优化F＃字符串连接

Question

I'm building a MySql query that batch inserts 4096 records at once. The actual insert is quite fast but the bottleneck is generating the query. Any hints on optimizing this? The string generation is currently taking about 18 times longer than the query.

                    let builder = StringBuilder(524288)
                    Printf.bprintf builder
                        "
                         INSERT INTO %s
                             (`ID`,
                              `Tag`,
                              `Port`,
                              `Excess`,
                              `Return`,
                              `StartDate`,
                              `EndDate`
                              ) 
                          values "
                        x.evaluationstable

                    evaluations
                    |> Seq.iter(fun (e) ->
                        Printf.bprintf builder 
                            " (%d, '%s', '%s', %A, %A, %A, %A), "
                            e.ID
                            e.Tag
                            e.Port
                            e.Excess
                            e.Return
                            (e.StartDate.ToString(x.datetimeformat))
                            (e.EndDate.ToString(x.datetimeformat))
                    )

Answer 1

Try using StringBuilder.AppendFormat instead of Printf.bprintf . When I made this change in my example of your question, I saw a huge performance increase (~80x).

evaluations
|> Seq.iter (fun (e) ->
    builder.AppendFormat(
        " ({0}, '{1}', '{2}', {3}, {4}, {5}, {6}), ",
        e.ID,
        e.Tag,
        e.Port,
        e.Excess,
        e.Return,
        (e.StartDate.ToString("MM/dd/yyyy")),
        (e.EndDate.ToString("MM/dd/yyyy"))
    ) |> ignore
)

Answer 2

I would try to avoid embedding the data directly into SQL to start with. Use a sequence of prepared statements with parameters, and set those parameters to the values (without formatting them). That's safer and is likely to be a lot more efficient.

Whether you can still do this in a batch rather than in several separate calls just within the same transaction, I'm not sure.