xuanyuanking commented on a change in pull request #33187: URL: https://github.com/apache/spark/pull/33187#discussion_r666022602
########## File path: sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/RocksDBStateStoreProvider.scala ########## @@ -0,0 +1,328 @@ +/* + * 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.execution.streaming.state + +import java.io._ + +import org.apache.hadoop.conf.Configuration + +import org.apache.spark.{SparkConf, SparkEnv} +import org.apache.spark.internal.Logging +import org.apache.spark.sql.catalyst.expressions.UnsafeRow +import org.apache.spark.sql.types.StructType +import org.apache.spark.unsafe.Platform +import org.apache.spark.util.Utils + +private[state] class RocksDBStateStoreProvider + extends StateStoreProvider with Logging with Closeable { + import RocksDBStateStoreProvider._ + + class RocksDBStateStore(lastVersion: Long) extends StateStore { + /** Trait and classes representing the internal state of the store */ + trait STATE + case object UPDATING extends STATE + case object COMMITTED extends STATE + case object ABORTED extends STATE + + @volatile private var state: STATE = UPDATING + @volatile private var isValidated = false + + override def id: StateStoreId = RocksDBStateStoreProvider.this.stateStoreId + + override def version: Long = lastVersion + + override def get(key: UnsafeRow): UnsafeRow = { + verify(key != null, "Key cannot be null") + val value = encoder.decodeValue(rocksDB.get(encoder.encode(key))) + if (!isValidated && value != null) { + StateStoreProvider.validateStateRowFormat( + key, keySchema, value, valueSchema, storeConf) + isValidated = true + } + value + } + + override def put(key: UnsafeRow, value: UnsafeRow): Unit = { + verify(state == UPDATING, "Cannot put after already committed or aborted") + verify(key != null, "Key cannot be null") + require(value != null, "Cannot put a null value") + logDebug(s"Storing $key => $value") + rocksDB.put(encoder.encode(key), encoder.encode(value)) + } + + override def remove(key: UnsafeRow): Unit = { + verify(state == UPDATING, "Cannot remove after already committed or aborted") + verify(key != null, "Key cannot be null") + rocksDB.remove(encoder.encode(key)) + } + + override def getRange( + start: Option[UnsafeRow], + end: Option[UnsafeRow]): Iterator[UnsafeRowPair] = { + verify(state == UPDATING, "Cannot call getRange() after already committed or aborted") + iterator() + } + + override def iterator(): Iterator[UnsafeRowPair] = { + rocksDB.iterator().map { kv => + val rowPair = encoder.decode(kv) + if (!isValidated && rowPair.value != null) { + StateStoreProvider.validateStateRowFormat( + rowPair.key, keySchema, rowPair.value, valueSchema, storeConf) + isValidated = true + } + rowPair + } + } + + override def commit(): Long = synchronized { + verify(state == UPDATING, "Cannot commit after already committed or aborted") + val newVersion = rocksDB.commit() + state = COMMITTED + logInfo(s"Committed $newVersion for $id") + newVersion + } + + override def abort(): Unit = { + verify(state == UPDATING || state == ABORTED, "Cannot abort after already committed") + logInfo(s"Aborting ${version + 1} for $id") + rocksDB.rollback() + state = ABORTED + } + + override def metrics: StateStoreMetrics = { + val rocksDBMetrics = rocksDB.metrics + def commitLatencyMs(typ: String): Long = rocksDBMetrics.lastCommitLatencyMs.getOrElse(typ, 0L) + def avgNativeOpsLatencyMs(typ: String): Long = { + rocksDBMetrics.nativeOpsLatencyMicros.get(typ).map(_.avg).getOrElse(0.0).toLong + } + + val stateStoreCustomMetrics = Map[StateStoreCustomMetric, Long]( + CUSTOM_METRIC_SST_FILE_SIZE -> rocksDBMetrics.totalSSTFilesBytes, + CUSTOM_METRIC_GET_TIME -> avgNativeOpsLatencyMs("get"), + CUSTOM_METRIC_PUT_TIME -> avgNativeOpsLatencyMs("put"), + CUSTOM_METRIC_WRITEBATCH_TIME -> commitLatencyMs("writeBatch"), + CUSTOM_METRIC_FLUSH_TIME -> commitLatencyMs("flush"), + CUSTOM_METRIC_PAUSE_TIME -> commitLatencyMs("pause"), + CUSTOM_METRIC_CHECKPOINT_TIME -> commitLatencyMs("checkpoint"), + CUSTOM_METRIC_FILESYNC_TIME -> commitLatencyMs("fileSync"), + CUSTOM_METRIC_BYTES_COPIED -> rocksDBMetrics.bytesCopied, + CUSTOM_METRIC_FILES_COPIED -> rocksDBMetrics.filesCopied, + CUSTOM_METRIC_FILES_REUSED -> rocksDBMetrics.filesReused + ) ++ rocksDBMetrics.zipFileBytesUncompressed.map(bytes => + Map(CUSTOM_METRIC_ZIP_FILE_BYTES_UNCOMPRESSED -> bytes)).getOrElse(Map()) + + StateStoreMetrics( + rocksDBMetrics.numUncommittedKeys, + rocksDBMetrics.memUsageBytes, + stateStoreCustomMetrics) + } + + override def hasCommitted: Boolean = state == COMMITTED + + override def toString: String = { + s"RocksDBStateStore[id=(op=${id.operatorId},part=${id.partitionId})," + + s"dir=${id.storeCheckpointLocation()}]" + } + + /** Return the [[RocksDB]] instance in this store. This is exposed mainly for testing. */ + def dbInstance(): RocksDB = rocksDB + } + + override def init( + stateStoreId: StateStoreId, + keySchema: StructType, + valueSchema: StructType, + indexOrdinal: Option[Int], + storeConf: StateStoreConf, + hadoopConf: Configuration): Unit = { + this.stateStoreId_ = stateStoreId + this.keySchema = keySchema + this.valueSchema = valueSchema + this.storeConf = storeConf + this.hadoopConf = hadoopConf + rocksDB // lazy initialization + } + + override def stateStoreId: StateStoreId = stateStoreId_ + + override def getStore(version: Long): StateStore = { + require(version >= 0, "Version cannot be less than 0") + rocksDB.load(version) + new RocksDBStateStore(version) + } + + override def doMaintenance(): Unit = { + rocksDB.cleanup() + } + + override def close(): Unit = { + rocksDB.close() + } + + override def supportedCustomMetrics: Seq[StateStoreCustomMetric] = ALL_CUSTOM_METRICS + + private[state] def latestVersion: Long = rocksDB.getLatestVersion() + + /** Internal fields and methods */ + + @volatile private var stateStoreId_ : StateStoreId = _ + @volatile private var keySchema: StructType = _ + @volatile private var valueSchema: StructType = _ + @volatile private var storeConf: StateStoreConf = _ + @volatile private var hadoopConf: Configuration = _ + + private[sql] lazy val rocksDB = { + val dfsRootDir = stateStoreId.storeCheckpointLocation().toString + val storeIdStr = s"StateStoreId(opId=${stateStoreId.operatorId}," + + s"partId=${stateStoreId.partitionId},name=${stateStoreId.storeName})" + val sparkConf = Option(SparkEnv.get).map(_.conf).getOrElse(new SparkConf) + val localRootDir = Utils.createTempDir(Utils.getLocalDir(sparkConf), storeIdStr) + new RocksDB(dfsRootDir, RocksDBConf(storeConf), localRootDir, hadoopConf, storeIdStr) + } + + private lazy val encoder = new StateEncoder + + private def verify(condition: => Boolean, msg: String): Unit = { + if (!condition) { throw new IllegalStateException(msg) } + } + + /** + * Encodes/decodes UnsafeRows to versioned byte arrays. + * It uses the first byte of the generated byte array to store the version the describes how the + * row is encoded in the rest of the byte array. Currently, the default version is 0, + * + * VERSION 0: [ VERSION (1 byte) | ROW (N bytes) ] + * The bytes of a UnsafeRow is written unmodified to starting from offset 1 + * (offset 0 is the version byte of value 0). That is, if the unsafe row has N bytes, + * then the generated array byte will be N+1 bytes. + */ + class StateEncoder { + import RocksDBStateStoreProvider._ + + // Reusable objects + private val keyRow = new UnsafeRow(keySchema.size) + private val valueRow = new UnsafeRow(valueSchema.size) + private val rowTuple = new UnsafeRowPair() + + /** + * Encode the UnsafeRow of N bytes as a N+1 byte array. + * @note This creates a new byte array and memcopies the UnsafeRow to the new array. + */ + def encode(row: UnsafeRow): Array[Byte] = { + val bytesToEncode = row.getBytes + val encodedBytes = new Array[Byte](bytesToEncode.length + STATE_ENCODING_NUM_VERSION_BYTES) + Platform.putByte(encodedBytes, Platform.BYTE_ARRAY_OFFSET, STATE_ENCODING_VERSION) + // Platform.BYTE_ARRAY_OFFSET is the recommended way to memcopy b/w byte arrays. See Platform. + Platform.copyMemory( + bytesToEncode, Platform.BYTE_ARRAY_OFFSET, + encodedBytes, Platform.BYTE_ARRAY_OFFSET + STATE_ENCODING_NUM_VERSION_BYTES, + bytesToEncode.length) + encodedBytes + } + + /** + * Decode byte array for a key to a UnsafeRow. + * @note The UnsafeRow returned is reused across calls, and the UnsafeRow just points to + * the given byte array. + */ + def decodeKey(keyBytes: Array[Byte]): UnsafeRow = { + if (keyBytes != null) { + // Platform.BYTE_ARRAY_OFFSET is the recommended way refer to the 1st offset. See Platform. + keyRow.pointTo( + keyBytes, + Platform.BYTE_ARRAY_OFFSET + STATE_ENCODING_NUM_VERSION_BYTES, + keyBytes.length - STATE_ENCODING_NUM_VERSION_BYTES) + keyRow + } else null + } + + /** + * Decode byte array for a value to a UnsafeRow. + * + * @note The UnsafeRow returned is reused across calls, and the UnsafeRow just points to + * the given byte array. + */ + def decodeValue(valueBytes: Array[Byte]): UnsafeRow = { + if (valueBytes != null) { + // Platform.BYTE_ARRAY_OFFSET is the recommended way refer to the 1st offset. See Platform. + valueRow.pointTo( + valueBytes, + Platform.BYTE_ARRAY_OFFSET + STATE_ENCODING_NUM_VERSION_BYTES, + valueBytes.size - STATE_ENCODING_NUM_VERSION_BYTES) + valueRow + } else null + } + + /** + * Decode pair of key-value byte arrays in a pair of key-value UnsafeRows. + * + * @note The UnsafeRow returned is reused across calls, and the UnsafeRow just points to + * the given byte array. + */ + def decode(byteArrayTuple: ByteArrayPair): UnsafeRowPair = { + rowTuple.withRows(decodeKey(byteArrayTuple.key), decodeValue(byteArrayTuple.value)) + } + } +} + +object RocksDBStateStoreProvider { + // Version as a single byte that specifies the encoding of the row data in RocksDB + val STATE_ENCODING_NUM_VERSION_BYTES = 1 + val STATE_ENCODING_VERSION: Byte = 0 + + // Native operation latencies report as latency per 1000 calls + // as SQLMetrics support ms latency whereas RocksDB reports it in microseconds. + val CUSTOM_METRIC_GET_TIME = StateStoreCustomTimingMetric( + "rocksdbGetLatency", "RocksDB: avg get latency (per 1000 calls)") + val CUSTOM_METRIC_PUT_TIME = StateStoreCustomTimingMetric( + "rocksdbPutLatency", "RocksDB: avg put latency (per 1000 calls)") + + // Commit latency detailed breakdown + val CUSTOM_METRIC_WRITEBATCH_TIME = StateStoreCustomTimingMetric( + "rocksdbCommitWriteBatchLatency", "RocksDB: commit - write batch time") + val CUSTOM_METRIC_FLUSH_TIME = StateStoreCustomTimingMetric( + "rocksdbCommitFlushLatency", "RocksDB: commit - flush time") + val CUSTOM_METRIC_PAUSE_TIME = StateStoreCustomTimingMetric( + "rocksdbCommitPauseLatency", "RocksDB: commit - pause bg time") + val CUSTOM_METRIC_CHECKPOINT_TIME = StateStoreCustomTimingMetric( + "rocksdbCommitCheckpointLatency", "RocksDB: commit - checkpoint time") + val CUSTOM_METRIC_FILESYNC_TIME = StateStoreCustomTimingMetric( + "rocksdbCommitPauseBgTime", "RocksDB: commit - file sync time") Review comment: Yea, changed to rocksdbFileSyncTime ########## File path: sql/core/src/test/scala/org/apache/spark/sql/execution/streaming/state/RocksDBStateStoreSuite.scala ########## @@ -0,0 +1,152 @@ +/* + * 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.execution.streaming.state + +import java.util.UUID + +import scala.util.Random + +import org.apache.hadoop.conf.Configuration +import org.scalatest.BeforeAndAfter + +import org.apache.spark.SparkConf +import org.apache.spark.sql.LocalSparkSession.withSparkSession +import org.apache.spark.sql.SparkSession +import org.apache.spark.sql.execution.streaming.StatefulOperatorStateInfo +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.types._ +import org.apache.spark.unsafe.Platform +import org.apache.spark.util.Utils + +class RocksDBStateStoreSuite extends StateStoreSuiteBase[RocksDBStateStoreProvider] + with BeforeAndAfter { + + import StateStoreTestsHelper._ + + test("version encoding") { + import RocksDBStateStoreProvider._ + + val provider = newStoreProvider() + val store = provider.getStore(0) + val keyRow = stringToRow("a") + val valueRow = intToRow(1) + store.put(keyRow, valueRow) + val iter = provider.rocksDB.iterator() + assert(iter.hasNext) + val kv = iter.next() + + // Verify the version encoded in first byte of the key and value byte arrays + assert(Platform.getByte(kv.key, Platform.BYTE_ARRAY_OFFSET) === STATE_ENCODING_VERSION) + assert(Platform.getByte(kv.value, Platform.BYTE_ARRAY_OFFSET) === STATE_ENCODING_VERSION) + } + + test("RocksDB confs are passed correctly from SparkSession to db instance") { + val sparkConf = new SparkConf().setMaster("local").setAppName(this.getClass.getSimpleName) + withSparkSession(SparkSession.builder.config(sparkConf).getOrCreate()) { spark => + // Set the session confs that should be passed into RocksDB + val testConfs = Seq( + ("spark.sql.streaming.stateStore.providerClass", + classOf[RocksDBStateStoreProvider].getName), + (RocksDBConf.ROCKSDB_CONF_NAME_PREFIX + ".compactOnCommit", "true"), + (RocksDBConf.ROCKSDB_CONF_NAME_PREFIX + ".lockAcquireTimeoutMs", "10") + ) + testConfs.foreach { case (k, v) => spark.conf.set(k, v) } + + // Prepare test objects for running task on state store + val testRDD = spark.sparkContext.makeRDD[String](Seq("a"), 1) + val testSchema = StructType(Seq(StructField("key", StringType, true))) + val testStateInfo = StatefulOperatorStateInfo( + checkpointLocation = Utils.createTempDir().getAbsolutePath, + queryRunId = UUID.randomUUID, operatorId = 0, storeVersion = 0, numPartitions = 5) + + // Create state store in a task and get the RocksDBConf from the instantiated RocksDB instance + val rocksDBConfInTask: RocksDBConf = testRDD.mapPartitionsWithStateStore[RocksDBConf]( + spark.sqlContext, testStateInfo, testSchema, testSchema, None) { + (store: StateStore, _: Iterator[String]) => + // Use reflection to get RockDB instance Review comment: Thanks, done in the next commit. -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
