amrishlal commented on a change in pull request #6811: URL: https://github.com/apache/incubator-pinot/pull/6811#discussion_r621789257
########## File path: pinot-core/src/main/java/org/apache/pinot/core/query/optimizer/filter/NumericalFilterOptimizer.java ########## @@ -0,0 +1,272 @@ +/** + * 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.pinot.core.query.optimizer.filter; + +import java.math.BigDecimal; +import java.util.List; +import javax.annotation.Nullable; +import org.apache.pinot.common.request.Expression; +import org.apache.pinot.common.request.ExpressionType; +import org.apache.pinot.common.request.Function; +import org.apache.pinot.common.request.Literal; +import org.apache.pinot.common.utils.request.FilterQueryTree; +import org.apache.pinot.common.utils.request.RequestUtils; +import org.apache.pinot.pql.parsers.pql2.ast.FilterKind; +import org.apache.pinot.spi.data.FieldSpec; +import org.apache.pinot.spi.data.Schema; + + +/** + * Numerical expressions of form "column = literal" or "column != literal" can compare a column of one datatype + * (say INT) with a literal of different datatype (say DOUBLE). These expressions can not be evaluated on the Server. + * Hence, we rewrite such expressions into an equivalent expression whose LHS and RHS are of the same datatype. + * + * Simple predicate examples: + * 1) WHERE "intColumn = 5.0" gets rewritten to "WHERE intColumn = 5" + * 2) WHERE "intColumn != 5.0" gets rewritten to "WHERE intColumn != 5" + * 3) WHERE "intColumn = 5.5" gets rewritten to "WHERE false" because INT values can not match 5.5. + * 4) WHERE "intColumn = 3000000000 gets rewritten to "WHERE false" because INT values can not match 3000000000. + * 5) WHERE "intColumn != 3000000000 gets rewritten to "WHERE true" becuase INT values always not equal to 3000000000. + * + * Compound predicate examples: + * 6) WHERE "intColumn1 = 5.5 AND intColumn2 = intColumn3" + * rewrite to "WHERE false AND intColumn2 = intColumn3" + * rewrite to "WHERE intColumn2 = intColumn3" + * 7) WHERE "intColumn1 != 5.5 OR intColumn2 = 5000000000" (5000000000 is out of bounds for integer column) + * rewrite to "WHERE true OR false" + * rewrite to "WHERE true" + * rewrite to query without any WHERE clause. + * + * When entire predicate gets rewritten to false (Example 3 above), the query will not return any data. Hence, it is + * better for the Broker itself to return an empty response rather than sending the query to servers for further + * evaluation. + */ +public class NumericalFilterOptimizer implements FilterOptimizer { + + private static final Expression TRUE = RequestUtils.getLiteralExpression(true); + private static final Expression FALSE = RequestUtils.getLiteralExpression(false); + + @Override + public FilterQueryTree optimize(FilterQueryTree filterQueryTree, @Nullable Schema schema) { + // Don't do anything here since this is for PQL queries which we no longer support. + return filterQueryTree; + } + + @Override + public Expression optimize(Expression expression, @Nullable Schema schema) { + ExpressionType type = expression.getType(); + if (type != ExpressionType.FUNCTION) { + // Not a function, so we have nothing to rewrite. + return expression; + } + + Function function = expression.getFunctionCall(); + List<Expression> operands = function.getOperands(); + String operator = function.getOperator(); + if (operator.equals(FilterKind.AND.name()) || operator.equals(FilterKind.OR.name())) { + // One of the operands may be an EQUALS or NOT_EQUALS function so recursively traverse the expression tree to see + // if we find an EQUALS or NOT_EQUALS function to rewrite. + operands.forEach(operand -> optimize(operand, schema)); + + // We have rewritten the child operands, so rewrite the parent if needed. + return optimizeCurrent(expression); + } else if (operator.equals(FilterKind.EQUALS.name()) || operator.equals(FilterKind.NOT_EQUALS.name())) { + // Verify that LHS is a numeric column and RHS is a numeric literal before rewriting. + Expression lhs = operands.get(0), rhs = operands.get(1); + if (isNumericColumn(lhs, schema) && isNumericLiteral(rhs)) { + // Rewrite the expression. + return rewrite(expression, lhs, rhs, schema); + } + } + + return expression; + } + + /** + * If any of the operands of AND function is "false", then the AND function itself is false and can be replaced with + * "false" literal. Otherwise, remove all the "true" operands of the AND function. Similarly, if any of the operands + * of OR function is "true", then the OR function itself is true and can be replaced with "true" literal. Otherwise, + * remove all the "false" operands of the OR function. + */ + private static Expression optimizeCurrent(Expression expression) { + Function function = expression.getFunctionCall(); + List<Expression> operands = function.getOperands(); + if (function.getOperator().equals(FilterKind.AND.name())) { + // If any of the literal operands are FALSE, then replace AND function with FALSE. + if (operands.stream().anyMatch(operand -> operand.equals(FALSE))) { + return setExpressionToBoolean(expression, false); + } + + // Remove all Literal operands that are TRUE. + operands.removeIf(x -> x.equals(TRUE)); + if (operands.size() == 0) { + return setExpressionToBoolean(expression, true); + } + } else if (function.getOperator().equals(FilterKind.OR.name())) { + // If any of the literal operands are TRUE, then replace OR function with TRUE + if (operands.stream().anyMatch(operand -> operand.equals(TRUE))) { + return setExpressionToBoolean(expression, true); + } + + // Remove all Literal operands that are FALSE. + operands.removeIf(x -> x.equals(FALSE)); + if (operands.size() == 0) { + return setExpressionToBoolean(expression, false); + } + } + + return expression; + } + + private boolean isNumericColumn(Expression expression, Schema schema) { + if (expression.getType() != ExpressionType.IDENTIFIER) { + // Expression can not be a column. + return false; + } + + String column = expression.getIdentifier().getName(); + FieldSpec fieldSpec = schema.getFieldSpecFor(column); + if (fieldSpec == null || !fieldSpec.isSingleValueField()) { + // Expression can not be a column name. + return false; + } + + return schema.getFieldSpecFor(column).getDataType().isNumeric(); + } + + private boolean isNumericLiteral(Expression expression) { + if (expression.getType() == ExpressionType.LITERAL) { + Literal._Fields type = expression.getLiteral().getSetField(); + switch (type) { + case SHORT_VALUE: + case INT_VALUE: + case LONG_VALUE: + case DOUBLE_VALUE: + return true; + } + } + return false; + } + + /** Change the expression value to boolean literal with given value. */ + private static Expression setExpressionToBoolean(Expression expression, boolean value) { + expression.unsetFunctionCall(); + expression.setType(ExpressionType.LITERAL); + expression.setLiteral(Literal.boolValue(value)); + + return expression; + } + + /** + * Rewrite expressions of form "column = literal" or "column != literal" to ensure that RHS literal is the same + * datatype as LHS column. + */ + private Expression rewrite(Expression equals, Expression lhs, Expression rhs, Schema schema) { + // Get expression operator + boolean result = equals.getFunctionCall().getOperator().equals(FilterKind.NOT_EQUALS.name()); + + // Get column data type. + FieldSpec.DataType dataType = schema.getFieldSpecFor(lhs.getIdentifier().getName()).getDataType(); + + switch (rhs.getLiteral().getSetField()) { + case SHORT_VALUE: + case INT_VALUE: + // No rewrites needed since SHORT and INT conversion to numeric column types (INT, LONG, FLOAT, and DOUBLE) is + // lossless and will be implicitly handled on the server side. + break; + case LONG_VALUE: { + long actual = rhs.getLiteral().getLongValue(); + switch (dataType) { + case INT: { + int converted = (int) actual; + if (converted != actual) { + // Long value does not fall within the bounds of INT column. + setExpressionToBoolean(equals, result); + } else { + // Replace long value with converted int value. + rhs.setLiteral(Literal.intValue(converted)); Review comment: Did in place modification since we need to preserve the reference to `rhs` to avoid destroying the expression tree that is being traversed. -- 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: commits-unsubscr...@pinot.apache.org For additional commands, e-mail: commits-h...@pinot.apache.org
