Slick 3.0批量插入或更新(upsert)

opu*_*111 26 mysql sql scala slick typesafe

在Slick 3.0中执行批量insertOrUpdate的正确方法是什么?

我正在使用适当查询的MySQL

INSERT INTO table (a,b,c) VALUES (1,2,3),(4,5,6)
ON DUPLICATE KEY UPDATE c=VALUES(a)+VALUES(b);
Run Code Online (Sandbox Code Playgroud)

MySQL批量INSERT或UPDATE

这是我目前的代码非常慢:-(

// FIXME -- this is slow but will stop repeats, an insertOrUpdate
// functions for a list would be much better
val rowsInserted = rows.map {
  row => await(run(TableQuery[FooTable].insertOrUpdate(row)))
}.sum
Run Code Online (Sandbox Code Playgroud)

我正在寻找的是相当于

def insertOrUpdate(values: Iterable[U]): DriverAction[MultiInsertResult, NoStream, Effect.Write]
Run Code Online (Sandbox Code Playgroud)

Sea*_*ira 40

有几种方法可以使这段代码更快(每一种都应该比前面的代码更快,但它逐渐减少了惯用的代码):

  • 在slick-pg 0.16.1+上运行insertOrUpdateAll而不是insertOrUpdateif

    await(run(TableQuery[FooTable].insertOrUpdateAll rows)).sum
    
    Run Code Online (Sandbox Code Playgroud)
  • 一次运行您的DBIO事件,而不是在运行下一个之前等待每个事件提交:

    val toBeInserted = rows.map { row => TableQuery[FooTable].insertOrUpdate(row) }
    val inOneGo = DBIO.sequence(toBeInserted)
    val dbioFuture = run(inOneGo)
    // Optionally, you can add a `.transactionally`
    // and / or `.withPinnedSession` here to pin all of these upserts
    // to the same transaction / connection
    // which *may* get you a little more speed:
    // val dbioFuture = run(inOneGo.transactionally)
    val rowsInserted = await(dbioFuture).sum
    
    Run Code Online (Sandbox Code Playgroud)
  • 下拉到JDBC级别并一次性运行upsert(通过这个答案的想法):

    val SQL = """INSERT INTO table (a,b,c) VALUES (?, ?, ?)
    ON DUPLICATE KEY UPDATE c=VALUES(a)+VALUES(b);"""
    
    SimpleDBIO[List[Int]] { session =>
      val statement = session.connection.prepareStatement(SQL)
      rows.map { row =>
        statement.setInt(1, row.a)
        statement.setInt(2, row.b)
        statement.setInt(3, row.c)
        statement.addBatch()
      }
      statement.executeBatch()
    }
    
    Run Code Online (Sandbox Code Playgroud)

  • `.transactionally`为+1 - 仅此一项改进了从3分钟到1分钟插入50,000行的性能. (2认同)