Prefetch a chunk of result for stream operation

Resolves #2132.

Use fs2 Stream's `prefetchN` to buffer query results so that slow
downstream operations don't slow down transaction commit.

The buffered result size is `chunkSize` (equals to `fetchSize`) by
default. If want a different buffer size, can use `transactNoPrefetc` to
get a stream without `prefetchN` and append any prefetch operations
wanted.

modules/core/src/main/scala/doobie/syntax/stream.scala

@@ -8,16 +8,23 @@ import doobie.util.compat.=:=
 import doobie.util.transactor.Transactor
 import doobie.free.connection.ConnectionIO
 import cats.data.Kleisli
+import cats.effect.Concurrent
 import cats.effect.kernel.{Async, MonadCancelThrow}
 import fs2.{Pipe, Stream}
 
 class StreamOps[F[_], A](fa: Stream[F, A]) {
-  def transact[M[_]: MonadCancelThrow](xa: Transactor[M])(implicit
+  def transactNoPrefetch[M[_]: MonadCancelThrow](xa: Transactor[M])(implicit
       ev: Stream[F, A] =:= Stream[ConnectionIO, A]
   ): Stream[M, A] = xa.transP.apply(fa)
+
+  def transact[M[_]: Concurrent](xa: Transactor[M])(implicit
+      ev: Stream[F, A] =:= Stream[ConnectionIO, A]
+  ): Stream[M, A] = transactNoPrefetch(xa).prefetchN(1)
+
 }
 class KleisliStreamOps[A, B](fa: Stream[Kleisli[ConnectionIO, A, *], B]) {
-  def transact[M[_]: MonadCancelThrow](xa: Transactor[M]): Stream[Kleisli[M, A, *], B] = xa.transPK[A].apply(fa)
+  def transactNoPrefetch[M[_]: MonadCancelThrow](xa: Transactor[M]): Stream[Kleisli[M, A, *], B] = xa.transPK[A].apply(fa)
+  def transact[M[_]: Concurrent](xa: Transactor[M]): Stream[Kleisli[M, A, *], B] = transactNoPrefetch(xa).prefetchN(1)
 }
 class PipeOps[F[_], A, B](inner: Pipe[F, A, B]) {
   def transact[M[_]: Async](xa: Transactor[M])(implicit ev: Pipe[F, A, B] =:= Pipe[ConnectionIO, A, B]): Pipe[M, A, B] =

modules/hikari/src/test/scala/doobie/postgres/StreamPrefetchSuite.scala

@@ -0,0 +1,101 @@
+// Copyright (c) 2013-2020 Rob Norris and Contributors
+// This software is licensed under the MIT License (MIT).
+// For more information see LICENSE or https://opensource.org/licenses/MIT
+
+package doobie.postgres
+
+import cats.effect.IO
+import com.zaxxer.hikari.HikariDataSource
+import doobie.*
+import doobie.implicits.*
+import doobie.util.transactor.Strategy
+import fs2.Stream
+
+import java.sql.SQLTransientConnectionException
+import java.util.concurrent.Executors
+import java.util.concurrent.atomic.AtomicBoolean
+import scala.concurrent.ExecutionContext
+import scala.concurrent.duration.*
+
+class StreamPrefetchSuite extends munit.FunSuite {
+
+  import cats.effect.unsafe.implicits.global
+
+  private var dataSource: HikariDataSource = null
+  private var xa: Transactor[IO] = null
+  private val count = 100
+
+  private def createDataSource() = {
+
+    Class.forName("org.postgresql.Driver")
+    val dataSource = new HikariDataSource
+
+    dataSource `setJdbcUrl` "jdbc:postgresql://localhost:5432/postgres"
+    dataSource `setUsername` "postgres"
+    dataSource `setPassword` "password"
+    dataSource `setMaximumPoolSize` 1
+    dataSource `setConnectionTimeout` 2000
+    dataSource
+  }
+
+  override def beforeAll(): Unit = {
+    super.beforeAll()
+    dataSource = createDataSource()
+    xa = Transactor.fromDataSource[IO](
+      dataSource,
+      ExecutionContext.fromExecutor(Executors.newFixedThreadPool(32))
+    )
+    val insert = for {
+      _ <- Stream.eval(sql"CREATE TABLE if not exists stream_cancel_test (i text)".update.run.transact(xa))
+      _ <- Stream.eval(sql"truncate table stream_cancel_test".update.run.transact(xa))
+      _ <- Stream.eval(sql"INSERT INTO stream_cancel_test select 1 from generate_series(1, $count)".update.run.transact(xa))
+    } yield ()
+
+    insert.compile.drain.unsafeRunSync()
+  }
+
+  override def afterAll(): Unit = {
+    dataSource.close()
+    super.afterAll()
+  }
+
+  test("Connection returned before stream is drained, if chunk size is larger than result count") {
+    val xa = Transactor.fromDataSource[IO](
+      dataSource,
+      ExecutionContext.fromExecutor(Executors.newFixedThreadPool(32))
+    )
+
+    val streamLargerBuffer = fr"select * from stream_cancel_test".query[Int].streamWithChunkSize(200).transact(xa)
+      .evalMap(_ => fr"select 1".query[Int].unique.transact(xa))
+      .compile.count
+
+    assertEquals(streamLargerBuffer.unsafeRunSync(), count.toLong)
+  }
+
+  test("Connection is not returned after consuming only 1 chunk, if chunk size is smaller than result count") {
+    val streamSmallerBuffer = fr"select * from stream_cancel_test".query[Int].streamWithChunkSize(10).transact(xa)
+      .evalMap(_ => fr"select 1".query[Int].unique.transact(xa))
+      .compile.count
+
+    intercept[SQLTransientConnectionException](streamSmallerBuffer.unsafeRunSync())
+  }
+
+  test("Connection returned before stream is drained, if chunk size is smaller than result count") {
+    val hasClosed = new AtomicBoolean(false)
+    val xaCopy = xa.copy(
+      strategy0 = Strategy.default.copy(
+        always = Strategy.default.always.flatMap(_ => FC.delay(hasClosed.set(true)))
+      )
+    )
+
+    val earlyClose = new AtomicBoolean(false)
+
+    val streamSmallerBufferValid = fr"select * from stream_cancel_test".query[Int].streamWithChunkSize(10).transact(xaCopy)
+      .evalMap{_ => IO{if (hasClosed.get()) earlyClose.set(true)} >> IO.sleep(10.milliseconds) }
+      .compile.count
+
+    assertEquals(streamSmallerBufferValid.unsafeRunSync(), count.toLong)
+    assertEquals(earlyClose.get(), true)
+  }
+
+}