Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Coral-Incremental] Cost calculation for RelNode #516

Open
wants to merge 18 commits into
base: master
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from 15 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
@@ -0,0 +1,285 @@
/**
* Copyright 2024 LinkedIn Corporation. All rights reserved.
* Licensed under the BSD-2 Clause license.
* See LICENSE in the project root for license information.
*/
package com.linkedin.coral.incremental;

import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Paths;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.Iterator;
import java.util.List;
import java.util.Map;

import com.google.gson.JsonElement;
import com.google.gson.JsonObject;
import com.google.gson.JsonParser;

import org.apache.calcite.plan.RelOptTable;
import org.apache.calcite.rel.RelNode;
import org.apache.calcite.rel.core.TableScan;
import org.apache.calcite.rel.logical.LogicalJoin;
import org.apache.calcite.rel.logical.LogicalProject;
import org.apache.calcite.rel.logical.LogicalUnion;
import org.apache.calcite.rel.type.RelDataType;
import org.apache.calcite.rel.type.RelDataTypeField;
import org.apache.calcite.rex.RexCall;
import org.apache.calcite.rex.RexInputRef;
import org.apache.calcite.rex.RexNode;

import static java.lang.Math.*;


/**
* RelNodeCostEstimator is a utility class designed to estimate the cost of executing relational operations
* in a query plan. It uses statistical information about table row counts and column distinct values
* to compute costs associated with different types of relational operations like table scans, joins,
* unions, and projections.
*
* <p>This class supports loading statistics from a JSON configuration file.
* For a relational operations (RelNode), the execution cost and row count are estimated based on
* these statistics and the input relational expressions.
*
* <p>The cost estimation takes into account factors such as I/O costs and data shuffling costs.
* The cost of writing a row to disk is IOCostValue, and the cost of shuffling a row
* between nodes is shuffleCostValue.
*
* <p>Cost is get from 'getCost' method, which returns the total cost of the query plan, and cost consists of
* execution cost and I/O cost.
*/
public class RelNodeCostEstimator {
yyy1000 marked this conversation as resolved.
Show resolved Hide resolved

class CostInfo {
// TODO: we may also need to add TableName field.
Double shuffleCost;
Double rowCount;

public CostInfo(Double shuffleCost, Double row) {
this.shuffleCost = shuffleCost;
this.rowCount = row;
}
}

class TableStatistic {
// The number of rows in the table
Double rowCount;
// The number of distinct values in each column
Map<String, Double> distinctCountByRow;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does this work for nested columns? If yes, what is the spec of the key?

Does it work for complex types?

If the answer is no, we should be explicit about this at least in the Java doc.


public TableStatistic() {
this.distinctCountByRow = new HashMap<>();
}
}

class JoinKey {
String leftTableName;
String rightTableName;
String leftFieldName;
String rightFieldName;

public JoinKey(String leftTableName, String rightTableName, String leftFieldName, String rightFieldName) {
this.leftTableName = leftTableName;
this.rightTableName = rightTableName;
this.leftFieldName = leftFieldName;
this.rightFieldName = rightFieldName;
}
}

private Map<String, TableStatistic> costStatistic = new HashMap<>();

private final Double IOCostValue;

private final Double shuffleCostValue;

public RelNodeCostEstimator(Double IOCostValue, Double shuffleCostValue) {
this.IOCostValue = IOCostValue;
this.shuffleCostValue = shuffleCostValue;
}

/**
* Loads statistics from a JSON configuration file and stores them in internal data structures.
*
* <p>This method reads a JSON file from the specified path, parses its content, and extracts
* statistical information. For each table in the JSON object, it retrieves the row count and
* distinct counts for each column. These values are then stored in the `stat` and `distinctStat`
* maps, respectively.
*
* @param configPath the path to the JSON configuration file
*/
public void loadStatistic(String configPath) throws IOException {
try {
String content = new String(Files.readAllBytes(Paths.get(configPath)));
JsonObject jsonObject = new JsonParser().parse(content).getAsJsonObject();
for (Map.Entry<String, JsonElement> entry : jsonObject.entrySet()) {
TableStatistic tableStatistic = new TableStatistic();
String tableName = entry.getKey();
JsonObject tableObject = entry.getValue().getAsJsonObject();

Double rowCount = tableObject.get("RowCount").getAsDouble();

JsonObject distinctCounts = tableObject.getAsJsonObject("DistinctCounts");

tableStatistic.rowCount = rowCount;

for (Map.Entry<String, JsonElement> distinctEntry : distinctCounts.entrySet()) {
String columnName = distinctEntry.getKey();
Double distinctCount = distinctEntry.getValue().getAsDouble();

tableStatistic.distinctCountByRow.put(columnName, distinctCount);
}
costStatistic.put(tableName, tableStatistic);

}
} catch (IOException e) {
throw new IOException("Failed to load statistics from the configuration file: " + configPath, e);
}

}

/**
* Returns the total cost of executing a relational operation.
*
* <p>This method computes the cost of executing a relational operation based on the input
* relational expression. The cost is calculated as the sum of the execution cost and the I/O cost.
* We assume that I/O only occurs at the root of the query plan (Project) where we write the output to disk.
* So the cost is the sum of the shuffle cost of all children RelNodes and IOCostValue * row count of the root Project RelNode.
*
* @param rel the input relational expression
* @return the total cost of executing the relational operation
*/
public Double getCost(RelNode rel) {
CostInfo executionCostInfo = getExecutionCost(rel);
Double IOCost = executionCostInfo.rowCount * IOCostValue;
return executionCostInfo.shuffleCost * shuffleCostValue + IOCost;
}

public CostInfo getExecutionCost(RelNode rel) {
if (rel instanceof TableScan) {
return getExecutionCostTableScan((TableScan) rel);
} else if (rel instanceof LogicalJoin) {
return getExecutionCostJoin((LogicalJoin) rel);
} else if (rel instanceof LogicalUnion) {
return getExecutionCostUnion((LogicalUnion) rel);
} else if (rel instanceof LogicalProject) {
return getExecutionCostProject((LogicalProject) rel);
}
throw new IllegalArgumentException("Unsupported relational operation: " + rel.getClass().getSimpleName());
}

private CostInfo getExecutionCostTableScan(TableScan scan) {
RelOptTable originalTable = scan.getTable();
String tableName = getTableName(originalTable);
try {
TableStatistic tableStat = costStatistic.get(tableName);
Double rowCount = tableStat.rowCount;
return new CostInfo(rowCount, rowCount);
} catch (NullPointerException e) {
throw new IllegalArgumentException("Table statistics not found for table: " + tableName);
yyy1000 marked this conversation as resolved.
Show resolved Hide resolved
}
}

private String getTableName(RelOptTable table) {
return String.join(".", table.getQualifiedName());
}

private CostInfo getExecutionCostJoin(LogicalJoin join) {
RelNode left = join.getLeft();
RelNode right = join.getRight();
CostInfo leftCost = getExecutionCost(left);
CostInfo rightCost = getExecutionCost(right);
Double joinSize = estimateJoinSize(join, leftCost.rowCount, rightCost.rowCount);
// The shuffle cost of a join is the maximum shuffle cost of its children because
// in modern distributed systems, the shuffle cost is dominated by the largest shuffle.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Terminology is not clear here. What is the shuffle cost of children? We might need to discuss this over a doc.

return new CostInfo(max(leftCost.shuffleCost, rightCost.shuffleCost), joinSize);
}

private List<JoinKey> getJoinKeys(LogicalJoin join) {
List<JoinKey> joinKeys = new ArrayList<>();
RexNode condition = join.getCondition();
if (condition instanceof RexCall) {
getJoinKeysFromJoinCondition((RexCall) condition, join, joinKeys);
}
// Assertion to check if joinKeys.size() is greater than or equal to 1
if (joinKeys.size() < 1) {
throw new IllegalArgumentException("Join keys size is less than 1");
}
return joinKeys;
yyy1000 marked this conversation as resolved.
Show resolved Hide resolved
}

private void getJoinKeysFromJoinCondition(RexCall call, LogicalJoin join, List<JoinKey> joinKeys) {
if (call.getOperator().getName().equalsIgnoreCase("AND")) {
// Process each operand of the AND separately
for (RexNode operand : call.getOperands()) {
if (operand instanceof RexCall) {
getJoinKeysFromJoinCondition((RexCall) operand, join, joinKeys);
}
}
} else {
// Process the join condition (e.g., EQUALS)
List<RexNode> operands = call.getOperands();
if (operands.size() == 2 && operands.get(0) instanceof RexInputRef && operands.get(1) instanceof RexInputRef) {
RexInputRef leftRef = (RexInputRef) operands.get(0);
RexInputRef rightRef = (RexInputRef) operands.get(1);
RelDataType leftType = join.getLeft().getRowType();
RelDataType rightType = join.getRight().getRowType();

int leftIndex = leftRef.getIndex();
int rightIndex = rightRef.getIndex() - leftType.getFieldCount();

RelDataTypeField leftField = leftType.getFieldList().get(leftIndex);
String leftTableName = getTableName(join.getLeft().getTable());
String leftFieldName = leftField.getName();
RelDataTypeField rightField = rightType.getFieldList().get(rightIndex);
String rightTableName = getTableName(join.getRight().getTable());
String rightFieldName = rightField.getName();

joinKeys.add(new JoinKey(leftTableName, rightTableName, leftFieldName, rightFieldName));
}
}
}

private Double estimateJoinSize(LogicalJoin join, Double leftSize, Double rightSize) {
List<JoinKey> joinKeys = getJoinKeys(join);
Double selectivity = 1.0;
for (JoinKey joinKey : joinKeys) {
String leftTableName = joinKey.leftTableName;
String rightTableName = joinKey.rightTableName;
String leftFieldName = joinKey.leftFieldName;
String rightFieldName = joinKey.rightFieldName;
try {
TableStatistic leftTableStat = costStatistic.get(leftTableName);
TableStatistic rightTableStat = costStatistic.get(rightTableName);
Double leftCardinality = leftTableStat.rowCount;
Double rightCardinality = rightTableStat.rowCount;
Double leftDistinct = leftTableStat.distinctCountByRow.getOrDefault(leftFieldName, leftCardinality);
Double rightDistinct = rightTableStat.distinctCountByRow.getOrDefault(rightFieldName, rightCardinality);
selectivity *= 1 / max(leftDistinct, rightDistinct);
} catch (NullPointerException e) {
throw new IllegalArgumentException(
"Table statistics not found for table: " + leftTableName + " or " + rightTableName);
}
}
return leftSize * rightSize * selectivity;
}

private CostInfo getExecutionCostUnion(LogicalUnion union) {
Double unionCost = 0.0;
Double unionSize = 0.0;
RelNode input;
for (Iterator var4 = union.getInputs().iterator(); var4.hasNext();) {
input = (RelNode) var4.next();
CostInfo inputCost = getExecutionCost(input);
unionSize += inputCost.rowCount;
unionCost = max(inputCost.shuffleCost, unionCost);
}
unionCost *= 2;
return new CostInfo(unionCost, unionSize);
}

private CostInfo getExecutionCostProject(LogicalProject project) {
return getExecutionCost(project.getInput());
}
}
Original file line number Diff line number Diff line change
@@ -0,0 +1,106 @@
/**
* Copyright 2024 LinkedIn Corporation. All rights reserved.
* Licensed under the BSD-2 Clause license.
* See LICENSE in the project root for license information.
*/
package com.linkedin.coral.incremental;

import java.io.File;
import java.io.IOException;
import java.util.HashMap;
import java.util.Map;

import org.apache.calcite.rel.RelNode;
import org.apache.commons.io.FileUtils;
import org.apache.hadoop.hive.conf.HiveConf;
import org.apache.hadoop.hive.metastore.api.MetaException;
import org.apache.hadoop.hive.ql.metadata.HiveException;
import org.testng.annotations.AfterTest;
import org.testng.annotations.BeforeClass;
import org.testng.annotations.Test;

import static com.linkedin.coral.incremental.TestUtils.*;
import static org.testng.Assert.*;


public class RelNodeCostEstimatorTest {
private HiveConf conf;

private RelNodeCostEstimator estimator;

static final String TEST_JSON_FILE_DIR = "src/test/resources/";

@BeforeClass
public void beforeClass() throws HiveException, MetaException, IOException {
conf = TestUtils.loadResourceHiveConf();
estimator = new RelNodeCostEstimator(2.0, 1.0);
TestUtils.initializeViews(conf);
}

@AfterTest
public void afterClass() throws IOException {
FileUtils.deleteDirectory(new File(conf.get(CORAL_INCREMENTAL_TEST_DIR)));
}

public Map<String, Double> fakeStatData() {
Map<String, Double> stat = new HashMap<>();
stat.put("hive.test.bar1", 80.0);
stat.put("hive.test.bar2", 20.0);
stat.put("hive.test.bar1_prev", 40.0);
stat.put("hive.test.bar2_prev", 10.0);
stat.put("hive.test.bar1_delta", 60.0);
stat.put("hive.test.bar2_delta", 10.0);
return stat;
}

@Test
public void testSimpleSelectAll() throws IOException {
String sql = "SELECT * FROM test.bar1";
RelNode relNode = hiveToRelConverter.convertSql(sql);
estimator.loadStatistic(TEST_JSON_FILE_DIR + "statistic.json");
assertEquals(estimator.getCost(relNode), 300.0);
}

@Test
public void testSimpleJoin() throws IOException {
String sql = "SELECT * FROM test.bar1 JOIN test.bar2 ON test.bar1.x = test.bar2.x";
yyy1000 marked this conversation as resolved.
Show resolved Hide resolved
RelNode relNode = hiveToRelConverter.convertSql(sql);
estimator.loadStatistic(TEST_JSON_FILE_DIR + "statistic.json");
assertEquals(estimator.getCost(relNode), 500.0);
}

@Test
public void testSimpleUnion() throws IOException {
String sql = "SELECT *\n" + "FROM test.bar1 AS bar1\n" + "INNER JOIN test.bar2 AS bar2 ON bar1.x = bar2.x\n"
+ "UNION ALL\n" + "SELECT *\n" + "FROM test.bar3 AS bar3\n" + "INNER JOIN test.bar2 AS bar2 ON bar3.x = bar2.x";
RelNode relNode = hiveToRelConverter.convertSql(sql);
estimator.loadStatistic(TEST_JSON_FILE_DIR + "statistic.json");
assertEquals(estimator.getCost(relNode), 680.0);
}

@Test
public void testUnsupportOperator() throws IOException {
String sql = "SELECT * FROM test.bar1 WHERE x = 1";
RelNode relNode = hiveToRelConverter.convertSql(sql);
estimator.loadStatistic(TEST_JSON_FILE_DIR + "statistic.json");
try {
estimator.getCost(relNode);
fail("Should throw exception");
} catch (RuntimeException e) {
assertEquals(e.getMessage(), "Unsupported relational operation: " + "LogicalFilter");
}
}

@Test
public void testNoStatistic() throws IOException {
String sql = "SELECT * FROM test.foo";
RelNode relNode = hiveToRelConverter.convertSql(sql);
estimator.loadStatistic(TEST_JSON_FILE_DIR + "statistic.json");
try {
estimator.getCost(relNode);
fail("Should throw exception");
} catch (RuntimeException e) {
assertEquals(e.getMessage(), "Table statistics not found for table: " + "hive.test.foo");
}
}
}
Loading