restapi（8）- restapi-sql：用户自主的服务

学习函数式编程初衷是看到自己熟悉的oop编程语言和sql数据库在现代商业社会中前景暗淡，准备完全放弃windows技术栈转到分布式大数据技术领域的。但是在现实中理想总是不如人意，本来想在一个规模较小的公司展展拳脚，以为小公司会少点历史包袱，有利于全面技术改造。但现实是：即使是小公司，一旦有个成熟的产品，那么进行全面的技术更新基本上是不可能的了，因为公司要生存，开发人员很难新旧技术之间随时切换。除非有狂热的热情，员工怠慢甚至抵制情绪不容易解决。只能采取逐步切换方式：保留原有产品的后期维护不动，新产品开发用一些新的技术。在我们这里的情况就是：以前一堆c#、sqlserver的东西必须保留，新的功能比如大数据、ai、识别等必须用新的手段如scala、python、dart、akka、kafka、cassandra、mongodb来开发。好了，新旧两个开发平台之间的软件系统对接又变成了一个问题。

现在我们这里有个需求：把在linux-ubuntu akka-cluster集群环境里mongodb里数据处理的结果传给windows server下SQLServer里。这是一种典型的异系统集成场景。我的解决方案是通过一个restapi服务作为两个系统的数据桥梁，这个restapi的最基本要求是：

1、支持任何操作系统前端：这个没什么问题，在http层上通过json交换数据

2、能读写mongodb：在前面讨论的restapi-mongo已经实现了这一功能

3、能读写windows server环境下的sqlserver：这个是本篇讨论的主题

4、用户能够比较方便的对平台数据库进行操作，最好免去前后双方每类操作都需要进行协定model这一过程，也就是能达到用户随意调用服务

前面曾经实现了一个jdbc-engine项目，基于scalikejdbc，不过只示范了slick-h2相关的功能。现在需要sqlserver-jdbc驱动，然后试试能不能在JVM里驱动windows下的sqlserver。maven里找不到sqlserver的驱动，但从微软官网可以下载mssql-jdbc-7.0.0.jre8.jar。这是个jar，在sbt里称作unmanagedjar，不能摆在build.sbt的dependency里。这个需要摆在项目根目录下的lib目录下即可（也可以在放在build.sbt里unmanagedBase :=?? 指定的路径下）。然后是数据库连接，下面是可以使用sqlserver的application.conf配置文件内容：

# JDBC settings
prod {
  db {
    h2 {
      driver = "org.h2.Driver"
      url = "jdbc:h2:tcp://localhost/~/slickdemo"
      user = ""
      password = ""
      poolFactoryName = "hikaricp"
      numThreads = 10
      maxConnections = 12
      minConnections = 4
      keepAliveConnection = true
    }
    mysql {
      driver = "com.mysql.cj.jdbc.Driver"
      url = "jdbc:mysql://localhost:3306/testdb"
      user = "root"
      password = "123"
      poolFactoryName = "hikaricp"
      numThreads = 10
      maxConnections = 12
      minConnections = 4
      keepAliveConnection = true
    }
    postgres {
      driver = "org.postgresql.Driver"
      url = "jdbc:postgresql://localhost:5432/testdb"
      user = "root"
      password = "123"
      poolFactoryName = "hikaricp"
      numThreads = 10
      maxConnections = 12
      minConnections = 4
      keepAliveConnection = true
    }
    mssql {
      driver = "com.microsoft.sqlserver.jdbc.SQLServerDriver"
      url = "jdbc:sqlserver://192.168.11.164:1433;integratedSecurity=false;Connect Timeout=3000"
      user = "sa"
      password = "Tiger2020"
      poolFactoryName = "hikaricp"
      numThreads = 10
      maxConnections = 12
      minConnections = 4
      keepAliveConnection = true
      connectionTimeout = 3000
    }
    termtxns {
      driver = "com.microsoft.sqlserver.jdbc.SQLServerDriver"
      url = "jdbc:sqlserver://192.168.11.164:1433;DATABASE=TERMTXNS;integratedSecurity=false;Connect Timeout=3000"
      user = "sa"
      password = "Tiger2020"
      poolFactoryName = "hikaricp"
      numThreads = 10
      maxConnections = 12
      minConnections = 4
      keepAliveConnection = true
      connectionTimeout = 3000
    }
    crmdb {
      driver = "com.microsoft.sqlserver.jdbc.SQLServerDriver"
      url = "jdbc:sqlserver://192.168.11.164:1433;DATABASE=CRMDB;integratedSecurity=false;Connect Timeout=3000"
      user = "sa"
      password = "Tiger2020"
      poolFactoryName = "hikaricp"
      numThreads = 10
      maxConnections = 12
      minConnections = 4
      keepAliveConnection = true
      connectionTimeout = 3000
    }
  }
  # scallikejdbc Global settings
  scalikejdbc.global.loggingSQLAndTime.enabled = true
  scalikejdbc.global.loggingSQLAndTime.logLevel = info
  scalikejdbc.global.loggingSQLAndTime.warningEnabled = true
  scalikejdbc.global.loggingSQLAndTime.warningThresholdMillis = 1000
  scalikejdbc.global.loggingSQLAndTime.warningLogLevel = warn
  scalikejdbc.global.loggingSQLAndTime.singleLineMode = false
  scalikejdbc.global.loggingSQLAndTime.printUnprocessedStackTrace = false
  scalikejdbc.global.loggingSQLAndTime.stackTraceDepth = 10
}

这个文件里的mssql,termtxns,crmdb段落都是给sqlserver的，它们都使用hikaricp线程池管理。

在jdbc-engine里启动数据库方式如下：

  ConfigDBsWithEnv("prod").setup('termtxns)
  ConfigDBsWithEnv("prod").setup('crmdb)
  ConfigDBsWithEnv("prod").loadGlobalSettings()

这段打开了在配置文件中用termtxns,crmdb注明的数据库。

下面是SqlHttpServer.scala的代码：

package com.datatech.rest.sql
import akka.http.scaladsl.Http
import akka.http.scaladsl.server.Directives._
import pdi.jwt._
import AuthBase._
import MockUserAuthService._
import com.datatech.sdp.jdbc.config.ConfigDBsWithEnv

import akka.actor.ActorSystem
import akka.stream.ActorMaterializer

import Repo._
import SqlRoute._

object SqlHttpServer extends App {

  implicit val httpSys = ActorSystem("sql-http-sys")
  implicit val httpMat = ActorMaterializer()
  implicit val httpEC = httpSys.dispatcher

  ConfigDBsWithEnv("prod").setup('termtxns)
  ConfigDBsWithEnv("prod").setup('crmdb)
  ConfigDBsWithEnv("prod").loadGlobalSettings()

  implicit val authenticator = new AuthBase()
    .withAlgorithm(JwtAlgorithm.HS256)
    .withSecretKey("OpenSesame")
    .withUserFunc(getValidUser)

  val route =
    path("auth") {
      authenticateBasic(realm = "auth", authenticator.getUserInfo) { userinfo =>
        post { complete(authenticator.issueJwt(userinfo))}
      }
    } ~
      pathPrefix("api") {
        authenticateOAuth2(realm = "api", authenticator.authenticateToken) { token =>
          new SqlRoute("sql", token)(new JDBCRepo)
            .route
          // ~ ...
        }
      }

  val (port, host) = (50081,"192.168.11.189")

  val bindingFuture = Http().bindAndHandle(route,host,port)

  println(s"Server running at $host $port. Press any key to exit ...")

  scala.io.StdIn.readLine()
  
  bindingFuture.flatMap(_.unbind())
    .onComplete(_ => httpSys.terminate())

}

服务入口在http://mydemo.com/api/sql，服务包括get,post,put三类，参考这个SqlRoute：

package com.datatech.rest.sql
import akka.http.scaladsl.server.Directives
import akka.stream.ActorMaterializer
import akka.http.scaladsl.model._
import akka.actor.ActorSystem
import com.datatech.rest.sql.Repo.JDBCRepo
import akka.http.scaladsl.common._
import spray.json.DefaultJsonProtocol
import akka.http.scaladsl.marshallers.sprayjson.SprayJsonSupport

trait JsFormats extends SprayJsonSupport with DefaultJsonProtocol
object JsConverters extends JsFormats {
  import SqlModels._
  implicit val brandFormat = jsonFormat2(Brand)
  implicit val customerFormat = jsonFormat6(Customer)
}

object SqlRoute {
  import JsConverters._
  implicit val jsonStreamingSupport = EntityStreamingSupport.json()
    .withParallelMarshalling(parallelism = 8, unordered = false)

  class SqlRoute(val pathName: String, val jwt: String)(repo: JDBCRepo)(
  implicit  sys: ActorSystem, mat: ActorMaterializer) extends Directives with JsonConverter {
    val route = pathPrefix(pathName) {
      path(Segment / Remaining) { case (db, tbl) =>
        (get & parameter('sqltext)) { sql => {
          val rsc = new RSConverter
          val rows = repo.query[Map[String,Any]](db, sql, rsc.resultSet2Map)
          complete(rows.map(m => toJson(m)))
        }
        } ~ (post & parameter('sqltext)) { sql =>
              entity(as[String]){ json =>
                repo.batchInsert(db,tbl,sql,json)
                complete(StatusCodes.OK)
              }
        } ~ put {
          entity(as[Seq[String]]) { sqls =>
            repo.update(db, sqls)
            complete(StatusCodes.OK)
          }
        }
      }
    }
  }
}

jdbc-engine的特点是可以用字符类型的sql语句来操作。所以我们可以通过传递字符串型的sql语句来实现服务调用，使用门槛低，方便通用。restapi-sql提供的是对服务器端sqlserver的普通操作，包括读get,写入post,更改put。这些sqlserver操作部分是在JDBCRepo里的：

package com.datatech.rest.sql
import com.datatech.sdp.jdbc.engine.JDBCEngine._
import com.datatech.sdp.jdbc.engine.{JDBCQueryContext, JDBCUpdateContext}
import scalikejdbc._
import akka.stream.ActorMaterializer
import com.datatech.sdp.result.DBOResult.DBOResult
import akka.stream.scaladsl._
import scala.concurrent._
import SqlModels._

object Repo {

  class JDBCRepo(implicit ec: ExecutionContextExecutor, mat: ActorMaterializer) {
    def query[R](db: String, sqlText: String, toRow: WrappedResultSet => R): Source[R,Any] = {
      //construct the context
      val ctx = JDBCQueryContext(
        dbName = Symbol(db),
        statement = sqlText
      )
      jdbcAkkaStream(ctx,toRow)
    }

    def query(db: String, tbl: String, sqlText: String) = {
      //construct the context
      val ctx = JDBCQueryContext(
        dbName = Symbol(db),
        statement = sqlText
      )
      jdbcQueryResult[Vector,RS](ctx,getConverter(tbl)).toFuture[Vector[RS]]
    }

    def update(db: String, sqlTexts: Seq[String]): DBOResult[Seq[Long]] = {
      val ctx = JDBCUpdateContext(
        dbName = Symbol(db),
        statements = sqlTexts
      )
      jdbcTxUpdates(ctx)
    }
    def bulkInsert[P](db: String, sqlText: String, prepParams: P => Seq[Any], params: Source[P,_]) = {
      val insertAction = JDBCActionStream(
        dbName = Symbol(db),
        parallelism = 4,
        processInOrder = false,
        statement = sqlText,
        prepareParams = prepParams
      )
      params.via(insertAction.performOnRow).to(Sink.ignore).run()
    }
    def batchInsert(db: String, tbl: String, sqlText: String, jsonParams: String):DBOResult[Seq[Long]] = {
      val ctx = JDBCUpdateContext(
        dbName = Symbol(db),
        statements = Seq(sqlText),
        batch = true,
        parameters = getSeqParams(jsonParams,sqlText)
      )
      jdbcBatchUpdate[Seq](ctx)
    }
  }
  import monix.execution.Scheduler.Implicits.global
  implicit class DBResultToFuture(dbr: DBOResult[_]){
    def toFuture[R] = {
      dbr.value.value.runToFuture.map {
        eor =>
          eor match {
            case Right(or) => or match {
              case Some(r) => r.asInstanceOf[R]
              case None => throw new RuntimeException("Operation produced None result!")
            }
            case Left(err) => throw new RuntimeException(err)
          }
      }
    }
  }
}

读query部分即 def query[R](db: String, sqlText: String, toRow: WrappedResultSet => R): Source[R,Any] = {...} 这个函数返回Source[R,Any]，下面我们好好谈谈这个R：R是读的结果，通常是某个类或model，比如读取Person记录返回一组Person类的实例。这里有一种强类型的感觉。一开始我也是随大流坚持建model后用toJson[E],fromJson[E]这样做线上数据转换。现在的问题是restapi-sql是一项公共服务，使用者知道sqlserver上有些什么表，然后希望通过sql语句来从这些表里读取数据。这些sql语句可能超出表的界限如sql join, union等，如果我们坚持每个返回结果都必须有个对应的model，那么显然就会牺牲这个服务的通用性。实际上，http线上数据交换本身就不可能是强类型的，因为经过了json转换。对于json转换来说，只要求字段名称、字段类型对称就行了。至于从什么类型转换成了另一个什么类型都没问题。所以，字段名+字段值的表现形式不就是Map[K,V]吗，我们就用Map[K,V]作为万能model就行了，没人知道。也就是说用户方通过sql语句指定返回的字段名称，它们可能是任何类型Any，具体类型自然会由数据库补上。服务方从数据库读取结果ResultSet后转成Map[K,V]然后再转成json返回给用户，用户可以用Map[String,Any]信息产生任何类型，这就是自主。好，就来看看如何将ResultSet转成Map[String,Any]：

package com.datatech.rest.sql
import scalikejdbc._
import java.sql.ResultSetMetaData
class RSConverter {
  import RSConverterUtil._
  var rsMeta: ResultSetMetaData = _
  var columnCount: Int = 0
  var rsFields: List[(String,String)] = List[(String,String)]()

  def getFieldsInfo:List[(String,String)] =
    ( 1 until columnCount).foldLeft(List[(String,String)]()) {
    case (cons,i) =>
      (rsMeta.getColumnLabel(i) -> rsMeta.getColumnTypeName(i)) :: cons
  }
  def resultSet2Map(rs: WrappedResultSet): Map[String,Any] = {
    if(columnCount == 0) {
      rsMeta =  rs.underlying.getMetaData
      columnCount = rsMeta.getColumnCount
      rsFields = getFieldsInfo
    }
    rsFields.foldLeft(Map[String,Any]()) {
      case (m,(n,t)) =>
        m + (n -> rsFieldValue(n,t,rs))
    }
  }
}
object RSConverterUtil {
  import scala.collection.immutable.TreeMap
  def map2Params(stm: String, m: Map[String,Any]): Seq[Any] = {
    val sortedParams = m.foldLeft(TreeMap[Int,Any]()) {
      case (t,(k,v)) => t + (stm.indexOfSlice(k) -> v)
    }
    sortedParams.map(_._2).toSeq
  }
  def rsFieldValue(fldname: String, fldType: String, rs: WrappedResultSet): Any = fldType match {
    case "LONGVARCHAR" => rs.string(fldname)
    case "VARCHAR" => rs.string(fldname)
    case "CHAR" => rs.string(fldname)
    case "BIT" => rs.boolean(fldname)
    case "TIME" => rs.time(fldname)
    case "TIMESTAMP" => rs.timestamp(fldname)
    case "ARRAY" => rs.array(fldname)
    case "NUMERIC" => rs.bigDecimal(fldname)
    case "BLOB" => rs.blob(fldname)
    case "TINYINT" => rs.byte(fldname)
    case "VARBINARY" => rs.bytes(fldname)
    case "BINARY" => rs.bytes(fldname)
    case "CLOB" => rs.clob(fldname)
    case "DATE" => rs.date(fldname)
    case "DOUBLE" => rs.double(fldname)
    case "REAL" => rs.float(fldname)
    case "FLOAT" => rs.float(fldname)
    case "INTEGER" => rs.int(fldname)
    case "SMALLINT" => rs.int(fldname)
    case "Option[Int]" => rs.intOpt(fldname)
    case "BIGINT" => rs.long(fldname)
  }
}

这段主要功能是将JDBC的ResultSet转换成Map[String,Any]。在前面讨论的restapi-mongo我们可以进行Document到Map[String,Any]的转换以实现同样的目的。

下面是个调用query服务的例子：

    val getAllRequest = HttpRequest(
      HttpMethods.GET,
      uri = "http://192.168.11.189:50081/api/sql/termtxns/brand?sqltext=SELECT%20*%20FROM%20BRAND",
    ).addHeader(authentication)

    (for {
      response <- Http().singleRequest(getAllRequest)
      json <- Unmarshal(response.entity).to[String]
    } yield message).andThen {
      case Success(msg) => println(s"Received json collection: $json")
      case Failure(err) => println(s"Error: ${err.getMessage}")
    }

特点是我只需要提供sql语句，服务就会返回一个json数组，然后我怎么把json转成任何类型就随我高兴了。

再看看post服务：在这里希望实现一种批次型插入表的功能，比如从一个数据表里把数据搬到另外一个表。一般来讲在jdbc操作里首先得提供一个模版，如：insert into person(fullname,code) values(?,?)，然后通过提供一组参数值来实现批次插入。当然，为安全起见，我们还是需要确定正确的参数位置，这个可以从sql语句里获取：

  def map2Params(stm: String, m: Map[String,Any]): Seq[Any] = {
    val sortedParams = m.foldLeft(TreeMap[Int,Any]()) {
      case (t,(k,v)) => t + (stm.toUpperCase.indexOfSlice(k.toUpperCase) -> v)
    }
    sortedParams.map(_._2).toSeq
  }

  def getSeqParams(json: String, sql: String): Seq[Seq[Any]] = {
    val seqOfjson = fromJson[Seq[String]](json)
    val prs = seqOfjson.map(fromJson[Map[String,Any]])
    prs.map(RSConverterUtil.map2Params(sql,_))
  }

下面是个批次插入的示范代码：

    val encodedSelect = URLEncode.encode("select id as code, name as fullname from members")
    val encodedInsert = URLEncode.encode("insert into person(fullname,code) values(?,?)")
    val getMembers = HttpRequest(
       HttpMethods.GET,
       uri = "http://192.168.0.189:50081/api/sql/h2/members?sqltext="+encodedSelect
      ).addHeader(authentication)
    val postRequest = HttpRequest(
      HttpMethods.POST,
      uri = "http://192.168.0.189:50081/api/sql/h2/person?sqltext="+encodedInsert,
    ).addHeader(authentication)
    
    (for {
      _ <- update("http://192.168.0.189:50081/api/sql/h2/person",Seq(createCTX))
      respMembers <- Http().singleRequest(getMembers)
      message <- Unmarshal(respMembers.entity).to[String]
      reqEntity <- Marshal(message).to[RequestEntity]
      respInsert <- Http().singleRequest(postRequest.copy(entity = reqEntity))
 //       HttpEntity(ContentTypes.`application/json`,ByteString(message))))
    } yield respInsert).onComplete {
      case Success(r@HttpResponse(StatusCodes.OK, _, entity, _)) =>
        println("builk insert successful!")
      case Success(_) => println("builk insert failed!")
      case Failure(err) => println(s"Error: ${err.getMessage}")
    }

你看，我特别把参数值清单里字段位置和insert sql里字段先后位置颠倒了，但还是得到正确的结果。

最后是put：这是为批次型的事物处理设计的。接受一条或者多条无参数sql指令，多条指令会在一个事物中执行。具体使用方式如下：

    def update(url: String, cmds: Seq[String])(implicit token: Authorization): Future[HttpResponse] =
    for {
      reqEntity <- Marshal(cmds).to[RequestEntity]
      response <- Http().singleRequest(HttpRequest(
        method=HttpMethods.PUT,uri=url,entity=reqEntity)
      .addHeader(token))
    } yield response

在上面的讨论里介绍了基于sqlserver的rest服务，与前面讨论的restapi-mongo从原理上区别并不大，重点是实现了用户主导的数据库操作。