CouchDB 中奇怪的 Map Reduce 行为。重新减少？

Question

我有一个 couchdb 的 mapreduce 问题（两个函数如下所示）：当我使用 grouplevel = 2（精确）运行它时，我得到准确的输出：

{"rows":[
 {"key":["2011-01-11","staff-1"],"value":{"total":895.72,"count":2,"services":6,"services_ignored":6,"services_liked":0,"services_disliked":0,"services_disliked_avg":0,"Revise":{"total":275.72,"count":1},"Review":{"total":620,"count":1}}},
 {"key":["2011-01-11","staff-2"],"value":{"total":8461.689999999999,"count":2,"services":41,"services_ignored":37,"services_liked":4,"services_disliked":0,"services_disliked_avg":0,"Revise":{"total":4432.4,"count":1},"Review":{"total":4029.29,"count":1}}},
 {"key":["2011-01-11","staff-3"],"value":{"total":2100.72,"count":1,"services":10,"services_ignored":4,"services_liked":3,"services_disliked":3,"services_disliked_avg":2.3333333333333335,"Revise":{"total":2100.72,"count":1}}},

但是，更改为 grouplevel=1 以便所有不同人员键的值都应按日期分组不再提供准确的输出（请注意总数是正确的，但所有其他人都是错误的）：

{"rows":[
  {"key":["2011-01-11"],"value":{"total":11458.130000000001,"count":2,"services":0,"services_ignored":0,"services_liked":0,"services_disliked":0,"services_disliked_avg":0,"None":{"total":11458.130000000001,"count":2}}},

我唯一的理论是这与rereduce有关，我还没有学过。我应该探索那个选项还是我在这里错过了其他东西？

这是地图功能：

function(doc) {
if(doc.doc_type == 'Feedback') {
    emit([doc.date.split('T')[0], doc.staff_id], doc);
}
}

这就是 Reduce：

function(keys, vals) {
// sum all key points by status: total, count, services (liked, rejected, ignored)
var ret = {
    'total':0,
    'count':0, 
    'services': 0,
    'services_ignored': 0,
    'services_liked': 0,
    'services_disliked': 0,
    'services_disliked_avg': 0,
};

var total_disliked_score = 0;

// handle status
function handle_status(doc) {
    if(!doc.status || doc.status == '' || doc.status == undefined) {
        status = 'None';
    } else if (doc.status == 'Declined') {
        status = 'Rejected';
    } else {
        status = doc.status;
    }
    if(!ret[status]) ret[status] = {'total':0, 'count':0};
    ret[status]['total'] += doc.total;  
    ret[status]['count'] += 1;
};

// handle likes / dislikes
function handle_services(services) {
    ret.services += services.length;
    for(var a in services) {
        if (services[a].user_likes == 10) {
            ret.services_liked += 1;
        } else if (services[a].user_likes >= 1) {
            ret.services_disliked += 1;
            total_disliked_score += services[a].user_likes;
            if (total_disliked_score >= ret.services_disliked) {
                ret.services_disliked_avg = total_disliked_score / ret.services_disliked;
            }
        } else {
            ret.services_ignored += 1;
        }
    }
}

// loop thru docs 
for(var i in vals) {
    // increment the total $
    ret.total += vals[i].total;
    ret.count += 1;

    // update totals and sums for the status of this route
    handle_status(vals[i]);

    // do the likes / dislikes stats
    if(vals[i].groups) {
        for(var ii in vals[i].groups) {
            if(vals[i].groups[ii].services) {
                handle_services(vals[i].groups[ii].services); 
            }
        }
    }

    // handle deleted services
    if(vals[i].hidden_services) {
        if (vals[i].hidden_services) {
            handle_services(vals[i].hidden_services);
        }
    }
}

return ret;
}

Answer 1

作为参考，在 var ret = { ... } 下面添加以下代码来处理 rereduce 工作！

function rereduce_status(row, ret, stat) 
{
    if(row[stat]) {
        if(!ret[stat]) ret[stat] = {'total':0, 'count':0};
        ret[stat]['total'] += row[stat].total;
        ret[stat]['count'] += row[stat].count;
    }   
    return ret;
}

if(rereduce) {
    for (var i in vals) {
        ret.total += vals[i].total;
        ret.count += vals[i].count;
        ret.services += vals[i].services;
        ret.services_ignored += vals[i].services_ignored;
        ret.services_liked += vals[i].services_liked;
        ret.services_disliked += vals[i].services_disliked;
        ret.services_disliked_score += vals[i].services_disliked_score;
        if (ret.services_disliked_score >= ret.services_disliked) {
            ret.services_disliked_avg = ret.services_disliked_score / ret.services_disliked;
        }
        ret = rereduce_status(vals[i], ret, 'None');
        ret = rereduce_status(vals[i], ret, 'Review');
        ret = rereduce_status(vals[i], ret, 'Revise');
        ret = rereduce_status(vals[i], ret, 'Rejected');
        ret = rereduce_status(vals[i], ret, 'Booked');
    }

    return ret;
}

Answer 2

这是一个典型的错误。请记住，CouchDB 缩减发生在几个步骤中，其中一些步骤将接收作为输入其他缩减步骤的结果。但是，您的代码似乎假定 vals[i] 将是 { "groups": _ , "hidden_services": _ , _ } 形式的对象，表示单个文档。当发生 rereduce 时，此代码将失败，因为 vals[i] 将采用 { "count" : _ , "services" : _ , _ } 的形式，表示前一个缩减步骤的结果。

因此，例如，通过使用ret.count += 1 进行计数，您计算的是中间缩减结果的数量，而不是文档的数量。

一种解决方案是编写两个版本的reduce 代码，一个用于处理原始reduce，另一个用于处理rereduce 步骤。您可以通过查看第三个参数来确定给定调用是初始调用还是重新减少调用（如果为初始则为假，如果为重新减少则为真）。

另一个解决方案是让 map 函数发出一个与 reduce 函数返回的形式相同的 { "count" : _ , "services" : _ , _ } 的预处理值，并让 reduce 函数只是将这些值的成员加在一起。