与SQL Server相比，MariaDB JDBC驱动程序无法有效批处理更新

Question

我对应用程序中的插入/更新/删除例程的性能进行了基准测试，我正在将该例程从SQL Server移植到MariaDB。

• 带有i7 2.80GHz CPU+16GB RAM
的本地Win10工作站上的Java 1.8
• jdbc org.mariadb.jdbc:mariadb-java-client:2.2.4
• 10.2.12-MariaDB-在AWS上记录MariaDB服务器

该基准触发50,000次插入、相同的更新和删除。

SQL Server通过Net.SourceForge.jtds JDBC驱动程序在1秒内处理所有这些数据。

这两个数据库中的模式都是相同的，我认为由于MariaDB中的插入速度很快，这可能排除了索引问题或服务器配置错误。

我已经尝试了JDBC连接字符串的多个变体,最后得到的结果是最快的：

  ?verifyServerCertificate=true\
  &useSSL=true\
  &requireSSL=true\
  &allowMultiQueries=true\
  &cachePrepStmts=true\
  &cacheResultSetMetadata=true\
  &cacheServerConfiguration=true\
  &elideSetAutoCommits=true\
  &maintainTimeStats=false\
  &prepStmtCacheSize=50000\
  &prepStmtCacheSqlLimit=204800\
  &rewriteBatchedStatements=false\
  &useBatchMultiSend=true\
  &useBatchMultiSendNumber=50000\
  &useBulkStmts=true\
  &useLocalSessionState=true\
  &useLocalTransactionState=true\
  &useServerPrepStmts=true

在mysql和mysql-connectorj中的性能在所有情况下都比MariaDB差。

我已经看了一个星期了，现在正在考虑采用我前面问题中建议的工作方法，如何提高mySQL vs SQL Server中一系列更新的速度？

万一是服务器配置错误，下面是关键变量：

key_buffer_size                16MB
innodb_buffer_pool_size        24GB (mem 30GB)
innodb_log_file_size           134MB
innodb_log_buffer_size         8MB
innodb_flush_log_at_trx_commit 0
max_allowed_packet             16MB

我的50,000次写入只是很小的数据量--大约2MB。但是对于SQL语法，当它通过JDBC连接时，这可能要大10倍--这是正确的吗？

下面是SQL和解释计划：

Describe `data`
+---------------+------------------+------+-----+---------------------+-------------------------------+
| Field         | Type             | Null | Key | Default             | Extra                         |
+---------------+------------------+------+-----+---------------------+-------------------------------+
| parentId      | int(10) unsigned | NO   | PRI | NULL                |                               |
| modifiedDate  | date             | NO   | PRI | NULL                |                               |
| valueDate     | date             | NO   | PRI | NULL                |                               |
| value         | float            | NO   |     | NULL                |                               |
| versionstamp  | int(10) unsigned | NO   |     | 1                   |                               |
| createdDate   | datetime         | YES  |     | current_timestamp() |                               |
| last_modified | datetime         | YES  |     | NULL                | on update current_timestamp() |
+---------------+------------------+------+-----+---------------------+-------------------------------+

INSERT INTO `data` (`value`, `parentId`, `modifiedDate`, `valueDate`) VALUES (4853.16314229298,52054,'20-Apr-18','28-Dec-18')

+------+-------------+-------+------+---------------+------+---------+------+------+-------+
| id   | select_type | table | type | possible_keys | key  | key_len | ref  | rows | Extra |
+------+-------------+-------+------+---------------+------+---------+------+------+-------+
|    1 | INSERT      | data  | ALL  | NULL          | NULL | NULL    | NULL | NULL | NULL  |
+------+-------------+-------+------+---------------+------+---------+------+------+-------+



UPDATE `data` SET `value` = 4853.16314229298 WHERE `parentId` = 52054 AND `modifiedDate` = '20-Apr-18' AND `valueDate` = '28-Dec-18'

+------+-------------+-------+-------+---------------+---------+---------+------+------+-------------+
| id   | select_type | table | type  | possible_keys | key     | key_len | ref  | rows | Extra       |
+------+-------------+-------+-------+---------------+---------+---------+------+------+-------------+
|    1 | SIMPLE      | data  | range | PRIMARY       | PRIMARY | 10      | NULL |    1 | Using where |
+------+-------------+-------+-------+---------------+---------+---------+------+------+-------------+


DELETE FROM `data` WHERE `parentId` = 52054 AND `modifiedDate` = '20-Apr-18' AND `valueDate` = '29-Jan-16'

+------+-------------+-------+-------+---------------+---------+---------+------+------+-------------+
| id   | select_type | table | type  | possible_keys | key     | key_len | ref  | rows | Extra       |
+------+-------------+-------+-------+---------------+---------+---------+------+------+-------------+
|    1 | SIMPLE      | data  | range | PRIMARY       | PRIMARY | 10      | NULL |    1 | Using where |
+------+-------------+-------+-------+---------------+---------+---------+------+------+-------------+

[更新]

JDBC用法--这是一个缩减的版本，因此请原谅任何严重的错误：

    final Connection connection = dataSource.getConnection();
    connection.setAutoCommit(false);
    try (PreparedStatement statement = connection.prepareStatement(
                 "UPDATE data SET value = ? " +
                         "WHERE parentId = ? " +
                         "AND modifiedDate = ? " +
                         "AND valueDate = ? ")) {
        // timeSeries is a list of 50,000 data points
        Arrays.stream(timeSeries)
                .forEach(ts -> {
            try {
                statement.setDouble(1, value);
                statement.setLong(2, parentId);
                statement.setDate(3, new java.sql.Date(
                        modifiedDate.getTime()));
                statement.setDate(4, new java.sql.Date(
                        valueDate.getTime()));
                statement.addBatch();
            } catch (SQLException e) {
                throw new RuntimeException(
                        "Bad batch statement handling", e);
            }
        });
        int[] results = statement.executeBatch();
        connection.commit();
    } catch (SQLException e) {
        connection.rollback();
        throw e;
    } finally {
        connection.close();
    }

我还有一些来自general_log的数据，显示了传入的JDBC调用，它看起来非常基本--一个'prepare'调用来设置语句，然后进行单独的更新。

13/06/2018 15:09    service_user_t[service_user_t] @  [9.177.2.31]  75954   298206495   Query   set autocommit=0
13/06/2018 15:09    service_user_t[service_user_t] @  [9.177.2.31]  75954   298206495   Prepare UPDATE `data` SET `value` = ? WHERE `parentId` = ? AND `modifiedDate` = ? AND `valueDate` = ?
13/06/2018 15:09    service_user_t[service_user_t] @  [9.177.2.31]  75954   298206495   Execute UPDATE `data` SET `value` = ? WHERE `parentId` = ? AND `modifiedDate` = ? AND `valueDate` = ?
13/06/2018 15:09    service_user_t[service_user_t] @  [9.177.2.31]  75954   298206495   Execute UPDATE `data` SET `value` = ? WHERE `parentId` = ? AND `modifiedDate` = ? AND `valueDate` = ?
13/06/2018 15:09    service_user_t[service_user_t] @  [9.177.2.31]  75954   298206495   Execute UPDATE `data` SET `value` = ? WHERE `parentId` = ? AND `modifiedDate` = ? AND `valueDate` = ?
13/06/2018 15:09    service_user_t[service_user_t] @  [9.177.2.31]  75954   298206495   Execute UPDATE `data` SET `value` = ? WHERE `parentId` = ? AND `modifiedDate` = ? AND `valueDate` = ?
13/06/2018 15:09    service_user_t[service_user_t] @  [9.177.2.31]  75954   298206495   Execute UPDATE `data` SET `value` = ? WHERE `parentId` = ? AND `modifiedDate` = ? AND `valueDate` = ?
etc
etc

Answer 1

在批处理中的某些行之间添加“BEGIN”和“COMMIT”语句。或者在批处理之前启动事务，然后在批处理之后提交。这将比成千上万个单独的声明要快得多。

如果您只执行插入操作，rewriteBatchStatements=true应该会大大加快它的速度，而不需要事务处理。你也可以把max_packet_size增加到1GB，这会做更多的批处理，也许你的整个批处理会被转换成1个非常大的multi-insert。