD 表;下游不得继续新增对旧表的依赖。
## 本次 review 重点
1. 表命名模型是否适用于整个数仓
2. 物理表改名采用兼容视图还是一次性联动下游
3. 替换调度任务前,在 Databricks 中对比新旧结果的行数、业务键重复数和空键数
1. `DWD -> DWS -> DM` 的职责边界是否符合现有调度设计
2. `dws_ext_td_ims_*` 中是否需要保留来源系统 `ims`
3. DWS 目标表的 catalog、字段类型、表属性和权限是否需要统一 DDL 模板
+84 -63
View File
@@ -2,16 +2,77 @@
## 1. 范围和原则
本规范适用于 CHPA Databricks 数仓脚本。第一阶段目标是在不改变业务结果的前提下提高可维护性。
本规范适用于 CHPA Databricks 数仓脚本。第一阶段目标是在不改变业务口径的前提下提高可维护性。
- 先保持业务口径,再做性能优化。
- 每个脚本只负责一个主要目标表。
- 文件头声明源表、目标表、数据粒度、写入方式和依赖
- 文件头声明源表、目标表、数据粒度、写入方式和迁移关系
- 持久化写入必须显式列出目标列和查询列,禁止使用 `SELECT *`
- 逻辑重构与生产表改名分开实施
- 为行数、唯一性、非空约束和未匹配记录建立验证查询或任务检查。
- 为行数、唯一性、非空约束和未匹配记录建立验证检查
## 2. 文件和目录命名
## 2. 数仓分层
数仓固定为三层,依赖方向为:
```text
DWD -> DWS -> DM
```
### DWD
- 保存清洗、标准化后的原子明细和基础主数据。
- 保持源数据可追溯性和稳定粒度。
- 不承载面向报表的层级拉宽、跨主题指标聚合。
### DWS
- 引用 DWD 构建公共维度、层级宽表和可复用事实表。
- 统一编码、公共口径和跨明细关联结果。
- 本批 ATC/NFC 层级宽维表属于 DWS,不再写回 DWD。
### DM
- 引用 DWS 构建具体业务主题、指标和报表数据集。
- 允许面向使用场景组织字段,但不得反向成为 DWD/DWS 的依赖。
读取关系保持清晰:DWS 读取 DWD,DM 读取 DWS。DWD 不读取 DWS/DMDWS 不读取 DMDWS/DM 的派生结果不得写回 DWD,DM 的派生结果不得写回 DWS。
## 3. 表命名
表名由“层级 + ext + 对象类型 + 业务实体”组成:
```text
[<catalog>.]<schema>.<layer>_ext_<object_type>_<business_entity>
```
固定前缀如下:
| 层级 | 对象 | 表名前缀 |
| --- | --- | --- |
| DWS | 维度/主数据 | `dws_ext_td_` |
| DWS | 事实数据 | `dws_ext_tf_` |
| DM | 维度/主数据 | `dm_ext_td_` |
| DM | 事实数据 | `dm_ext_tf_` |
命名规则:
- `<schema>``<layer>` 保持一致,例如 `dws.dws_ext_td_xxx`
- `td` 表示维度或主数据,`tf` 表示事实数据。
- `<business_entity>` 使用小写 snake_case。
- 需要区分来源系统时,将来源放在业务实体开头,例如 `ims_atc_hierarchy`
- `atc``nfc``ims` 等已形成业务共识的缩写可以保留。
首批及后续建议映射:
| 旧表 | 新表 | 类型 |
| --- | --- | --- |
| `dwd.dwd_ims_atc_hierarchy` | `dws.dws_ext_td_ims_atc_hierarchy` | DWS 维度 |
| `dwd.dwd_ims_nfc_hierarchy` | `dws.dws_ext_td_ims_nfc_hierarchy` | DWS 维度 |
| `dwd.dwd_ims_td_manufacturer_corp` | `dws.dws_ext_td_ims_manufacturer_corporation` | DWS 维度,后续处理 |
| `dwd.dwd_ims_td_pack_property` | `dws.dws_ext_td_ims_pack_property` | DWS 维度,后续处理 |
| `dwd.dwd_gnd_pharbers_prov_fact` | `dws.dws_ext_tf_pharbers_province_sales` | DWS 事实,需先确认粒度 |
## 4. 文件和目录命名
目录结构统一为:
@@ -21,63 +82,31 @@ sql/<业务域>/<阶段>/<序号>_<目标表>.sql
规则:
- 目录和文件名使用小写 snake_case。
- 路径和文件名不使用空格
- 使用两位序号明确 notebook/job 执行顺序。
- 一个脚本只写入一个主要持久化目标。`01 dwd_update.sql` 这种多目标脚本,应按目标表和职责拆分。
- 目录和文件名使用小写 snake_case,不使用空格
- 目录阶段必须和输出层一致,例如 DWS 脚本放在 `02_dws`
- 使用两位序号明确 notebook/job 执行顺序。
- 一个脚本只写入一个主要持久化目标。
- 多目标脚本按目标表和职责拆分。
示例:
```text
sql/chpa/01_dwd/01_dwd_ims_atc_hierarchy.sql
sql/chpa/02_dws/01_dws_ext_td_ims_atc_hierarchy.sql
```
## 3. 物理表命名
新表使用以下格式:
```text
[<catalog>.]<分层_schema>.<来源系统>_<对象类型>_<业务实体>[_<限定词>]
```
未启用 Unity Catalog 或依赖默认 catalog 时可以省略 catalog。分层已经由 schema 表达,表名不再重复 `dwd``dws``dm`
- 来源系统:如 `ims``gnd``pharbers`
- 对象类型:`td` 表示维度/主数据,`tf` 表示事实数据。
- 业务实体:使用稳定的业务名词,如 `atc_hierarchy``pack_property`
- 限定词:按需表达粒度或变体,如 `monthly``province`
`01` 组表名建议:
| 现有物理表名 | 推荐规范表名 | 首批处理方式 |
| --- | --- | --- |
| `dwd.dwd_ims_atc_hierarchy` | `dwd.ims_td_atc_hierarchy` | 保留现名 |
| `dwd.dwd_ims_nfc_hierarchy` | `dwd.ims_td_nfc_hierarchy` | 保留现名 |
| `dwd.dwd_ims_td_manufacturer_corp` | `dwd.ims_td_manufacturer_corporation` | 后续 review |
| `dwd.dwd_ims_td_pack_property` | `dwd.ims_td_pack_property` | 后续 review |
| `dwd.dwd_gnd_pharbers_prov_fact` | `dwd.pharbers_tf_province_sales` | 先确认业务粒度 |
推荐迁移顺序:
1. 创建规范名称的新表,并用兼容视图暴露旧表名。
2. 新旧表并行运行,对比数据结果。
3. 在一次受控发布中修改全部下游引用。
4. 所有消费者迁移后再移除旧名称兼容视图。
## 4. 字段命名
## 5. 字段命名
- 新字段使用小写 snake_case。
- 标识使用 `_id`,业务编码使用 `_code`,名称或描述使用 `_name``_description`
- 日期时间后缀按真实类型使用 `_date``_timestamp``_at`
- `atc``nfc``prod``pack` 等已形成业务共识或受输出契约约束的缩写可以保留
- 纯逻辑重构中不直接修改持久化输出字段名。
- 纯逻辑重构中不直接修改已有输出字段契约
## 5. SQL 结构和格式
## 6. SQL 结构和格式
脚本统一按以下顺序组织:
1. Databricks notebook 标记和脚本契约头。
2. 带显式目标列的 `INSERT OVERWRITE`
2. 带显式目标列的写入语句
3. 源数据标准化 CTE。
4. 业务规则 CTE。
5. 按目标表字段顺序显式编写最终 `SELECT`
@@ -86,27 +115,21 @@ sql/chpa/01_dwd/01_dwd_ims_atc_hierarchy.sql
- SQL 关键字使用大写。
- CTE 和别名使用小写 snake_case。
- 使用四个空格缩进。
- `SELECT` 每行一个字段
- 每个 JOIN 条件单独成行。
- 使用 `atc_level_1` 等有业务意义的别名,维护代码中不使用 `t1``a` 等无语义别名。
- 使用四个空格缩进`SELECT` 每行一个字段
- 使用有业务意义的别名,不使用 `t1``a` 等无语义别名
- 同一作用域出现多个关系时,所有字段都带关系限定符。
## 6. 写入和数据质量规则
## 7. 写入和数据质量
- 只有可确定性重跑的全量任务可以使用 `INSERT OVERWRITE`
- 不在一个脚本中混合无关的 `UPDATE` 和目标表构建逻辑。
- 编码补零等标准化逻辑集中到一个明确的处理阶段,避免在多个任务中重复表达式
- 编码补零等标准化逻辑集中到一个明确阶段
- 每个脚本必须声明预期目标粒度。
- 最少验证以下指标:
- 新结果与旧结果的总行数;
- 声明业务键的重复数;
- 必填键的空值数;
- 层级关联新增的未匹配记录数。
- 最少验证新旧行数、双向差集、业务键重复、必填键空值和层级未匹配数。
## 7. 首批兼容说明
## 8. 首批兼容说明
ATC 和 NFC 脚本原样保留以下父级编码规则:
ATC 和 NFC 脚本保留旧逻辑中的父级编码规则:
- ATC 2 级关联 1 级:取前 1 位。
- ATC 3 级关联 2 级:取前 3 位。
@@ -114,6 +137,4 @@ ATC 和 NFC 脚本原样保留以下父级编码规则:
- NFC 2 级关联 1 级:取前 1 位。
- NFC 3 级关联 2 级:取前 2 位。
任何口径简化都应先用生产样本验证这些规则
首批上线前运行 `validation/chpa/01_dwd/validate_hierarchy_refactor.sql`。新旧双向差集必须为 0,行数必须一致;重复路径和未匹配层级作为业务 review 指标记录,不在未确认阈值前自动判定失败。
两个 DWS 脚本执行后运行 `validation/chpa/02_dws/validate_hierarchy_refactor.sql`。新 DWS 输出与旧 DWD 基线的行数必须一致,双向差集必须为 0;重复路径和未匹配层级作为业务 review 指标记录
@@ -2,14 +2,15 @@
-- =============================================================================
-- Purpose : Build the flattened IMS ATC level 1-4 hierarchy.
-- Source : dwd.dwd_ims_td_therapeutic_class
-- Target : dwd.dwd_ims_atc_hierarchy
-- Target : dws.dws_ext_td_ims_atc_hierarchy
-- Grain : One row per hierarchy path rooted at ATC1_CODE; lower levels may be null.
-- Write mode : Full refresh (INSERT OVERWRITE).
-- Downstream : dwd.dwd_ims_td_pack_property, dws.dws_ims_td_atc_cn
-- Replaces : dwd.dwd_ims_atc_hierarchy (legacy output).
-- Consumers : Pack-property and ATC-CN jobs must migrate to the DWS target.
-- Notes : Parent-code matching is preserved from the legacy script.
-- =============================================================================
INSERT OVERWRITE TABLE dwd.dwd_ims_atc_hierarchy (
INSERT OVERWRITE TABLE dws.dws_ext_td_ims_atc_hierarchy (
ATC1_ID,
ATC1_CODE,
ATC1_DES,
@@ -2,14 +2,15 @@
-- =============================================================================
-- Purpose : Build the flattened IMS NFC level 1-3 hierarchy.
-- Source : dwd.dwd_ims_td_new_form_class
-- Target : dwd.dwd_ims_nfc_hierarchy
-- Target : dws.dws_ext_td_ims_nfc_hierarchy
-- Grain : One row per hierarchy path rooted at NFC1_CODE; lower levels may be null.
-- Write mode : Full refresh (INSERT OVERWRITE).
-- Downstream : dwd.dwd_ims_td_pack_property, dws.dws_ims_td_nfc_cn
-- Replaces : dwd.dwd_ims_nfc_hierarchy (legacy output).
-- Consumers : Pack-property and NFC-CN jobs must migrate to the DWS target.
-- Notes : Parent-code matching is preserved from the legacy script.
-- =============================================================================
INSERT OVERWRITE TABLE dwd.dwd_ims_nfc_hierarchy (
INSERT OVERWRITE TABLE dws.dws_ext_td_ims_nfc_hierarchy (
NFC1_ID,
NFC1_CODE,
NFC1_DES,
@@ -1,396 +0,0 @@
-- Databricks notebook source
-- MAGIC %md
-- MAGIC # ATC/NFC hierarchy refactor validation
-- MAGIC Run this notebook before executing the refactored overwrite scripts.
-- MAGIC The legacy targets and source tables must represent the same data snapshot.
-- COMMAND ----------
-- Confirm the ATC target contains the expected 12 non-partition output columns.
DESCRIBE TABLE dwd.dwd_ims_atc_hierarchy;
-- COMMAND ----------
-- Confirm the NFC target contains the expected 9 non-partition output columns.
DESCRIBE TABLE dwd.dwd_ims_nfc_hierarchy;
-- COMMAND ----------
CREATE OR REPLACE TEMP VIEW validation_atc_hierarchy_candidate AS
WITH therapeutic_class AS (
SELECT
therapeutic_id,
therapeutic_code,
therapeutic_name,
therapeutic_level
FROM dwd.dwd_ims_td_therapeutic_class
),
atc_level_1 AS (
SELECT
therapeutic_id AS atc1_id,
therapeutic_code AS atc1_code,
therapeutic_name AS atc1_des
FROM therapeutic_class
WHERE therapeutic_level = '1'
),
atc_level_2 AS (
SELECT
therapeutic_id AS atc2_id,
therapeutic_code AS atc2_code,
therapeutic_name AS atc2_des
FROM therapeutic_class
WHERE therapeutic_level = '2'
),
atc_level_3 AS (
SELECT
therapeutic_id AS atc3_id,
therapeutic_code AS atc3_code,
therapeutic_name AS atc3_des
FROM therapeutic_class
WHERE therapeutic_level = '3'
),
atc_level_4 AS (
SELECT
therapeutic_id AS atc4_id,
therapeutic_code AS atc4_code,
therapeutic_name AS atc4_des
FROM therapeutic_class
WHERE therapeutic_level = '4'
)
SELECT
atc_level_1.atc1_id AS ATC1_ID,
atc_level_1.atc1_code AS ATC1_CODE,
atc_level_1.atc1_des AS ATC1_DES,
atc_level_2.atc2_id AS ATC2_ID,
atc_level_2.atc2_code AS ATC2_CODE,
atc_level_2.atc2_des AS ATC2_DES,
atc_level_3.atc3_id AS ATC3_ID,
atc_level_3.atc3_code AS ATC3_CODE,
atc_level_3.atc3_des AS ATC3_DES,
atc_level_4.atc4_id AS ATC4_ID,
atc_level_4.atc4_code AS ATC4_CODE,
atc_level_4.atc4_des AS ATC4_DES
FROM atc_level_1
LEFT JOIN atc_level_2
ON atc_level_1.atc1_code = LEFT(atc_level_2.atc2_code, 1)
LEFT JOIN atc_level_3
ON atc_level_2.atc2_code = LEFT(atc_level_3.atc3_code, 3)
LEFT JOIN atc_level_4
ON atc_level_3.atc3_code = LEFT(atc_level_4.atc4_code, 4)
;
-- COMMAND ----------
CREATE OR REPLACE TEMP VIEW validation_nfc_hierarchy_candidate AS
WITH new_form_class AS (
SELECT
newformclass_id,
newformclass_code,
newformclass_name,
newformclass_level
FROM dwd.dwd_ims_td_new_form_class
),
nfc_level_1 AS (
SELECT
newformclass_id AS nfc1_id,
newformclass_code AS nfc1_code,
newformclass_name AS nfc1_des
FROM new_form_class
WHERE newformclass_level = '1'
),
nfc_level_2 AS (
SELECT
newformclass_id AS nfc2_id,
newformclass_code AS nfc2_code,
newformclass_name AS nfc2_des
FROM new_form_class
WHERE newformclass_level = '2'
),
nfc_level_3 AS (
SELECT
newformclass_id AS nfc3_id,
newformclass_code AS nfc3_code,
newformclass_name AS nfc3_des
FROM new_form_class
WHERE newformclass_level = '3'
)
SELECT
nfc_level_1.nfc1_id AS NFC1_ID,
nfc_level_1.nfc1_code AS NFC1_CODE,
nfc_level_1.nfc1_des AS NFC1_DES,
nfc_level_2.nfc2_id AS NFC2_ID,
nfc_level_2.nfc2_code AS NFC2_CODE,
nfc_level_2.nfc2_des AS NFC2_DES,
nfc_level_3.nfc3_id AS NFC3_ID,
nfc_level_3.nfc3_code AS NFC3_CODE,
nfc_level_3.nfc3_des AS NFC3_DES
FROM nfc_level_1
LEFT JOIN nfc_level_2
ON nfc_level_1.nfc1_code = LEFT(nfc_level_2.nfc2_code, 1)
LEFT JOIN nfc_level_3
ON nfc_level_2.nfc2_code = LEFT(nfc_level_3.nfc3_code, 2)
;
-- COMMAND ----------
-- Every check must return passed = true before replacing the scheduled jobs.
CREATE OR REPLACE TEMP VIEW validation_hierarchy_compatibility_checks AS
WITH compatibility_checks AS (
SELECT
'atc_row_count' AS check_name,
(SELECT COUNT(*) FROM validation_atc_hierarchy_candidate) AS actual_value,
(SELECT COUNT(*) FROM dwd.dwd_ims_atc_hierarchy) AS expected_value
UNION ALL
SELECT
'atc_legacy_minus_candidate' AS check_name,
(
SELECT COUNT(*)
FROM (
SELECT
ATC1_ID,
ATC1_CODE,
ATC1_DES,
ATC2_ID,
ATC2_CODE,
ATC2_DES,
ATC3_ID,
ATC3_CODE,
ATC3_DES,
ATC4_ID,
ATC4_CODE,
ATC4_DES
FROM dwd.dwd_ims_atc_hierarchy
EXCEPT ALL
SELECT
ATC1_ID,
ATC1_CODE,
ATC1_DES,
ATC2_ID,
ATC2_CODE,
ATC2_DES,
ATC3_ID,
ATC3_CODE,
ATC3_DES,
ATC4_ID,
ATC4_CODE,
ATC4_DES
FROM validation_atc_hierarchy_candidate
) AS differences
) AS actual_value,
0 AS expected_value
UNION ALL
SELECT
'atc_candidate_minus_legacy' AS check_name,
(
SELECT COUNT(*)
FROM (
SELECT
ATC1_ID,
ATC1_CODE,
ATC1_DES,
ATC2_ID,
ATC2_CODE,
ATC2_DES,
ATC3_ID,
ATC3_CODE,
ATC3_DES,
ATC4_ID,
ATC4_CODE,
ATC4_DES
FROM validation_atc_hierarchy_candidate
EXCEPT ALL
SELECT
ATC1_ID,
ATC1_CODE,
ATC1_DES,
ATC2_ID,
ATC2_CODE,
ATC2_DES,
ATC3_ID,
ATC3_CODE,
ATC3_DES,
ATC4_ID,
ATC4_CODE,
ATC4_DES
FROM dwd.dwd_ims_atc_hierarchy
) AS differences
) AS actual_value,
0 AS expected_value
UNION ALL
SELECT
'nfc_row_count' AS check_name,
(SELECT COUNT(*) FROM validation_nfc_hierarchy_candidate) AS actual_value,
(SELECT COUNT(*) FROM dwd.dwd_ims_nfc_hierarchy) AS expected_value
UNION ALL
SELECT
'nfc_legacy_minus_candidate' AS check_name,
(
SELECT COUNT(*)
FROM (
SELECT
NFC1_ID,
NFC1_CODE,
NFC1_DES,
NFC2_ID,
NFC2_CODE,
NFC2_DES,
NFC3_ID,
NFC3_CODE,
NFC3_DES
FROM dwd.dwd_ims_nfc_hierarchy
EXCEPT ALL
SELECT
NFC1_ID,
NFC1_CODE,
NFC1_DES,
NFC2_ID,
NFC2_CODE,
NFC2_DES,
NFC3_ID,
NFC3_CODE,
NFC3_DES
FROM validation_nfc_hierarchy_candidate
) AS differences
) AS actual_value,
0 AS expected_value
UNION ALL
SELECT
'nfc_candidate_minus_legacy' AS check_name,
(
SELECT COUNT(*)
FROM (
SELECT
NFC1_ID,
NFC1_CODE,
NFC1_DES,
NFC2_ID,
NFC2_CODE,
NFC2_DES,
NFC3_ID,
NFC3_CODE,
NFC3_DES
FROM validation_nfc_hierarchy_candidate
EXCEPT ALL
SELECT
NFC1_ID,
NFC1_CODE,
NFC1_DES,
NFC2_ID,
NFC2_CODE,
NFC2_DES,
NFC3_ID,
NFC3_CODE,
NFC3_DES
FROM dwd.dwd_ims_nfc_hierarchy
) AS differences
) AS actual_value,
0 AS expected_value
UNION ALL
SELECT
'except_all_null_semantics' AS check_name,
(
SELECT COUNT(*)
FROM (
SELECT CAST(NULL AS STRING) AS nullable_value
EXCEPT ALL
SELECT CAST(NULL AS STRING) AS nullable_value
) AS differences
) AS actual_value,
0 AS expected_value
)
SELECT
check_name,
actual_value,
expected_value,
actual_value = expected_value AS passed
FROM compatibility_checks
;
-- COMMAND ----------
SELECT
check_name,
actual_value,
expected_value,
passed
FROM validation_hierarchy_compatibility_checks
ORDER BY check_name;
-- COMMAND ----------
-- Profile duplicates and unmatched levels. These are review metrics, not automatic failures.
WITH atc_duplicate_paths AS (
SELECT COUNT(*) AS path_count
FROM validation_atc_hierarchy_candidate
GROUP BY ATC1_CODE, ATC2_CODE, ATC3_CODE, ATC4_CODE
HAVING COUNT(*) > 1
),
nfc_duplicate_paths AS (
SELECT COUNT(*) AS path_count
FROM validation_nfc_hierarchy_candidate
GROUP BY NFC1_CODE, NFC2_CODE, NFC3_CODE
HAVING COUNT(*) > 1
)
SELECT
'atc_duplicate_path_rows' AS metric_name,
COALESCE(SUM(path_count - 1), 0) AS metric_value
FROM atc_duplicate_paths
UNION ALL
SELECT
'atc_source_level_1_required_key_null' AS metric_name,
COUNT_IF(therapeutic_id IS NULL OR therapeutic_code IS NULL) AS metric_value
FROM dwd.dwd_ims_td_therapeutic_class
WHERE therapeutic_level = '1'
UNION ALL
SELECT
'atc_unmatched_level_2' AS metric_name,
COUNT_IF(ATC2_CODE IS NULL) AS metric_value
FROM validation_atc_hierarchy_candidate
UNION ALL
SELECT
'atc_unmatched_level_3' AS metric_name,
COUNT_IF(ATC3_CODE IS NULL) AS metric_value
FROM validation_atc_hierarchy_candidate
UNION ALL
SELECT
'atc_unmatched_level_4' AS metric_name,
COUNT_IF(ATC4_CODE IS NULL) AS metric_value
FROM validation_atc_hierarchy_candidate
UNION ALL
SELECT
'nfc_duplicate_path_rows' AS metric_name,
COALESCE(SUM(path_count - 1), 0) AS metric_value
FROM nfc_duplicate_paths
UNION ALL
SELECT
'nfc_source_level_1_required_key_null' AS metric_name,
COUNT_IF(newformclass_id IS NULL OR newformclass_code IS NULL) AS metric_value
FROM dwd.dwd_ims_td_new_form_class
WHERE newformclass_level = '1'
UNION ALL
SELECT
'nfc_unmatched_level_2' AS metric_name,
COUNT_IF(NFC2_CODE IS NULL) AS metric_value
FROM validation_nfc_hierarchy_candidate
UNION ALL
SELECT
'nfc_unmatched_level_3' AS metric_name,
COUNT_IF(NFC3_CODE IS NULL) AS metric_value
FROM validation_nfc_hierarchy_candidate
ORDER BY metric_name;
-- COMMAND ----------
-- Abort the notebook when any compatibility check fails.
SELECT IF(
COUNT_IF(NOT passed) = 0,
'All hierarchy compatibility checks passed.',
raise_error(
CONCAT(
'Hierarchy compatibility validation failed: ',
CAST(COUNT_IF(NOT passed) AS STRING),
' check(s) did not pass.'
)
)
) AS validation_result
FROM validation_hierarchy_compatibility_checks;
@@ -0,0 +1,250 @@
-- Databricks notebook source
-- MAGIC %md
-- MAGIC # ATC/NFC DWS hierarchy validation
-- MAGIC Run this notebook after both DWS hierarchy scripts finish.
-- MAGIC The DWS outputs and legacy DWD baselines must represent the same source snapshot.
-- COMMAND ----------
DESCRIBE TABLE dws.dws_ext_td_ims_atc_hierarchy;
-- COMMAND ----------
DESCRIBE TABLE dws.dws_ext_td_ims_nfc_hierarchy;
-- COMMAND ----------
-- Fix the comparison column set and order at the view boundary.
CREATE OR REPLACE TEMP VIEW validation_atc_legacy AS
SELECT
ATC1_ID,
ATC1_CODE,
ATC1_DES,
ATC2_ID,
ATC2_CODE,
ATC2_DES,
ATC3_ID,
ATC3_CODE,
ATC3_DES,
ATC4_ID,
ATC4_CODE,
ATC4_DES
FROM dwd.dwd_ims_atc_hierarchy;
-- COMMAND ----------
CREATE OR REPLACE TEMP VIEW validation_atc_dws AS
SELECT
ATC1_ID,
ATC1_CODE,
ATC1_DES,
ATC2_ID,
ATC2_CODE,
ATC2_DES,
ATC3_ID,
ATC3_CODE,
ATC3_DES,
ATC4_ID,
ATC4_CODE,
ATC4_DES
FROM dws.dws_ext_td_ims_atc_hierarchy;
-- COMMAND ----------
CREATE OR REPLACE TEMP VIEW validation_nfc_legacy AS
SELECT
NFC1_ID,
NFC1_CODE,
NFC1_DES,
NFC2_ID,
NFC2_CODE,
NFC2_DES,
NFC3_ID,
NFC3_CODE,
NFC3_DES
FROM dwd.dwd_ims_nfc_hierarchy;
-- COMMAND ----------
CREATE OR REPLACE TEMP VIEW validation_nfc_dws AS
SELECT
NFC1_ID,
NFC1_CODE,
NFC1_DES,
NFC2_ID,
NFC2_CODE,
NFC2_DES,
NFC3_ID,
NFC3_CODE,
NFC3_DES
FROM dws.dws_ext_td_ims_nfc_hierarchy;
-- COMMAND ----------
CREATE OR REPLACE TEMP VIEW validation_hierarchy_compatibility_checks AS
WITH compatibility_checks AS (
SELECT
'atc_row_count' AS check_name,
(SELECT COUNT(*) FROM validation_atc_dws) AS actual_value,
(SELECT COUNT(*) FROM validation_atc_legacy) AS expected_value
UNION ALL
SELECT
'atc_legacy_minus_dws' AS check_name,
(
SELECT COUNT(*)
FROM (
SELECT * FROM validation_atc_legacy
EXCEPT ALL
SELECT * FROM validation_atc_dws
) AS differences
) AS actual_value,
0 AS expected_value
UNION ALL
SELECT
'atc_dws_minus_legacy' AS check_name,
(
SELECT COUNT(*)
FROM (
SELECT * FROM validation_atc_dws
EXCEPT ALL
SELECT * FROM validation_atc_legacy
) AS differences
) AS actual_value,
0 AS expected_value
UNION ALL
SELECT
'nfc_row_count' AS check_name,
(SELECT COUNT(*) FROM validation_nfc_dws) AS actual_value,
(SELECT COUNT(*) FROM validation_nfc_legacy) AS expected_value
UNION ALL
SELECT
'nfc_legacy_minus_dws' AS check_name,
(
SELECT COUNT(*)
FROM (
SELECT * FROM validation_nfc_legacy
EXCEPT ALL
SELECT * FROM validation_nfc_dws
) AS differences
) AS actual_value,
0 AS expected_value
UNION ALL
SELECT
'nfc_dws_minus_legacy' AS check_name,
(
SELECT COUNT(*)
FROM (
SELECT * FROM validation_nfc_dws
EXCEPT ALL
SELECT * FROM validation_nfc_legacy
) AS differences
) AS actual_value,
0 AS expected_value
UNION ALL
SELECT
'except_all_null_semantics' AS check_name,
(
SELECT COUNT(*)
FROM (
SELECT CAST(NULL AS STRING) AS nullable_value
EXCEPT ALL
SELECT CAST(NULL AS STRING) AS nullable_value
) AS differences
) AS actual_value,
0 AS expected_value
)
SELECT
check_name,
actual_value,
expected_value,
actual_value = expected_value AS passed
FROM compatibility_checks;
-- COMMAND ----------
SELECT
check_name,
actual_value,
expected_value,
passed
FROM validation_hierarchy_compatibility_checks
ORDER BY check_name;
-- COMMAND ----------
-- These quality metrics require business review; they do not fail automatically.
WITH atc_duplicate_paths AS (
SELECT COUNT(*) AS path_count
FROM validation_atc_dws
GROUP BY ATC1_CODE, ATC2_CODE, ATC3_CODE, ATC4_CODE
HAVING COUNT(*) > 1
),
nfc_duplicate_paths AS (
SELECT COUNT(*) AS path_count
FROM validation_nfc_dws
GROUP BY NFC1_CODE, NFC2_CODE, NFC3_CODE
HAVING COUNT(*) > 1
)
SELECT
'atc_duplicate_path_rows' AS metric_name,
COALESCE(SUM(path_count - 1), 0) AS metric_value
FROM atc_duplicate_paths
UNION ALL
SELECT
'atc_source_level_1_required_key_null' AS metric_name,
COUNT_IF(therapeutic_id IS NULL OR therapeutic_code IS NULL) AS metric_value
FROM dwd.dwd_ims_td_therapeutic_class
WHERE therapeutic_level = '1'
UNION ALL
SELECT
'atc_unmatched_level_2' AS metric_name,
COUNT_IF(ATC2_CODE IS NULL) AS metric_value
FROM validation_atc_dws
UNION ALL
SELECT
'atc_unmatched_level_3' AS metric_name,
COUNT_IF(ATC3_CODE IS NULL) AS metric_value
FROM validation_atc_dws
UNION ALL
SELECT
'atc_unmatched_level_4' AS metric_name,
COUNT_IF(ATC4_CODE IS NULL) AS metric_value
FROM validation_atc_dws
UNION ALL
SELECT
'nfc_duplicate_path_rows' AS metric_name,
COALESCE(SUM(path_count - 1), 0) AS metric_value
FROM nfc_duplicate_paths
UNION ALL
SELECT
'nfc_source_level_1_required_key_null' AS metric_name,
COUNT_IF(newformclass_id IS NULL OR newformclass_code IS NULL) AS metric_value
FROM dwd.dwd_ims_td_new_form_class
WHERE newformclass_level = '1'
UNION ALL
SELECT
'nfc_unmatched_level_2' AS metric_name,
COUNT_IF(NFC2_CODE IS NULL) AS metric_value
FROM validation_nfc_dws
UNION ALL
SELECT
'nfc_unmatched_level_3' AS metric_name,
COUNT_IF(NFC3_CODE IS NULL) AS metric_value
FROM validation_nfc_dws
ORDER BY metric_name;
-- COMMAND ----------
-- Block downstream migration when any DWS-to-DWD compatibility check fails.
SELECT IF(
COUNT_IF(NOT passed) = 0,
'All DWS hierarchy compatibility checks passed.',
raise_error(
CONCAT(
'DWS hierarchy compatibility validation failed: ',
CAST(COUNT_IF(NOT passed) AS STRING),
' check(s) did not pass.'
)
)
) AS validation_result
FROM validation_hierarchy_compatibility_checks;