本人博客同步發(fā)表，排版更佳
教訓(xùn)：MongoDB在分片后的集合上進(jìn)行db.collection.count()操作時(shí)，出現(xiàn)結(jié)果不準(zhǔn)確的現(xiàn)象，需要采用聚合的方法獲取集合的count結(jié)果
在使用MongoDB-Java客戶端做簡(jiǎn)單的插入操作（10W條）以后，使用Studio 3T查看插入結(jié)果時(shí)，發(fā)現(xiàn)顯示的count結(jié)果與插入的數(shù)據(jù)不一致，偶然會(huì)多出幾條或十幾條，插入操作很簡(jiǎn)單，其中table3被分片

//創(chuàng)建客戶端連接
MongoClient mongoClient = new MongoClient( "xxx.xxx.xxx.xxx" , 20000 );
MongoDatabase database = mongoClient.getDatabase("testdb");
MongoCollection<Document> collection = database.getCollection("table3");

//插入文檔
long start = System.currentTimeMillis();
Random r = new Random();
for (int i=0;i<100000;i++){
    Document doc = new Document("id", i)
            .append("type", "database"+i)
            .append("test","testval"+i)
            .append("name", "name"+i);

    System.out.println("insertOne = " + i);
    collection.insertOne(doc);
}
System.out.println("use time = " + (System.currentTimeMillis()- start));
System.out.println("count = " + collection.count());

發(fā)現(xiàn)問題后，通過(guò)在shell里面查詢count，命令如下

db.table3.count()

使用該命令依然會(huì)出現(xiàn)統(tǒng)計(jì)信息不準(zhǔn)確的現(xiàn)象，通過(guò)谷歌發(fā)現(xiàn)，官方文檔——(https://docs.mongodb.com/manual/reference/method/db.collection.count/)中有解釋這種現(xiàn)象：

• On a sharded cluster, db.collection.count() can result in an inaccurate count if orphaned documentsexist or if a chunk migration is in progress.

• To avoid these situations, on a sharded cluster, use the $group stage of the db.collection.aggregate() method to $sum the documents. For example, the following operation counts the documents in a collection:

db.collection.aggregate(
   [
      { $group: { _id: null, count: { $sum: 1 } } }
   ]
)

官方文檔解釋了這種現(xiàn)象的原因以及解決方法：
不準(zhǔn)確的原因：

• 操作的是分片的集合（前提）；
• shard分片正在做塊遷移，導(dǎo)致有重復(fù)數(shù)據(jù)出現(xiàn)
• 存在孤立文檔（因?yàn)椴徽ｊP(guān)機(jī)、塊遷移失敗等原因?qū)е拢?/li>

解決方法
使用聚合aggregate的方式查詢count數(shù)量，shell命令如下：

db.collection.aggregate(
   [
      { $group: { _id: null, count: { $sum: 1 } } }
   ]
)

java代碼
所以在Java中也可以采用聚合的方式獲取count結(jié)果，使用聚合aggregate的方式可以準(zhǔn)確的獲取sharding后的集合的count結(jié)果。

DBObject groupFields = new BasicDBObject("_id", null);
groupFields.put("count", new BasicDBObject("$sum", 1));
BasicDBObject group = new BasicDBObject("$group", groupFields);

List<BasicDBObject> aggreList = new ArrayList<BasicDBObject>();
aggreList.add(group);
AggregateIterable<Document> output = collection.aggregate(aggreList);

for (Document dbObject : output)
{
      System.out.println("Aggregates count = "+ dbObject);
}

完整代碼

        //創(chuàng)建客戶端連接
        MongoClient mongoClient = new MongoClient( "xxx.xxx.xxx.xxx" , 20000 );
        MongoDatabase database = mongoClient.getDatabase("testdb");
        MongoCollection<Document> collection = database.getCollection("table3");

        //插入文檔
        long start = System.currentTimeMillis();
        Random r = new Random();
        for (int i=0;i<100000;i++){
            Document doc = new Document("id", i)
                    .append("type", "database"+i)
                    .append("test","testval"+i)
                    .append("name", "name"+i);

            System.out.println("insertOne = " + i);
            collection.insertOne(doc);
        }

        System.out.println("use time = " + (System.currentTimeMillis()- start));

        System.out.println("count = " + collection.count());

        DBObject groupFields = new BasicDBObject("_id", null);
        groupFields.put("count", new BasicDBObject("$sum", 1));
        BasicDBObject group = new BasicDBObject("$group", groupFields);

        List<BasicDBObject> aggreList = new ArrayList<BasicDBObject>();
        aggreList.add(group);
        AggregateIterable<Document> output = collection.aggregate(aggreList);

        for (Document dbObject : output)
        {
            System.out.println("Aggregates count = "+ dbObject);
        }