[GitHub] [spark] juliuszsompolski commented on a diff in pull request #40160: [SPARK-41725][CONNECT] Eager Execution of DF.sql()

["juliuszsompolski (via GitHub)" <gi...@apache.org>]

juliuszsompolski commented on code in PR #40160:
URL: https://github.com/apache/spark/pull/40160#discussion_r1172607370


##########
connector/connect/server/src/main/scala/org/apache/spark/sql/connect/planner/SparkConnectPlanner.scala:
##########
@@ -1450,10 +1458,79 @@ class SparkConnectPlanner(val session: SparkSession) {
         handleWriteOperationV2(command.getWriteOperationV2)
       case proto.Command.CommandTypeCase.EXTENSION =>
         handleCommandPlugin(command.getExtension)
+      case proto.Command.CommandTypeCase.SQL_COMMAND =>
+        handleSqlCommand(command.getSqlCommand, clientId, responseObserver)
       case _ => throw new UnsupportedOperationException(s"$command not supported.")
     }
   }
 
+  def handleSqlCommand(
+      getSqlCommand: SqlCommand,
+      clientId: String,
+      responseObserver: StreamObserver[ExecutePlanResponse]): Unit = {
+    // Eagerly execute commands of the provided SQL string.
+    val df = session.sql(getSqlCommand.getSql, getSqlCommand.getArgsMap)
+    // Check if commands have been executed.
+    val isCommand = df.queryExecution.commandExecuted.isInstanceOf[CommandResult]
+    val rows = df.logicalPlan match {
+      case lr: LocalRelation => lr.data
+      case cr: CommandResult => cr.rows
+      case _ => Seq.empty
+    }
+
+    // Convert the results to Arrow.
+    val schema = df.schema
+    val maxRecordsPerBatch = session.sessionState.conf.arrowMaxRecordsPerBatch
+    val maxBatchSize = (SparkEnv.get.conf.get(CONNECT_GRPC_ARROW_MAX_BATCH_SIZE) * 0.7).toLong
+    val timeZoneId = session.sessionState.conf.sessionLocalTimeZone
+
+    // Convert the data.
+    val bytes = if (rows.isEmpty) {
+      ArrowConverters.createEmptyArrowBatch(schema, timeZoneId)
+    } else {
+      val batches = ArrowConverters.toBatchWithSchemaIterator(
+        rows.iterator,
+        schema,
+        maxRecordsPerBatch,
+        maxBatchSize,
+        timeZoneId)
+      assert(batches.size == 1)
+      batches.next()
+    }
+
+    // To avoid explicit handling of the result on the client, we build the expected input
+    // of the relation on the server. The client has to simply forward the result.
+    val result = SqlCommandResult.newBuilder()
+    if (isCommand) {
+      result.setRelation(
+        proto.Relation
+          .newBuilder()
+          .setLocalRelation(
+            proto.LocalRelation
+              .newBuilder()
+              .setData(ByteString.copyFrom(bytes))))
+    } else {
+      result.setRelation(
+        proto.Relation
+          .newBuilder()
+          .setSql(
+            proto.SQL
+              .newBuilder()
+              .setQuery(getSqlCommand.getSql)
+              .putAllArgs(getSqlCommand.getArgsMap)))
+    }
+    // Exactly one SQL Command Result Batch
+    responseObserver.onNext(
+      ExecutePlanResponse
+        .newBuilder()
+        .setClientId(clientId)
+        .setSqlCommandResult(result)
+        .build())
+
+    // Send Metrics
+    SparkConnectStreamHandler.sendMetricsToResponse(clientId, df)

Review Comment:
   @grundprinzip note: this creates the ExecutePlansResponse proto object, but doesnt send it anywhere.
   Should be `responseObserver.onNext(SparkConnectStreamHandler.sendMetricsToResponse(clientId, df))`



##########
connector/connect/server/src/main/scala/org/apache/spark/sql/connect/service/SparkConnectStreamHandler.scala:
##########
@@ -163,45 +195,17 @@ class SparkConnectStreamHandler(responseObserver: StreamObserver[ExecutePlanResp
         response.setArrowBatch(batch)
         responseObserver.onNext(response.build())
       }
-
-      responseObserver.onNext(sendMetricsToResponse(clientId, dataframe))
-      responseObserver.onCompleted()
     }
   }
 
-  private def sendMetricsToResponse(clientId: String, rows: DataFrame): ExecutePlanResponse = {
+  def sendMetricsToResponse(clientId: String, rows: DataFrame): ExecutePlanResponse = {
     // Send a last batch with the metrics

Review Comment:
   maybe best rename this function, since it doesn't send anything?



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org