java - mongo java 驱动程序比 shell 慢(15 倍)

Question

尝试仅转储 _id 列。使用 mongo shell，大约 2 分钟即可完成:

time mongoexport -h localhost -d db1 -c collec1 -f _id -o u.text --csv
connected to: localhost
exported 68675826 records
real    2m20.970s

使用java大约需要30分钟:java -cp mongo-test- assembly-0.1.jar com.poshmark.Test

class Test {
public static void main(String[] args) {


    MongoClient mongoClient = new MongoClient("localhost");
    MongoDatabase database = mongoClient.getDatabase("db1");
    MongoCollection<Document> collection = database.getCollection("collec1");

    MongoCursor<Document> iterator = collection.find().projection(new Document("_id", 1)).iterator();
    while (iterator.hasNext()) {
        System.out.println(iterator.next().toString());
    }
}

}

盒子上的 CPU 使用率很低，没有看到任何网络延迟问题，因为两个测试都在同一个盒子上运行

更新:

使用 Files.newBufferedWriter 而不是 System.out.println 但最终获得相同的性能。

查看 db.currentOp()，让我认为 mongo 正在访问磁盘，因为它有太多 numYields

{
"inprog" : [
    {
        "desc" : "conn8636699",
        "threadId" : "0x79a70c0",
        "connectionId" : 8636699,
        "opid" : 1625079940,
        "active" : true,
        "secs_running" : 12,
        "microsecs_running" : NumberLong(12008522),
        "op" : "getmore",
        "ns" : "users.users",
        "query" : {
            "_id" : {
                "$exists" : true
            }
        },
        "client" : "10.1.166.219:60324",
        "numYields" : 10848,
        "locks" : {

        },
        "waitingForLock" : false,
        "lockStats" : {
            "Global" : {
                "acquireCount" : {
                    "r" : NumberLong(21696)
                },
                "acquireWaitCount" : {
                    "r" : NumberLong(26)
                },
                "timeAcquiringMicros" : {
                    "r" : NumberLong(28783)
                }
            },
            "MMAPV1Journal" : {
                "acquireCount" : {
                    "r" : NumberLong(10848)
                },
                "acquireWaitCount" : {
                    "r" : NumberLong(5)
                },
                "timeAcquiringMicros" : {
                    "r" : NumberLong(40870)
                }
            },
            "Database" : {
                "acquireCount" : {
                    "r" : NumberLong(10848)
                }
            },
            "Collection" : {
                "acquireCount" : {
                    "R" : NumberLong(10848)
                }
            }
        }
    }
  ]
}

Answer 1

问题出在 STDOUT 中。

Printing to stdout is not inherently slow. It is the terminal you work with that is slow.

https://stackoverflow.com/a/3860319/3710490

The disk appears to be faster, because it is highly buffered.

The terminal, on the other hand, does little or no buffering: each individual print / write(line) waits for the full write (i.e. display to output device) to complete.

https://stackoverflow.com/a/3857543/3710490

我已经用足够的相似数据集重现了您的用例。

mongoexport 到 FILE

$ time "C:\Program Files\MongoDB\Server\4.2\bin\mongoexport.exe" -h localhost -d test -c collec1 -f _id -o u.text --csv
2020-03-28T13:03:01.550+0100    csv flag is deprecated; please use --type=csv instead
2020-03-28T13:03:02.433+0100    connected to: mongodb://localhost/
2020-03-28T13:03:03.479+0100    [........................]  test.collec1  0/21028330  (0.0%)
2020-03-28T13:05:02.934+0100    [########################]  test.collec1  21028330/21028330  (100.0%)
2020-03-28T13:05:02.934+0100    exported 21028330 records

real    2m1,936s
user    0m0,000s
sys     0m0,000s

mongoexport 到 STDOUT

$ time "C:\Program Files\MongoDB\Server\4.2\bin\mongoexport.exe" -h localhost -d test -c collec1 -f _id --csv
2020-03-28T14:43:16.479+0100    connected to: mongodb://localhost/
2020-03-28T14:43:16.545+0100    [........................]  test.collec1  0/21028330  (0.0%)
2020-03-28T14:53:02.361+0100    [########################]  test.collec1  21028330/21028330  (100.0%)
2020-03-28T14:53:02.361+0100    exported 21028330 records

real    9m45,962s
user    0m0,015s
sys     0m0,000s

JAVA 到文件

$ time "C:\Program Files\Java\jdk1.8.0_211\bin\java.exe" -jar mongo-test-assembly-0.1.jar FILE
Wasted time for [FILE] -  271,57 sec

real    4m32,174s
user    0m0,015s
sys     0m0,000s

JAVA 到 STDOUT 到文件

$ time "C:\Program Files\Java\jdk1.8.0_211\bin\java.exe" -jar mongo-test-assembly-0.1.jar SYSOUT > u.text

real    6m50,962s
user    0m0,015s
sys     0m0,000s

JAVA 到 STDOUT

$ time "C:\Program Files\Java\jdk1.8.0_211\bin\java.exe" -jar mongo-test-assembly-0.1.jar SYSOUT > u.text

Wasted time for [SYSOUT] -  709,33 sec

real    11m51,276s
user    0m0,000s
sys     0m0,015s

Java代码

long init = System.currentTimeMillis();
try (MongoClient mongoClient = new MongoClient("localhost");
    BufferedWriter writer = Files.newBufferedWriter(Files.createTempFile("benchmarking", ".tmp"))) {

    MongoDatabase database = mongoClient.getDatabase("test");
    MongoCollection<Document> collection = database.getCollection("collec1");

    MongoCursor<Document> iterator = collection.find().projection(new Document("_id", 1)).iterator();
    while (iterator.hasNext()) {
        if ("SYSOUT".equals(args[0])) {
            System.out.println(iterator.next().get("_id"));

        } else {
            writer.write(iterator.next().get("_id") + "\n");
        }
    }
} catch (Exception e) {
    e.printStackTrace();
}
long end = System.currentTimeMillis();
System.out.println(String.format("Wasted time for [%s] -  %.2f sec", args[0], (end - init) / 1_000.));