Question

考虑下面存储在CouchDB中的示例文档

 {
"_id":....,
"rev":....,
"type":"orders",
"Period":"2013-01",
"Region":"East",
"Category":"Stationary",
"Product":"Pen",
"Rate":1,
"Qty":10,
"Amount":10
}

{
"_id":....,
"rev":....,
"type":"orders",
"Period":"2013-02",
"Region":"South",
"Category":"Food",
"Product":"Biscuit",
"Rate":7,
"Qty":5,
"Amount":35
}

考虑遵循SQL查询

SELECT Period, Region,Category, Product, Min(Rate),Max(Rate),Count(Rate), Sum(Qty),Sum(Amount)
FROM Sales
GROUP BY Period,Region,Category, Product;

是否可以在couchdb中创建与上述SQL查询等效的map / reduce视图，并生成类似

的输出

[
    {
        "Period":"2013-01",
        "Region":"East",
        "Category":"Stationary",
        "Product":"Pen",
        "MinRate":1,
        "MaxRate":2,
        "OrdersCount":20,
        "TotQty":1000,
        "Amount":1750
    },
    {
    ... 
    }

]

Answer 1

我将提出一个非常简单的解决方案，要求在“select”子句中聚合每个变量一个视图。虽然可以在单个视图中聚合所有变量，但reduce函数会复杂得多。

设计文档如下所示：

{
    "_id": "_design/ddoc",
    "_rev": "...",
    "language": "javascript",
    "views": {
        "rates": {
            "map": "function(doc) {\n  emit([doc.Period, doc.Region, doc.Category, doc.Product], doc.Rate);\n}",
            "reduce": "_stats"
        },
        "qty": {
            "map": "function(doc) {\n  emit([doc.Period, doc.Region, doc.Category, doc.Product], doc.Qty);\n}",
            "reduce": "_stats"
        }
    }
}

现在，您可以查询<couchdb>/<database>/_design/ddoc/_view/rates?group_level=4以获取有关“费率”变量的统计信息。结果应如下所示：

{"rows":[
{"key":["2013-01","East","Stationary","Pen"],"value":{"sum":4,"count":3,"min":1,"max":2,"sumsqr":6}},
{"key":["2013-01","North","Stationary","Pen"],"value":{"sum":1,"count":1,"min":1,"max":1,"sumsqr":1}},
{"key":["2013-01","South","Stationary","Pen"],"value":{"sum":0.5,"count":1,"min":0.5,"max":0.5,"sumsqr":0.25}},
{"key":["2013-02","South","Food","Biscuit"],"value":{"sum":7,"count":1,"min":7,"max":7,"sumsqr":49}}
]}

对于“Qty”变量，查询将为<couchdb>/<database>/_design/ddoc/_view/qty?group_level=4。

使用group_level属性，您可以控制要执行聚合的级别。例如，使用group_level=2查询将汇总到“Period”和“Region”。

Answer 2

在前面，我相信@benedolph的答案是最佳实践和最佳案例。理想情况下，每个reduce应返回1个标量值，以使代码尽可能简单。

但是，您必须发出多个查询才能检索问题所描述的完整结果集。如果您没有并行运行查询的选项，或者保持查询数量下降非常重要，则可以一次完成所有查询。

您的地图功能将非常简单：

function (doc) {
    emit([ doc.Period, doc.Region, doc.Category, doc.Product ], doc);
}

reduce函数是冗长的地方：

function (key, values, rereduce) {
    // helper function to sum all the values of a specified field in an array of objects
    function sumField(arr, field) {
        return arr.reduce(function (prev, cur) {
            return prev + cur[field];
        }, 0);
    }

    // helper function to create an array of just a single property from an array of objects
    // (this function came from underscore.js, at least it's name and concept)
    function pluck(arr, field) {
        return arr.map(function (item) {
            return item[field];
        });
    }

    // rereduce made this more challenging, and I could not thoroughly test this right now
    // see the CouchDB wiki for more information
    if (rereduce) {
        // a rereduce handles transitionary values
        // (so the "values" below are the results of previous reduce functions, not the map function)
        return {
            OrdersCount: sumField(values, "OrdersCount"),
            MinRate: Math.min.apply(Math, pluck(values, "MinRate")),
            MaxRate: Math.max.apply(Math, pluck(values, "MaxRate")),
            TotQty: sumField(values, "TotQty"),
            Amount: sumField(values, "Amount")
        };
    } else {
        var rates = pluck(values, "Rate");

        // This takes a group of documents and gives you the stats you were asking for
        return {
            OrdersCount: values.length,
            MinRate: Math.min.apply(Math, rates),
            MaxRate: Math.max.apply(Math, rates),
            TotQty: sumField(values, "Qty"),
            Amount: sumField(values, "Amount")
        };
    }
}

我根本无法测试此代码的“rereduce”分支，你必须在你的最后完成。（但这应该有效）有关reduce vs rereduce的信息，请参阅the wiki。

我在顶部添加的辅助函数实际上使代码整体更短更容易阅读，它们在很大程度上受我Underscore.js经验的影响。但是，您不能在reduce函数中包含CommonJS模块，因此必须手动编写。

同样，最好的情况是让每个聚合字段获得它自己的map / reduce索引，但是如果这不是你的选项，那么上面的代码应该能够得到你在这个问题中所描述的内容。

CouchDB - Map Reduce类似于SQL Group by

2 个答案: