ERP报表系统设计与数据仓库

报表是ERP里最被低估的模块。

很多企业上ERP，只关注业务流程，报表随便做做。上线后发现：要的数据查不到，查到的数据不准，做报表比做业务还累。

这篇文章从架构层面讲清楚ERP报表系统怎么设计。

一、报表系统的核心问题

实时性 vs 性能

业务数据实时变化，但报表查询不能卡死业务系统。

直接在业务库上跑报表，并发高的时候会把数据库拖垮。

灵活性 vs 标准化

每个部门要的报表不一样。销售要业绩排名，仓库要库存周转，财务要利润分析。

如果每个报表都单独开发，工作量巨大。如果不开发，用户就觉得系统不好用。

准确性 vs 复杂性

业务数据经过多环节流转，报表计算涉及多层关联。关联越多，出错的可能性越大。

二、报表架构设计

三层架构

复制代码

业务数据库 → 数据仓库 → 报表服务层 → 前端展示

业务数据库只负责业务操作，不跑报表。

数据仓库负责数据汇总和计算。

报表服务层负责查询和分发。

这样业务和报表互不影响。

业务数据库设计

报表相关的字段要在业务表里预留。

复制代码

CREATE TABLE sa_order (
    id BIGINT PRIMARY KEY AUTO_INCREMENT,
    order_no VARCHAR(30) NOT NULL,
    customer_id BIGINT NOT NULL,
    order_date DATE NOT NULL,
    total_amount DECIMAL(18,2) NOT NULL,
    
    -- 报表预留字段
    region_code VARCHAR(10),          -- 区域编码，用于区域销售分析
    product_category VARCHAR(30),     -- 产品大类，用于品类分析
    sales_channel VARCHAR(20),        -- 销售渠道，用于渠道分析
    department_id BIGINT,             -- 部门，用于部门业绩
    
    -- 时间维度
    create_time DATETIME NOT NULL,
    update_time DATETIME,
    
    INDEX idx_date (order_date),
    INDEX idx_customer (customer_id),
    INDEX idx_region (region_code)
);

这些字段在业务操作时可能用不到，但对报表至关重要。

数据仓库（DW）设计

数据仓库和业务数据库的核心区别：面向查询优化，不是面向写入优化。

3.1 维度建模

使用星型模型，事实表+维度表。

复制代码

-- 销售事实表
CREATE TABLE dw_sales_fact (
    id BIGINT PRIMARY KEY AUTO_INCREMENT,
    
    -- 维度外键
    date_key INT NOT NULL,            -- 关联时间维度
    customer_key INT NOT NULL,        -- 关联客户维度
    product_key INT NOT NULL,         -- 关联产品维度
    region_key INT,                   -- 关联区域维度
    salesperson_key INT,              -- 关联销售员维度
    channel_key INT,                  -- 关联渠道维度
    
    -- 度量值
    order_count INT DEFAULT 0,        -- 订单数
    order_amount DECIMAL(18,2) DEFAULT 0,  -- 订单金额
    cost_amount DECIMAL(18,2) DEFAULT 0,   -- 成本金额
    profit_amount DECIMAL(18,2) DEFAULT 0, -- 利润金额
    
    INDEX idx_date (date_key),
    INDEX idx_customer (customer_key),
    INDEX idx_product (product_key)
);

-- 时间维度表
CREATE TABLE dim_date (
    date_key INT PRIMARY KEY,
    full_date DATE NOT NULL,
    year INT,
    quarter INT,
    month INT,
    week_of_year INT,
    day_of_week INT,
    is_weekend TINYINT,
    is_holiday TINYINT,
    fiscal_year INT,
    fiscal_quarter INT,
    fiscal_month INT
);

-- 产品维度表
CREATE TABLE dim_product (
    product_key INT PRIMARY KEY,
    product_id BIGINT NOT NULL,
    product_code VARCHAR(30),
    product_name VARCHAR(100),
    category_level1 VARCHAR(30),      -- 一级分类
    category_level2 VARCHAR(30),      -- 二级分类
    category_level3 VARCHAR(30),      -- 三级分类
    brand VARCHAR(50),
    unit VARCHAR(10),
    is_active TINYINT DEFAULT 1,
    created_date DATE,
    modified_date DATE
);

-- 客户维度表
CREATE TABLE dim_customer (
    customer_key INT PRIMARY KEY,
    customer_id BIGINT NOT NULL,
    customer_code VARCHAR(30),
    customer_name VARCHAR(100),
    region VARCHAR(30),
    city VARCHAR(30),
    industry VARCHAR(50),
    customer_level VARCHAR(20),       -- 客户等级
    credit_limit DECIMAL(18,2),
    is_active TINYINT DEFAULT 1
);

-- 销售员维度表
CREATE TABLE dim_salesperson (
    salesperson_key INT PRIMARY KEY,
    user_id BIGINT NOT NULL,
    user_name VARCHAR(50),
    department VARCHAR(50),
    position VARCHAR(30),
    region VARCHAR(30),
    entry_date DATE,
    is_active TINYINT DEFAULT 1
);

3.2 ETL过程

数据从业务库抽取到数据仓库，需要经过清洗和转换。

复制代码

class SalesETL:
    """销售数据ETL"""
    
    def extract(self, start_date, end_date):
        """从业务库抽取数据"""
        sql = """
            SELECT 
                o.id AS order_id,
                o.order_no,
                o.customer_id,
                o.order_date,
                o.total_amount AS order_amount,
                od.product_id,
                od.quantity,
                od.unit_price,
                od.cost_price,
                od.quantity * od.cost_price AS cost_amount,
                o.total_amount - SUM(od.quantity * od.cost_price) OVER 
                    (PARTITION BY o.id) AS profit_amount,
                o.region_code,
                o.salesperson_id,
                o.sales_channel
            FROM sa_order o
            JOIN sa_order_detail od ON o.id = od.order_id
            WHERE o.order_date BETWEEN %s AND %s
              AND o.status = 'CONFIRMED'
        """
        return self.source_db.query(sql, (start_date, end_date))
    
    def transform(self, raw_data):
        """数据转换"""
        result = []
        for row in raw_data:
            # 日期维度
            date_key = self.get_date_key(row['order_date'])
            # 客户维度
            customer_key = self.get_customer_key(row['customer_id'])
            # 产品维度
            product_key = self.get_product_key(row['product_id'])
            # 销售员维度
            salesperson_key = self.get_salesperson_key(row['salesperson_id'])
            # 区域维度
            region_key = self.get_region_key(row['region_code'])
            # 渠道维度
            channel_key = self.get_channel_key(row['sales_channel'])
            
            result.append({
                'date_key': date_key,
                'customer_key': customer_key,
                'product_key': product_key,
                'region_key': region_key,
                'salesperson_key': salesperson_key,
                'channel_key': channel_key,
                'order_amount': row['order_amount'],
                'cost_amount': row['cost_amount'],
                'profit_amount': row['profit_amount'],
                'order_count': 1
            })
        return result
    
    def load(self, transformed_data):
        """加载到数据仓库"""
        # 先删除当天的数据（支持重跑）
        # 再插入新数据
        self.dw_db.batch_insert('dw_sales_fact', transformed_data)

3.3 调度策略

ETL不能实时跑，太耗资源。按以下频率调度：

复制代码

etl_schedule:
  # 日汇总表：每天凌晨1点跑前一天的数据
  dw_sales_daily:
    schedule: "0 1 * * *"
    time_range: "yesterday"
    
  # 月汇总表：每月1号凌晨3点跑上个月的数据
  dw_sales_monthly:
    schedule: "0 3 1 * *"
    time_range: "last_month"
    
  # 实时快照表：每15分钟跑一次（只增量）
  dw_sales_realtime:
    schedule: "*/15 * * * *"
    time_range: "last_15_minutes"

三、汇总表设计

直接在事实表上聚合查询，数据量大的时候很慢。需要预先建汇总表。

日汇总

复制代码

CREATE TABLE dw_sales_daily_summary (
    date_key INT NOT NULL,
    region_key INT,
    product_category VARCHAR(30),
    
    order_count INT,
    order_amount DECIMAL(18,2),
    cost_amount DECIMAL(18,2),
    profit_amount DECIMAL(18,2),
    
    avg_order_amount DECIMAL(18,2),
    profit_rate DECIMAL(5,4),
    
    PRIMARY KEY (date_key, region_key, product_category)
);

月汇总

复制代码

CREATE TABLE dw_sales_monthly_summary (
    year_month INT NOT NULL,          -- 202605
    region_key INT,
    product_category VARCHAR(30),
    
    order_count INT,
    order_amount DECIMAL(18,2),
    cost_amount DECIMAL(18,2),
    profit_amount DECIMAL(18,2),
    
    -- 环比增长
    mom_order_count_rate DECIMAL(5,4),
    mom_order_amount_rate DECIMAL(5,4),
    
    PRIMARY KEY (year_month, region_key, product_category)
);

环比增长率怎么算：

复制代码

INSERT INTO dw_sales_monthly_summary
SELECT 
    DATE_FORMAT(s.order_date, '%%Y%%m') AS year_month,
    r.region_key,
    p.category_level1 AS product_category,
    COUNT(DISTINCT s.id) AS order_count,
    SUM(s.total_amount) AS order_amount,
    -- ... 其他字段
    (SUM(s.total_amount) - prev.order_amount) / prev.order_amount AS mom_rate
FROM dw_sales_fact f
-- 关联各维度表
JOIN fact_orders s ON ...
JOIN dim_region r ON ...
JOIN dim_product p ON ...
LEFT JOIN dw_sales_monthly_summary prev 
    ON DATE_FORMAT(s.order_date, '%%Y%%m') - INTERVAL 1 MONTH = prev.year_month
GROUP BY DATE_FORMAT(s.order_date, '%%Y%%m'), r.region_key, p.category_level1;

四、报表服务层

报表定义

复制代码

CREATE TABLE rpt_report_config (
    id BIGINT PRIMARY KEY AUTO_INCREMENT,
    report_code VARCHAR(30) NOT NULL UNIQUE,
    report_name VARCHAR(100) NOT NULL,
    
    -- 数据源
    data_source VARCHAR(50),          -- DW/REALTIME/CUSTOM
    sql_template TEXT NOT NULL,        -- SQL模板，支持参数
    
    -- 维度配置
    dimensions JSON,                   -- 可选的分组维度
    measures JSON,                     -- 度量字段
    
    -- 权限
    role_ids JSON,                     -- 允许查看的角色
    
    -- 缓存
    cache_ttl INT DEFAULT 300,         -- 缓存时间（秒）
    
    -- 状态
    is_active TINYINT DEFAULT 1
);

动态查询

报表不是写死的SQL，是动态拼装的。

用户选了哪些维度、哪些度量，系统自动拼SQL。

复制代码

public class ReportQueryBuilder
{
    private readonly ReportConfig _config;
    private readonly List<string> _selectedDimensions = new();
    private readonly List<string> _selectedMeasures = new();
    private readonly List<string> _whereConditions = new();
    
    public ReportQueryBuilder(ReportConfig config)
    {
        _config = config;
    }
    
    public ReportQueryBuilder AddDimension(string dimension)
    {
        // 校验维度是否在报表配置中
        if (_config.Dimensions.Contains(dimension))
            _selectedDimensions.Add(dimension);
        return this;
    }
    
    public ReportQueryBuilder AddMeasure(string measure)
    {
        if (_config.Measures.Contains(measure))
            _selectedMeasures.Add(measure);
        return this;
    }
    
    public ReportQueryBuilder AddFilter(string field, string op, object value)
    {
        // 防注入：只允许预定义字段
        if (!_config.AllFields.Contains(field))
            throw new SecurityException($"Field {field} not allowed");
        
        _whereConditions.Add($"{field} {op} @{field}");
        return this;
    }
    
    public (string sql, object param) Build()
    {
        var select = new StringBuilder("SELECT ");
        
        // 维度
        select.Append(string.Join(", ", _selectedDimensions));
        
        // 度量
        foreach (var measure in _selectedMeasures)
        {
            var agg = _config.GetAggregation(measure);
            select.Append($", {agg} AS {measure}");
        }
        
        var sql = select.ToString();
        sql += $" FROM {_config.DataSource}";
        
        // WHERE
        if (_whereConditions.Any())
            sql += " WHERE " + string.Join(" AND ", _whereConditions);
        
        // GROUP BY
        sql += " GROUP BY " + string.Join(", ", _selectedDimensions);
        
        // ORDER BY
        sql += " ORDER BY " + string.Join(", ", _selectedDimensions);
        
        return (sql, null);
    }
}

报表缓存

不是每次查询都查数据库，命中缓存就直接返回。

复制代码

public class ReportCacheService
{
    private readonly IDatabase _db;
    private readonly ICache _cache;
    
    public ReportData GetReport(string reportCode, ReportQuery query)
    {
        // 生成缓存Key
        var cacheKey = $"rpt:{reportCode}:{query.ToHash()}";
        
        // 查缓存
        var cached = _cache.Get<ReportData>(cacheKey);
        if (cached != null)
            return cached;
        
        // 查数据库
        var config = _db.GetReportConfig(reportCode);
        var (sql, param) = new ReportQueryBuilder(config)
            .AddDimensions(query.Dimensions)
            .AddMeasures(query.Measures)
            .AddFilters(query.Filters)
            .Build();
        
        var data = _db.Query<ReportData>(sql, param);
        
        // 写缓存
        _cache.Set(cacheKey, data, TimeSpan.FromSeconds(config.CacheTtl));
        
        return data;
    }
    
    // 报表数据变更时，清除相关缓存
    public void InvalidateCache(string reportCode)
    {
        var pattern = $"rpt:{reportCode}:*";
        _cache.DeleteByPattern(pattern);
    }
}

五、常见报表设计

销售日报

复制代码

SELECT 
    d.full_date AS 日期,
    r.region_name AS 区域,
    COUNT(DISTINCT f.order_id) AS 订单数,
    SUM(f.order_amount) AS 销售额,
    SUM(f.profit_amount) AS 利润,
    ROUND(SUM(f.profit_amount) / SUM(f.order_amount) * 100, 2) AS 利润率
FROM dw_sales_fact f
JOIN dim_date d ON f.date_key = d.date_key
JOIN dim_region r ON f.region_key = r.region_key
WHERE d.full_date = ?
GROUP BY d.full_date, r.region_key;

产品销量排名

复制代码

SELECT 
    p.product_name AS 产品名称,
    p.category_level1 AS 产品分类,
    SUM(f.order_count) AS 销售数量,
    SUM(f.order_amount) AS 销售金额,
    SUM(f.profit_amount) AS 利润,
    ROUND(SUM(f.profit_amount) / SUM(f.order_amount) * 100, 2) AS 利润率
FROM dw_sales_fact f
JOIN dim_product p ON f.product_key = p.product_key
WHERE f.date_key BETWEEN ? AND ?
GROUP BY p.product_key
ORDER BY SUM(f.order_amount) DESC
LIMIT 20;

客户ABC分析

复制代码

SELECT 
    c.customer_name AS 客户名称,
    SUM(f.order_amount) AS 累计销售额,
    SUM(f.order_amount) * 100.0 / (
        SELECT SUM(order_amount) FROM dw_sales_fact 
        WHERE date_key BETWEEN ? AND ?
    ) AS 占比,
    @running_total := @running_total + SUM(f.order_amount) AS 累计金额,
    CASE 
        WHEN @running_total <= (SELECT SUM(order_amount) * 0.7 FROM dw_sales_fact WHERE date_key BETWEEN ? AND ?)
            THEN 'A'
        WHEN @running_total <= (SELECT SUM(order_amount) * 0.9 FROM dw_sales_fact WHERE date_key BETWEEN ? AND ?)
            THEN 'B'
        ELSE 'C'
    END AS 客户等级
FROM dw_sales_fact f
JOIN dim_customer c ON f.customer_key = c.customer_key
CROSS JOIN (SELECT @running_total := 0) AS init
WHERE f.date_key BETWEEN ? AND ?
GROUP BY c.customer_key
ORDER BY SUM(f.order_amount) DESC;

库存周转分析

复制代码

SELECT 
    p.product_name AS 产品名称,
    p.category_level1 AS 分类,
    i.current_qty AS 当前库存,
    ROUND(
        COALESCE(s.total_out_qty, 0) / GREATEST(i.current_qty, 1), 2
    ) AS 月周转率,
    ROUND(
        GREATEST(i.current_qty, 0) / 
        GREATEST(COALESCE(s.avg_daily_out_qty, 1), 0.01), 0
    ) AS 可售天数
FROM dw_inventory_current i
JOIN dim_product p ON i.product_key = p.product_key
LEFT JOIN dw_inventory_summary_monthly s 
    ON i.product_key = s.product_key
    AND s.year_month = DATE_FORMAT(CURDATE(), '%%Y%%m')
ORDER BY 可售天数 ASC;

六、报表导出

Excel导出

大数据量报表不能一次全部加载到内存。用流式写入。

复制代码

public async Task ExportToExcel(string reportCode, ReportQuery query, Stream output)
{
    using var package = new ExcelPackage();
    var sheet = package.Workbook.Worksheets.Add("报表");
    
    // 写表头
    var headers = query.SelectedDimensions.Concat(query.SelectedMeasures).ToList();
    for (int i = 0; i < headers.Count; i++)
        sheet.Cells[1, i + 1].Value = headers[i];
    
    // 流式查询，逐行写入
    int row = 2;
    await foreach (var record in _db.StreamQuery(reportCode, query))
    {
        for (int i = 0; i < headers.Count; i++)
            sheet.Cells[row, i + 1].Value = record[headers[i]];
        row++;
    }
    
    await package.SaveAsAsync(output);
}

定时推送

有些报表需要定时发邮件给相关人员。

复制代码

report_schedule:
  - name: "每日销售日报"
    report_code: "sales_daily"
    schedule: "0 8 * * *"            # 每天早上8点
    recipients: ["sales_manager@company.com", "gm@company.com"]
    format: "excel"
    
  - name: "每月库存周转分析"
    report_code: "inventory_turnover"
    schedule: "0 9 1 * *"            # 每月1号早上9点
    recipients: ["warehouse_manager@company.com"]
    format: "pdf"

七、性能优化要点

分区表

事实表按月分区，查询时自动裁剪。

列式存储

OLAP场景用列式存储效率更高。ClickHouse、Doris都是好选择。

物化视图

复杂的聚合查询，用物化视图预计算。

复制代码

CREATE MATERIALIZED VIEW mv_sales_monthly_region
REFRESH COMPLETE ON DEMAND
AS
SELECT 
    DATE_FORMAT(order_date, '%%Y%%m') AS year_month,
    region_code,
    COUNT(*) AS order_count,
    SUM(total_amount) AS total_amount
FROM sa_order
GROUP BY DATE_FORMAT(order_date, '%%Y%%m'), region_code;

慢查询监控

复制代码

-- 记录报表查询时间
INSERT INTO rpt_query_log (report_code, query_time_ms, row_count, user_id)
VALUES (?, ?, ?, ?);

-- 找出慢查询
SELECT report_code, AVG(query_time_ms) AS avg_time, COUNT(*) AS query_count
FROM rpt_query_log
WHERE query_time > DATE_SUB(NOW(), INTERVAL 7 DAY)
GROUP BY report_code
HAVING avg_time > 3000
ORDER BY avg_time DESC;

报表系统做好了，ERP的价值才能真正体现。用户看到的不是一堆单据，而是能支撑决策的数据。

------云策数链