cloud-fan commented on a change in pull request #31905: URL: https://github.com/apache/spark/pull/31905#discussion_r655400419
########## File path: sql/core/src/main/scala/org/apache/spark/sql/Observation.scala ########## @@ -0,0 +1,170 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql + +import java.util.UUID +import java.util.concurrent.TimeUnit + +import org.apache.spark.sql.execution.QueryExecution +import org.apache.spark.sql.util.QueryExecutionListener + +/** + * Helper class to simplify usage of [[Dataset.observe(String, Column, Column*)]]: + * + * {{{ + * // Observe row count (rows) and highest id (maxid) in the Dataset while writing it + * val observation = Observation("my_metrics") + * val observed_ds = ds.observe(observation, count(lit(1)).as("rows"), max($"id").as("maxid")) + * observed_ds.write.parquet("ds.parquet") + * val metrics = observation.get Review comment: shall we also mention the `on` API here? ########## File path: sql/core/src/main/scala/org/apache/spark/sql/Observation.scala ########## @@ -0,0 +1,170 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql + +import java.util.UUID +import java.util.concurrent.TimeUnit + +import org.apache.spark.sql.execution.QueryExecution +import org.apache.spark.sql.util.QueryExecutionListener + +/** + * Helper class to simplify usage of [[Dataset.observe(String, Column, Column*)]]: + * + * {{{ + * // Observe row count (rows) and highest id (maxid) in the Dataset while writing it + * val observation = Observation("my_metrics") + * val observed_ds = ds.observe(observation, count(lit(1)).as("rows"), max($"id").as("maxid")) + * observed_ds.write.parquet("ds.parquet") + * val metrics = observation.get + * }}} + * + * This collects the metrics while the first action is executed on the obseerved dataset. Subsequent + * actions do not modify the metrics returned by [[org.apache.spark.sql.Observation.get]]. Retrieval + * of the metric via [[org.apache.spark.sql.Observation.get]] blocks until the first action has + * finished and metrics become available. You can add a timeout to that blocking via + * [[org.apache.spark.sql.Observation.waitCompleted]]: + * + * {{{ + * if (observation.waitCompleted(100, TimeUnit.MILLISECONDS)) { + * observation.get + * } + * }}} + * + * This class does not support streaming datasets. + * + * @param name name of the metric + * @since 3.2.0 + */ +class Observation(name: String) { + + private val listener: ObservationListener = ObservationListener(this) + + private var sparkSession: Option[SparkSession] = None + + @volatile private var row: Option[Row] = None + + /** + * Attaches this observation to the given [[Dataset]] to observe aggregation expressions. + * + * @param ds dataset + * @param expr first aggregation expression + * @param exprs more aggregation expressions + * @tparam T dataset type + * @return observed dataset + * @throws IllegalArgumentException If this is a streaming Dataset (ds.isStreaming == true) + */ + def on[T](ds: Dataset[T], expr: Column, exprs: Column*): Dataset[T] = { + if (ds.isStreaming) { + throw new IllegalArgumentException("Observation does not support streaming Datasets") + } + register(ds.sparkSession) + ds.observe(name, expr, exprs: _*) + } + + /** + * Waits for the first action on the observed dataset to complete and returns true. + * The result is then available through the get method. + * This method times out after the given amount of time returning false. + * + * @param time timeout + * @param unit timeout time unit + * @return true if action completed within timeout, false otherwise + * @throws InterruptedException interrupted while waiting + */ + def waitCompleted(time: Long, unit: TimeUnit): Boolean = waitCompleted(Some(unit.toMillis(time))) + + /** + * Get the observed metrics. This waits until the observed dataset finishes its first action. + * If you want to wait for the result and provide a timeout, use [[waitCompleted]]. Only the + * result of the first action is available. Subsequent actions do not modify the result. + * + * @return the observed metrics as a [[Row]] + * @throws InterruptedException interrupted while waiting + */ + def get: Row = { + assert(waitCompleted(None), "waitCompleted without timeout returned false") + row.get + } + + private def waitCompleted(millis: Option[Long]): Boolean = { Review comment: shall we use `-1` to indicate no timeout? ########## File path: sql/core/src/main/scala/org/apache/spark/sql/Observation.scala ########## @@ -0,0 +1,170 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql + +import java.util.UUID +import java.util.concurrent.TimeUnit + +import org.apache.spark.sql.execution.QueryExecution +import org.apache.spark.sql.util.QueryExecutionListener + +/** + * Helper class to simplify usage of [[Dataset.observe(String, Column, Column*)]]: + * + * {{{ + * // Observe row count (rows) and highest id (maxid) in the Dataset while writing it + * val observation = Observation("my_metrics") + * val observed_ds = ds.observe(observation, count(lit(1)).as("rows"), max($"id").as("maxid")) + * observed_ds.write.parquet("ds.parquet") + * val metrics = observation.get + * }}} + * + * This collects the metrics while the first action is executed on the obseerved dataset. Subsequent + * actions do not modify the metrics returned by [[org.apache.spark.sql.Observation.get]]. Retrieval + * of the metric via [[org.apache.spark.sql.Observation.get]] blocks until the first action has + * finished and metrics become available. You can add a timeout to that blocking via + * [[org.apache.spark.sql.Observation.waitCompleted]]: + * + * {{{ + * if (observation.waitCompleted(100, TimeUnit.MILLISECONDS)) { + * observation.get + * } + * }}} + * + * This class does not support streaming datasets. + * + * @param name name of the metric + * @since 3.2.0 + */ +class Observation(name: String) { + + private val listener: ObservationListener = ObservationListener(this) + + private var sparkSession: Option[SparkSession] = None + + @volatile private var row: Option[Row] = None + + /** + * Attaches this observation to the given [[Dataset]] to observe aggregation expressions. + * + * @param ds dataset + * @param expr first aggregation expression + * @param exprs more aggregation expressions + * @tparam T dataset type + * @return observed dataset + * @throws IllegalArgumentException If this is a streaming Dataset (ds.isStreaming == true) + */ + def on[T](ds: Dataset[T], expr: Column, exprs: Column*): Dataset[T] = { + if (ds.isStreaming) { + throw new IllegalArgumentException("Observation does not support streaming Datasets") + } + register(ds.sparkSession) + ds.observe(name, expr, exprs: _*) + } + + /** + * Waits for the first action on the observed dataset to complete and returns true. + * The result is then available through the get method. + * This method times out after the given amount of time returning false. + * + * @param time timeout + * @param unit timeout time unit + * @return true if action completed within timeout, false otherwise + * @throws InterruptedException interrupted while waiting + */ + def waitCompleted(time: Long, unit: TimeUnit): Boolean = waitCompleted(Some(unit.toMillis(time))) + + /** + * Get the observed metrics. This waits until the observed dataset finishes its first action. + * If you want to wait for the result and provide a timeout, use [[waitCompleted]]. Only the + * result of the first action is available. Subsequent actions do not modify the result. + * + * @return the observed metrics as a [[Row]] + * @throws InterruptedException interrupted while waiting + */ + def get: Row = { + assert(waitCompleted(None), "waitCompleted without timeout returned false") + row.get + } + + private def waitCompleted(millis: Option[Long]): Boolean = { Review comment: Actually `Object.wait(0)` means wait forever. ########## File path: sql/core/src/main/scala/org/apache/spark/sql/Observation.scala ########## @@ -0,0 +1,170 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql + +import java.util.UUID +import java.util.concurrent.TimeUnit + +import org.apache.spark.sql.execution.QueryExecution +import org.apache.spark.sql.util.QueryExecutionListener + +/** + * Helper class to simplify usage of [[Dataset.observe(String, Column, Column*)]]: + * + * {{{ + * // Observe row count (rows) and highest id (maxid) in the Dataset while writing it + * val observation = Observation("my_metrics") + * val observed_ds = ds.observe(observation, count(lit(1)).as("rows"), max($"id").as("maxid")) + * observed_ds.write.parquet("ds.parquet") + * val metrics = observation.get + * }}} + * + * This collects the metrics while the first action is executed on the obseerved dataset. Subsequent + * actions do not modify the metrics returned by [[org.apache.spark.sql.Observation.get]]. Retrieval + * of the metric via [[org.apache.spark.sql.Observation.get]] blocks until the first action has + * finished and metrics become available. You can add a timeout to that blocking via + * [[org.apache.spark.sql.Observation.waitCompleted]]: + * + * {{{ + * if (observation.waitCompleted(100, TimeUnit.MILLISECONDS)) { + * observation.get + * } + * }}} + * + * This class does not support streaming datasets. + * + * @param name name of the metric + * @since 3.2.0 + */ +class Observation(name: String) { + + private val listener: ObservationListener = ObservationListener(this) + + private var sparkSession: Option[SparkSession] = None + + @volatile private var row: Option[Row] = None + + /** + * Attaches this observation to the given [[Dataset]] to observe aggregation expressions. + * + * @param ds dataset + * @param expr first aggregation expression + * @param exprs more aggregation expressions + * @tparam T dataset type + * @return observed dataset + * @throws IllegalArgumentException If this is a streaming Dataset (ds.isStreaming == true) + */ + def on[T](ds: Dataset[T], expr: Column, exprs: Column*): Dataset[T] = { + if (ds.isStreaming) { + throw new IllegalArgumentException("Observation does not support streaming Datasets") + } + register(ds.sparkSession) + ds.observe(name, expr, exprs: _*) + } + + /** + * Waits for the first action on the observed dataset to complete and returns true. + * The result is then available through the get method. + * This method times out after the given amount of time returning false. + * + * @param time timeout + * @param unit timeout time unit + * @return true if action completed within timeout, false otherwise + * @throws InterruptedException interrupted while waiting + */ + def waitCompleted(time: Long, unit: TimeUnit): Boolean = waitCompleted(Some(unit.toMillis(time))) + + /** + * Get the observed metrics. This waits until the observed dataset finishes its first action. + * If you want to wait for the result and provide a timeout, use [[waitCompleted]]. Only the + * result of the first action is available. Subsequent actions do not modify the result. + * + * @return the observed metrics as a [[Row]] + * @throws InterruptedException interrupted while waiting + */ + def get: Row = { + assert(waitCompleted(None), "waitCompleted without timeout returned false") + row.get + } + + private def waitCompleted(millis: Option[Long]): Boolean = { + synchronized { + if (row.isEmpty) { + if (millis.isDefined) { + this.wait(millis.get) + } else { + this.wait() + } + } + row.isDefined + } + } + + private def register(sparkSession: SparkSession): Unit = { + // makes this class thread-safe: + // only the first thread entering this block can set sparkSession + // all other threads will see the exception, because it is only allowed to do this once + synchronized { + if (this.sparkSession.isDefined) { + throw new IllegalStateException("An Observation can be used with a Dataset only once") + } + this.sparkSession = Some(sparkSession) + } + + sparkSession.listenerManager.register(listener) + } + + private def unregister(): Unit = { + this.sparkSession.foreach(_.listenerManager.unregister(listener)) + } + + private[spark] def onFinish(qe: QueryExecution): Unit = { + synchronized { + this.row = qe.observedMetrics.get(name) Review comment: It's possible that two DataFrame actions happen within a short time and we receive the second execution success event before we unregister the listener. It's safer to do ``` if (this.row.isEmpty) { this.row = qe.observedMetrics.get(name) assert(this.row.isDefined, "No metric provided by QueryExecutionListener") } this.notifyAll() ``` ########## File path: sql/core/src/main/scala/org/apache/spark/sql/Observation.scala ########## @@ -0,0 +1,139 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql + +import java.util.UUID +import java.util.concurrent.TimeUnit + +import org.apache.spark.sql.execution.QueryExecution +import org.apache.spark.sql.util.QueryExecutionListener + +/** + * Not thread-safe. + * @param name + * @param sparkSession + */ +class Observation(name: String) { + + private val listener: ObservationListener = ObservationListener(this) + + private var sparkSession: Option[SparkSession] = None + + @volatile private var row: Option[Row] = None + + /** + * Attach this observation to the given Dataset. + * + * @param ds dataset + * @tparam T dataset type + * @return observed dataset + */ + def on[T](ds: Dataset[T])(expr: Column, exprs: Column*): Dataset[T] = { + if (ds.isStreaming) { + throw new IllegalArgumentException("Observation does not support streaming Datasets") + } + register(ds.sparkSession) + ds.observe(name, expr, exprs: _*) + } + + /** + * Wait for the first action on the observed dataset to complete and returns true. + * This method times out after the given amount of time and returns false. + * + * @param time timeout + * @param unit timeout time unit + * @return true if action complete within timeout, false on timeout + */ + def waitCompleted(time: Long, unit: TimeUnit): Boolean = waitCompleted(Some(time), unit) + + /** + * Get the observation results. This waits until the observed dataset finishes its first action. + * If you want to wait for the result and provide a timeout, use waitCompleted. + * Only the result of the first action is available. Subsequent actions do not modify the result. + */ + def get: Row = { + assert(waitCompleted(None, TimeUnit.SECONDS), "waitCompleted without timeout returned false") + row.get + } + + private def waitCompleted(time: Option[Long], unit: TimeUnit): Boolean = { + synchronized { + if (row.isEmpty) { + if (time.isDefined) { + this.wait(unit.toMillis(time.get)) + } else { + this.wait() + } + } + row.isDefined + } + } + + private def register(sparkSession: SparkSession): Unit = { + // makes this class thread-safe: + // only the first thread entering this block can set sparkSession + // all other threads will see the exception, because it is only allowed to do this once + synchronized { + if (this.sparkSession.isDefined) { + throw new IllegalStateException("An Observation can be used with a Dataset only once") + } + this.sparkSession = Some(sparkSession) + } + + sparkSession.listenerManager.register(listener) + } + + private def unregister(): Unit = { + this.sparkSession.foreach(_.listenerManager.unregister(listener)) + } + + private[spark] def onFinish(qe: QueryExecution): Unit = { + synchronized { + this.row = qe.observedMetrics.get(name) + assert(this.row.isDefined, "No metric provided by QueryExecutionListener") + this.notifyAll() + } + unregister() Review comment: > two threads are involved in updating and accessing sparkSession, but not concurrently. My understanding is that, even if two things can't happen concurrently, we still need to add `volatile` or do lock as long as the update and read of the value happens in different threads. Otherwise, it's non-deterministic that how soon other threads can see the value update. ########## File path: sql/core/src/main/scala/org/apache/spark/sql/Observation.scala ########## @@ -0,0 +1,170 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql + +import java.util.UUID +import java.util.concurrent.TimeUnit + +import org.apache.spark.sql.execution.QueryExecution +import org.apache.spark.sql.util.QueryExecutionListener + +/** + * Helper class to simplify usage of [[Dataset.observe(String, Column, Column*)]]: + * + * {{{ + * // Observe row count (rows) and highest id (maxid) in the Dataset while writing it + * val observation = Observation("my_metrics") + * val observed_ds = ds.observe(observation, count(lit(1)).as("rows"), max($"id").as("maxid")) + * observed_ds.write.parquet("ds.parquet") + * val metrics = observation.get + * }}} + * + * This collects the metrics while the first action is executed on the obseerved dataset. Subsequent + * actions do not modify the metrics returned by [[org.apache.spark.sql.Observation.get]]. Retrieval + * of the metric via [[org.apache.spark.sql.Observation.get]] blocks until the first action has + * finished and metrics become available. You can add a timeout to that blocking via + * [[org.apache.spark.sql.Observation.waitCompleted]]: + * + * {{{ + * if (observation.waitCompleted(100, TimeUnit.MILLISECONDS)) { + * observation.get + * } + * }}} + * + * This class does not support streaming datasets. + * + * @param name name of the metric + * @since 3.2.0 + */ +class Observation(name: String) { Review comment: That's a good point! We can simply leverage the standard `Future` API to represent this async result. ########## File path: sql/core/src/main/scala/org/apache/spark/sql/Observation.scala ########## @@ -0,0 +1,170 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql + +import java.util.UUID +import java.util.concurrent.TimeUnit + +import org.apache.spark.sql.execution.QueryExecution +import org.apache.spark.sql.util.QueryExecutionListener + +/** + * Helper class to simplify usage of [[Dataset.observe(String, Column, Column*)]]: + * + * {{{ + * // Observe row count (rows) and highest id (maxid) in the Dataset while writing it + * val observation = Observation("my_metrics") + * val observed_ds = ds.observe(observation, count(lit(1)).as("rows"), max($"id").as("maxid")) + * observed_ds.write.parquet("ds.parquet") + * val metrics = observation.get + * }}} + * + * This collects the metrics while the first action is executed on the obseerved dataset. Subsequent + * actions do not modify the metrics returned by [[org.apache.spark.sql.Observation.get]]. Retrieval + * of the metric via [[org.apache.spark.sql.Observation.get]] blocks until the first action has + * finished and metrics become available. You can add a timeout to that blocking via + * [[org.apache.spark.sql.Observation.waitCompleted]]: + * + * {{{ + * if (observation.waitCompleted(100, TimeUnit.MILLISECONDS)) { + * observation.get + * } + * }}} + * + * This class does not support streaming datasets. + * + * @param name name of the metric + * @since 3.2.0 + */ +class Observation(name: String) { Review comment: We can still keep the `get` API, but returning `Future` should be a more standard way for users to do async than `waitCompleted`. -- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org
