Github user Ben-Zvi commented on a diff in the pull request: https://github.com/apache/drill/pull/717#discussion_r98805535 --- Diff: exec/java-exec/src/main/java/org/apache/drill/exec/physical/impl/xsort/managed/ExternalSortBatch.java --- @@ -0,0 +1,1321 @@ +/* + * 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.drill.exec.physical.impl.xsort.managed; + +import java.io.IOException; +import java.util.Collection; +import java.util.LinkedList; +import java.util.List; + +import org.apache.drill.common.AutoCloseables; +import org.apache.drill.common.config.DrillConfig; +import org.apache.drill.common.exceptions.UserException; +import org.apache.drill.exec.ExecConstants; +import org.apache.drill.exec.exception.OutOfMemoryException; +import org.apache.drill.exec.exception.SchemaChangeException; +import org.apache.drill.exec.memory.BufferAllocator; +import org.apache.drill.exec.ops.FragmentContext; +import org.apache.drill.exec.ops.MetricDef; +import org.apache.drill.exec.physical.config.ExternalSort; +import org.apache.drill.exec.physical.impl.sort.RecordBatchData; +import org.apache.drill.exec.physical.impl.xsort.MSortTemplate; +import org.apache.drill.exec.physical.impl.xsort.SingleBatchSorter; +import org.apache.drill.exec.physical.impl.xsort.managed.BatchGroup.InputBatch; +import org.apache.drill.exec.record.AbstractRecordBatch; +import org.apache.drill.exec.record.BatchSchema; +import org.apache.drill.exec.record.BatchSchema.SelectionVectorMode; +import org.apache.drill.exec.record.RecordBatch; +import org.apache.drill.exec.record.SchemaUtil; +import org.apache.drill.exec.record.VectorContainer; +import org.apache.drill.exec.record.VectorWrapper; +import org.apache.drill.exec.record.WritableBatch; +import org.apache.drill.exec.record.selection.SelectionVector2; +import org.apache.drill.exec.record.selection.SelectionVector4; +import org.apache.drill.exec.testing.ControlsInjector; +import org.apache.drill.exec.testing.ControlsInjectorFactory; +import org.apache.drill.exec.vector.ValueVector; +import org.apache.drill.exec.vector.complex.AbstractContainerVector; + +import com.google.common.collect.Lists; + +/** + * External sort batch: a sort batch which can spill to disk in + * order to operate within a defined memory footprint. + * <p> + * <h4>Basic Operation</h4> + * The operator has three key phases: + * <p> + * <ul> + * <li>The load phase in which batches are read from upstream.</li> + * <li>The merge phase in which spilled batches are combined to + * reduce the number of files below the configured limit. (Best + * practice is to configure the system to avoid this phase.) + * <li>The delivery phase in which batches are combined to produce + * the final output.</li> + * </ul> + * During the load phase: + * <p> + * <ul> + * <li>The incoming (upstream) operator provides a series of batches.</li> + * <li>This operator sorts each batch, and accumulates them in an in-memory + * buffer.</li> + * <li>If the in-memory buffer becomes too large, this operator selects + * a subset of the buffered batches to spill.</li> --- End diff -- A design comment: This "subset" (i.e., half the batches) -- why not spill __all__ the in-memory batches ? This would likely reduce the IO and the number of runs, which is good. There is no benefit to maintaining half the batches in memory (unlike Hash Aggregation), accept some cases where the end of the input is very near.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---