Mailing-List: contact dev-help@drill.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@drill.apache.org
From: paul-rogers <git@git.apache.org>
To: dev@drill.apache.org
Reply-To: dev@drill.apache.org
References: <git-pr-729-drill@git.apache.org>
In-Reply-To: <git-pr-729-drill@git.apache.org>
Subject: [GitHub] drill pull request #729: Drill 1328: Support table statistics for Parquet
Content-Type: text/plain
Message-Id: <20170212015123.2C84EDFD8C@git1-us-west.apache.org>
Date: Sun, 12 Feb 2017 01:51:23 +0000 (UTC)
archived-at: Sun, 12 Feb 2017 01:51:27 -0000

Github user paul-rogers commented on a diff in the pull request:

    https://github.com/apache/drill/pull/729#discussion_r100681030
  
    --- Diff: exec/java-exec/src/main/java/org/apache/drill/exec/physical/impl/statistics/StatisticsMergeBatch.java ---
    @@ -0,0 +1,532 @@
    +/**
    + * 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
    + * <p/>
    + * http://www.apache.org/licenses/LICENSE-2.0
    + * <p/>
    + * 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.statistics;
    +
    +import com.google.common.collect.Lists;
    +import com.google.common.collect.Maps;
    +import org.apache.drill.common.expression.LogicalExpression;
    +import org.apache.drill.common.expression.ErrorCollector;
    +import org.apache.drill.common.expression.ErrorCollectorImpl;
    +import org.apache.drill.common.expression.FunctionCallFactory;
    +import org.apache.drill.common.expression.SchemaPath;
    +import org.apache.drill.common.expression.ValueExpressions;
    +import org.apache.drill.common.types.TypeProtos;
    +import org.apache.drill.common.types.TypeProtos.MajorType;
    +import org.apache.drill.exec.exception.OutOfMemoryException;
    +import org.apache.drill.exec.exception.SchemaChangeException;
    +import org.apache.drill.exec.expr.ExpressionTreeMaterializer;
    +import org.apache.drill.exec.expr.TypeHelper;
    +import org.apache.drill.exec.expr.holders.BigIntHolder;
    +import org.apache.drill.exec.expr.holders.NullableFloat8Holder;
    +import org.apache.drill.exec.expr.holders.ObjectHolder;
    +import org.apache.drill.exec.expr.holders.ValueHolder;
    +import org.apache.drill.exec.ops.FragmentContext;
    +import org.apache.drill.exec.physical.config.StatisticsMerge;
    +import org.apache.drill.exec.record.AbstractSingleRecordBatch;
    +import org.apache.drill.exec.record.MaterializedField;
    +import org.apache.drill.exec.record.RecordBatch;
    +import org.apache.drill.exec.record.TransferPair;
    +import org.apache.drill.exec.record.VectorContainer;
    +import org.apache.drill.exec.record.VectorWrapper;
    +import org.apache.drill.exec.vector.ValueVector;
    +import org.apache.drill.exec.vector.DateVector;
    +import org.apache.drill.exec.vector.BigIntVector;
    +import org.apache.drill.exec.vector.NullableBigIntVector;
    +import org.apache.drill.exec.vector.NullableFloat8Vector;
    +import org.apache.drill.exec.vector.NullableVarBinaryVector;
    +import org.apache.drill.exec.vector.VarCharVector;
    +import org.apache.drill.exec.vector.complex.MapVector;
    +
    +import java.io.ByteArrayInputStream;
    +import java.io.DataInputStream;
    +import java.io.IOException;
    +import java.util.List;
    +import java.util.GregorianCalendar;
    +import java.util.HashMap;
    +import java.util.Map;
    +import java.util.TimeZone;
    +
    +import com.clearspring.analytics.stream.cardinality.HyperLogLog;
    +
    +public class StatisticsMergeBatch extends AbstractSingleRecordBatch<StatisticsMerge> {
    +  private static final org.slf4j.Logger logger = org.slf4j.LoggerFactory.getLogger(StatisticsMergeBatch.class);
    +  private Map<String, String> functions;
    +  private boolean first = true;
    +  private boolean finished = false;
    +  private int schema = 0;
    +  private int recordCount = 0;
    +  private List<String> keyList = null;
    +  private Map<MaterializedField, ValueVector> dataSrcVecMap = null;
    +  // Map of non-map fields to VV in the incoming schema
    +  private Map<MaterializedField, ValueVector> copySrcVecMap = null;
    +  private Map<String, Map<String, ValueHolder>> aggregationMap = null;
    +  public StatisticsMergeBatch(StatisticsMerge popConfig, RecordBatch incoming,
    +                              FragmentContext context) throws OutOfMemoryException {
    +    super(popConfig, context, incoming);
    +    this.functions = new HashMap<>();
    +    this.aggregationMap = new HashMap<>();
    +
    +    /*for (String key : popConfig.getFunctions()) {
    +      aggregationMap.put(key, new HashMap<String, ValueHolder>());
    +      if (key.equalsIgnoreCase("statcount") || key.equalsIgnoreCase("nonnullstatcount")) {
    +        functions.put(key, "sum");
    +      } else if (key.equalsIgnoreCase("hll")) {
    +        functions.put(key, "hll_merge");
    +      } else if (key.equalsIgnoreCase("sum_width")) {
    +        functions.put(key, "avg_width");
    +      }
    +    }*/
    +    for (String key : popConfig.getFunctions()) {
    +      if (key.equalsIgnoreCase("sum_width")) {
    +        functions.put(key, "avg_width");
    +      } else if (key.equalsIgnoreCase("hll")) {
    +        functions.put(key, "hll_merge");
    +      } else {
    +        functions.put(key, key);
    +      }
    +      aggregationMap.put(functions.get(key), new HashMap<String, ValueHolder>());
    +    }
    +  }
    +
    +  private void createKeyColumn(String name, LogicalExpression expr, Map<MaterializedField, ValueVector> parentMap)
    +      throws SchemaChangeException {
    +    ErrorCollector collector = new ErrorCollectorImpl();
    +
    +    LogicalExpression mle = ExpressionTreeMaterializer.materialize(expr, incoming, collector,
    +        context.getFunctionRegistry());
    +
    +    MaterializedField outputField = MaterializedField.create(name, mle.getMajorType());
    +    ValueVector vector = TypeHelper.getNewVector(outputField, oContext.getAllocator());
    +    container.add(vector);
    +
    +    if (collector.hasErrors()) {
    +      throw new SchemaChangeException("Failure while materializing expression. "
    +          + collector.toErrorString());
    +    }
    +    parentMap.put(outputField, vector);
    +  }
    +
    +  private ValueVector addMapVector(String name, MapVector parent, LogicalExpression expr) throws SchemaChangeException {
    +    ErrorCollector collector = new ErrorCollectorImpl();
    +
    +    LogicalExpression mle = ExpressionTreeMaterializer.materialize(expr, incoming, collector,
    +        context.getFunctionRegistry());
    +
    +    Class<? extends ValueVector> vvc =
    +        TypeHelper.getValueVectorClass(mle.getMajorType().getMinorType(), mle.getMajorType().getMode());
    +    ValueVector vector = parent.addOrGet(name, mle.getMajorType(), vvc);
    +
    +    if (collector.hasErrors()) {
    +      throw new SchemaChangeException("Failure while materializing expression. "
    +          + collector.toErrorString());
    +    }
    +    return vector;
    +  }
    +
    +  /**
    +   * Identify the list of fields within a map which are unpivoted as columns in output
    +   */
    +  private void buildKeyList() {
    +    List<String> lastMapKeyList = null;
    +    for (VectorWrapper<?> vw : incoming) {
    +      if (vw.getField().getType().getMinorType() != TypeProtos.MinorType.MAP) {
    +        continue;
    +      }
    +
    +      keyList = Lists.newArrayList();
    +
    +      for (ValueVector vv : vw.getValueVector()) {
    +        keyList.add(vv.getField().getLastName());
    +      }
    +
    +      if (lastMapKeyList == null) {
    +        lastMapKeyList = keyList;
    +      } else {
    +        if (keyList.size() != lastMapKeyList.size() || !lastMapKeyList.containsAll(keyList)) {
    +          throw new UnsupportedOperationException("Maps have different fields");
    +        }
    +      }
    +    }
    +  }
    +
    +  private void buildOutputContainer() throws SchemaChangeException {
    +    dataSrcVecMap = Maps.newHashMap();
    +    copySrcVecMap = Maps.newHashMap();
    +    MajorType mt = null;
    +
    +    ErrorCollector collector = new ErrorCollectorImpl();
    +    GregorianCalendar calendar = new GregorianCalendar(TimeZone.getTimeZone("UTC"));
    +
    +    calendar.setTimeInMillis(System.currentTimeMillis());
    +    createKeyColumn("schema", ValueExpressions.getBigInt(schema++), copySrcVecMap);
    +    createKeyColumn("computed", ValueExpressions.getDate(calendar), copySrcVecMap);
    +
    +    for (VectorWrapper<?> vw : incoming) {
    +      addVectorToOutgoingContainer(vw.getField().getLastName(), vw, collector);
    +      /*MaterializedField ds = vw.getField();
    +      String field = vw.getField().getLastName();
    +      // Input map vector
    +      MapVector mapVector = (MapVector) vw.getValueVector();
    +      assert mapVector.getPrimitiveVectors().size() > 0;
    +      // Proceed to create output map vector with same name e.g. statcount etc.
    +      mt = mapVector.getField().getType();
    +      MaterializedField mf = MaterializedField.create(functions.get(field), mt);
    +      assert !dataSrcVecMap.containsKey(mf);
    +      ValueVector vector = TypeHelper.getNewVector(mf, oContext.getAllocator());
    +      container.add(vector);
    +      MapVector outputMapVector = (MapVector) vector;
    +
    +      for (ValueVector vv : mapVector) {
    +        String fieldName = vv.getField().getLastName();
    +        if (!keyList.contains(fieldName)) {
    +          throw new UnsupportedOperationException("Unpivot data vector " +
    +              ds + " contains key " + fieldName + " not contained in key source!");
    +        }
    +        if (vv.getField().getType().getMinorType() == TypeProtos.MinorType.MAP) {
    +          throw new UnsupportedOperationException("Unpivot of nested map is not supported!");
    +        }
    +        if (field.equals("column")) {
    +          outputMapVector.addOrGet(fieldName, vv.getField().getType(), vv.getClass());
    +        } else {
    +          List<LogicalExpression> args = Lists.newArrayList();
    +          //TODO: Something else to access value of col such as emp_id?
    +          args.add(SchemaPath.getSimplePath(vv.getField().getPath()));
    +          //TODO: Put in the mapVector
    +          LogicalExpression call = FunctionCallFactory.createExpression(functions.get(field), args);
    +          //TODO: Is this sufficient to add to new Map?
    +          ValueVector vector1 = addMapVector(fieldName, outputMapVector, call);
    +          if (collector.hasErrors()) {
    +            throw new SchemaChangeException("Failure while materializing expression. "
    +                + collector.toErrorString());
    +          }
    +        }
    +      }
    +      dataSrcVecMap.put(ds, outputMapVector);*/
    +    }
    +    //Now create NDV in the outgoing container which was not avaliable in the incoming
    +    for (VectorWrapper<?> vw : incoming) {
    +      if (vw.getField().getLastName().equalsIgnoreCase("sum_width")) {//NullableFloat8 type
    +        addVectorToOutgoingContainer("ndv", vw, collector);
    +        break;
    +      }
    +    }
    +    container.setRecordCount(0);
    +    recordCount = 0;
    +    container.buildSchema(incoming.getSchema().getSelectionVectorMode());
    +  }
    +
    +  private void addVectorToOutgoingContainer(String field, VectorWrapper vw, ErrorCollector collector)
    +     throws SchemaChangeException {
    +    // Input map vector
    +    MapVector mapVector = (MapVector) vw.getValueVector();
    +    MaterializedField mf;
    +    assert mapVector.getPrimitiveVectors().size() > 0;
    +    // Proceed to create output map vector with same name e.g. statcount etc.
    +    MajorType mt = mapVector.getField().getType();
    +    if (functions.get(field) != null) {
    +      mf = MaterializedField.create(functions.get(field), mt);
    +    } else {
    +      mf = MaterializedField.create(field, mt);
    +    }
    +    assert !dataSrcVecMap.containsKey(mf);
    +    ValueVector vector = TypeHelper.getNewVector(mf, oContext.getAllocator());
    +    container.add(vector);
    +    MapVector outputMapVector = (MapVector) vector;
    +
    +    for (ValueVector vv : mapVector) {
    +      String fieldName = vv.getField().getLastName();
    +      if (!keyList.contains(fieldName)) {
    +        throw new UnsupportedOperationException("Unpivot data vector " +
    +                field + " contains key " + fieldName + " not contained in key source!");
    +      }
    +      if (vv.getField().getType().getMinorType() == TypeProtos.MinorType.MAP) {
    +        throw new UnsupportedOperationException("Unpivot of nested map is not supported!");
    +      }
    +      if (field.equals("column")) {
    +        outputMapVector.addOrGet(fieldName, vv.getField().getType(), vv.getClass());
    +      } else {
    +        List<LogicalExpression> args = Lists.newArrayList();
    +        LogicalExpression call;
    +        //TODO: Something else to access value of col such as emp_id?
    +        args.add(SchemaPath.getSimplePath(vv.getField().getPath()));
    +        //TODO: Put in the mapVector
    +        if (functions.get(field) != null) {
    +          call = FunctionCallFactory.createExpression(functions.get(field), args);
    +        } else {
    +          call = FunctionCallFactory.createExpression(field, args);
    +        }
    +        //TODO: Is this sufficient to add to new Map?
    +        ValueVector vector1 = addMapVector(fieldName, outputMapVector, call);
    +        if (collector.hasErrors()) {
    +          throw new SchemaChangeException("Failure while materializing expression. "
    +                  + collector.toErrorString());
    +        }
    +      }
    +    }
    +  }
    +
    +  @Override
    +  protected boolean setupNewSchema() throws SchemaChangeException {
    +    container.clear();
    +    buildKeyList();
    +    buildOutputContainer();
    +    return true;
    +  }
    +
    +  @Override
    +  protected IterOutcome doWork() {
    +    int outRecordCount = incoming.getRecordCount();
    +    HashMap<String, Long> nonNullRows = new HashMap<>();
    +
    +    for (VectorWrapper<?> vw : incoming) {
    +      ValueVector vv = vw.getValueVector();
    +      if (vv.getField().getType().getMinorType() != TypeProtos.MinorType.MAP) {
    +        //We only expect Maps in the incoming. STOP, if this is not the case.
    +        return IterOutcome.STOP;
    +      }
    +      MapVector mapVec = (MapVector) vv;
    +      Map<String, ValueHolder> statMap = aggregationMap.get(vv.getField().getLastName());
    +
    +      for (ValueVector mapElt : mapVec) {
    +        if (vv.getField().getLastName().equalsIgnoreCase("column")) {
    +          BigIntHolder nameHolder;
    +          if (statMap.get(mapElt.getField().getLastName()) != null) {
    +            nameHolder = (BigIntHolder) statMap.get(mapElt.getField().getLastName());
    +          } else {
    +            nameHolder = new BigIntHolder();
    +            statMap.put(mapElt.getField().getLastName(), nameHolder);
    +          }
    +          nameHolder.value = 1;
    +        } else if (vv.getField().getLastName().equalsIgnoreCase("statcount")
    +            || vv.getField().getLastName().equalsIgnoreCase("nonnullstatcount")) {
    +          BigIntHolder sumHolder;
    +          String colName = mapElt.getField().getLastName();
    +          if (statMap.get(colName) != null) {
    +            sumHolder = (BigIntHolder) statMap.get(colName);
    +          } else {
    +            sumHolder = new BigIntHolder();
    +            statMap.put(colName, sumHolder);
    +          }
    +          //TODO: assert size = 1
    +          //TODO: logger
    +          Object val = mapElt.getAccessor().getObject(0);
    +          if (val != null) {
    +            sumHolder.value += (long)val;
    +          }
    +        } else if (vv.getField().getLastName().equalsIgnoreCase("sum_width")) {
    +          NullableFloat8Holder sumHolder;
    +          String colName = mapElt.getField().getLastName();
    +          if (statMap == null) {
    +            statMap = aggregationMap.get(functions.get(vv.getField().getLastName()));
    +          }
    +          if (statMap.get(colName) != null) {
    +            sumHolder = (NullableFloat8Holder) statMap.get(colName);
    +          } else {
    +            sumHolder = new NullableFloat8Holder();
    +            statMap.put(colName, sumHolder);
    +          }
    +          //TODO: assert size = 1
    +          //TODO: logger
    +          Object val = mapElt.getAccessor().getObject(0);
    +          if (val != null) {
    +            sumHolder.value += (double) val;
    +            sumHolder.isSet = 1;
    +          }
    +        } else if (vv.getField().getLastName().equalsIgnoreCase("hll")) {
    +          ObjectHolder hllHolder;
    +          String colName = mapElt.getField().getLastName();
    +          if (statMap == null) {
    +            statMap = aggregationMap.get(functions.get(vv.getField().getLastName()));
    +          }
    +          if (statMap.get(colName) != null) {
    +            hllHolder = (ObjectHolder) statMap.get(colName);
    +          } else {
    +            hllHolder = new ObjectHolder();
    +            hllHolder.obj = new HyperLogLog(context.getContextInformation().getHllAccuracy());
    +            statMap.put(colName, hllHolder);
    +          }
    +          NullableVarBinaryVector hllVector = (NullableVarBinaryVector) mapElt;
    +          try {
    +            if (hllVector.getAccessor().isSet(0) == 1) {
    +              ByteArrayInputStream bais = new ByteArrayInputStream(hllVector.getAccessor().getObject(0), 0,
    +                  mapElt.getBufferSize());
    +              HyperLogLog other = HyperLogLog.Builder.build(new DataInputStream(bais));
    +              ((HyperLogLog) hllHolder.obj).addAll(other);
    +            }
    +          } catch (Exception ex) {
    +            //TODO: Catch IOException/CardinalityMergeException
    +            //TODO: logger
    +            return IterOutcome.STOP;
    +          }
    +        }
    +      }
    +      // Add NDV value vector map using HLL map (since the NDV map is directly generated from HLL and not produced by the underlying
    +      // Statistics Agg)
    +      Map<String, ValueHolder> hllMap = aggregationMap.get("hll");
    +      if (hllMap != null) {
    +        aggregationMap.put("ndv", hllMap);
    +      }
    +    }
    +    return IterOutcome.OK;
    +  }
    +
    +  public VectorContainer getOutgoingContainer() {
    +    return this.container;
    +  }
    +
    +  @Override
    +  public IterOutcome innerNext() {
    +    IterOutcome outcome;
    +    boolean didSomeWork = false;
    +    if (finished) {
    +      return IterOutcome.NONE;
    +    }
    +    try {
    +      outer: while (true) {
    +        outcome = next(incoming);
    +        switch (outcome) {
    +          case NONE:
    +            break outer;
    +          case OUT_OF_MEMORY:
    +          case NOT_YET:
    +          case STOP:
    +            return outcome;
    +          case OK_NEW_SCHEMA:
    +            if (first) {
    +              first =false;
    --- End diff --
    
    The multiple-fag approach to state management is awkward. Maybe define an enum with states:
    ```
    enum OperatorState { START, RUN, END };
    ```


---
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 infrastructure@apache.org or file a JIRA ticket
with INFRA.
---