- netty-0001: DnsResolveContext.finalResult ArrayList.contains O(R²) dedup on DNS records File: resolver-dns/.../dns/DnsResolveContext.java line ~914 Fix: LinkedHashSet gives O(1) dedup with preserved insertion order Ratio: 24.5x at R=50 records - dubbo-0002: MethodWalker.walkHierarchy no visited guard — O(2^D) diamond recursion File: dubbo-rpc-triple/.../rest/util/MethodWalker.java walkHierarchy() Fix: add visited HashSet, return early if already visited Ratio: 8x at D=3 (common Spring proxy depth) - doris-0004: NormalizeRepeat.buildContextWithAlias ImmutableList.contains O(S×G) for GROUPING SETS File: fe-core/.../nereids/rules/analysis/NormalizeRepeat.java buildContextWithAlias() Fix: convert groupingSetExpressions to HashSet before loop — O(1) lookup Ratio: 49x for CUBE(c1..c8), 1000x+ for CUBE(c1..c10)
192 lines
7.8 KiB
Java
192 lines
7.8 KiB
Java
/**
|
||
* doris-0004: NormalizeRepeat.buildContextWithAlias — List.contains O(S×G) for GROUPING SETS
|
||
*
|
||
* Demonstrates the defect: O(S×G) ImmutableList.contains per source expression
|
||
* vs O(G + S) with HashSet for groupingSetExpressions lookup.
|
||
*
|
||
* Compile: javac NormalizeRepeatGroupingSetAlgorithm.java
|
||
* Run: java NormalizeRepeatGroupingSetAlgorithm
|
||
*/
|
||
import java.util.*;
|
||
|
||
public class NormalizeRepeatGroupingSetAlgorithm {
|
||
|
||
// Simulate GROUPING SETS expressions as strings (column references)
|
||
// flatExpressions() returns ImmutableList in the real code
|
||
|
||
// ---- DEFECT: O(S×G) — List.contains per source expression ----
|
||
static long defectOps = 0;
|
||
|
||
static Map<String, String> buildContextDefect(
|
||
List<String> groupingSetExpressions, // ImmutableList in real code
|
||
List<String> sourceExpressions) {
|
||
Map<String, String> result = new LinkedHashMap<>();
|
||
for (String expression : sourceExpressions) {
|
||
// O(G) scan — the defect
|
||
defectOps += groupingSetExpressions.size();
|
||
if (groupingSetExpressions.contains(expression)) {
|
||
result.put(expression, "GROUPING_SLOT:" + expression);
|
||
} else {
|
||
result.put(expression, "ALIAS:" + expression);
|
||
}
|
||
}
|
||
return result;
|
||
}
|
||
|
||
// ---- FIX: O(G + S) — HashSet for O(1) membership test ----
|
||
static long fixOps = 0;
|
||
|
||
static Map<String, String> buildContextFixed(
|
||
List<String> groupingSetExpressionsRaw, // same ImmutableList
|
||
List<String> sourceExpressions) {
|
||
// One-time O(G) conversion to HashSet
|
||
fixOps += groupingSetExpressionsRaw.size();
|
||
Set<String> groupingSetExpressions = new HashSet<>(groupingSetExpressionsRaw);
|
||
|
||
Map<String, String> result = new LinkedHashMap<>();
|
||
for (String expression : sourceExpressions) {
|
||
// O(1) lookup
|
||
fixOps += 1;
|
||
if (groupingSetExpressions.contains(expression)) {
|
||
result.put(expression, "GROUPING_SLOT:" + expression);
|
||
} else {
|
||
result.put(expression, "ALIAS:" + expression);
|
||
}
|
||
}
|
||
return result;
|
||
}
|
||
|
||
// Generate flattened groupingSetExpressions for CUBE(c1..cK)
|
||
// CUBE(c1..cK) generates 2^K grouping sets, each being a subset of columns
|
||
static List<String> generateCubeGroupingSets(int k) {
|
||
List<String> flat = new ArrayList<>();
|
||
int numSets = 1 << k; // 2^k
|
||
for (int mask = 0; mask < numSets; mask++) {
|
||
for (int bit = 0; bit < k; bit++) {
|
||
if ((mask & (1 << bit)) != 0) {
|
||
flat.add("c" + (bit + 1));
|
||
}
|
||
}
|
||
}
|
||
return flat;
|
||
}
|
||
|
||
// Generate source expressions for a query with K grouping columns + extra output cols
|
||
static List<String> generateSourceExpressions(int k, int extra) {
|
||
List<String> sources = new ArrayList<>();
|
||
for (int i = 1; i <= k; i++) sources.add("c" + i); // grouping columns
|
||
for (int i = 1; i <= extra; i++) sources.add("agg" + i); // aggregate expressions
|
||
return sources;
|
||
}
|
||
|
||
static long benchmarkDefect(List<String> groupingSets, List<String> sources, int iters) {
|
||
long start = System.nanoTime();
|
||
for (int i = 0; i < iters; i++) {
|
||
Map<String, String> result = new LinkedHashMap<>();
|
||
for (String expression : sources) {
|
||
if (groupingSets.contains(expression)) {
|
||
result.put(expression, "GROUPING:" + expression);
|
||
} else {
|
||
result.put(expression, "ALIAS:" + expression);
|
||
}
|
||
}
|
||
}
|
||
return System.nanoTime() - start;
|
||
}
|
||
|
||
static long benchmarkFixed(List<String> groupingSetsRaw, List<String> sources, int iters) {
|
||
long start = System.nanoTime();
|
||
for (int i = 0; i < iters; i++) {
|
||
Set<String> groupingSets = new HashSet<>(groupingSetsRaw);
|
||
Map<String, String> result = new LinkedHashMap<>();
|
||
for (String expression : sources) {
|
||
if (groupingSets.contains(expression)) {
|
||
result.put(expression, "GROUPING:" + expression);
|
||
} else {
|
||
result.put(expression, "ALIAS:" + expression);
|
||
}
|
||
}
|
||
}
|
||
return System.nanoTime() - start;
|
||
}
|
||
|
||
public static void main(String[] args) {
|
||
System.out.println("=== doris-0004: NormalizeRepeat buildContextWithAlias List.contains O(S×G) ===");
|
||
System.out.println();
|
||
|
||
// Test correctness with simple ROLLUP(c1, c2, c3) = 4 grouping sets
|
||
// Grouping sets: {c1,c2,c3}, {c1,c2}, {c1}, {}
|
||
List<String> smallGroupingSets = Arrays.asList(
|
||
"c1", "c2", "c3", "c1", "c2", "c1" // flattened
|
||
);
|
||
List<String> sources = Arrays.asList("c1", "c2", "c3", "sum_v", "count_v");
|
||
|
||
System.out.println("--- Correctness check (ROLLUP(c1,c2,c3)) ---");
|
||
defectOps = 0;
|
||
fixOps = 0;
|
||
Map<String, String> defectResult = buildContextDefect(smallGroupingSets, sources);
|
||
Map<String, String> fixResult = buildContextFixed(smallGroupingSets, sources);
|
||
|
||
System.out.printf("Defect result: %s%n", defectResult);
|
||
System.out.printf("Fix result: %s%n", fixResult);
|
||
|
||
boolean correct = defectResult.equals(fixResult);
|
||
System.out.printf("Results match: %s%n%n", correct ? "PASS" : "FAIL");
|
||
|
||
// Algorithmic operation counts
|
||
System.out.println("--- Algorithmic operation counts ---");
|
||
System.out.printf("%-8s %10s %12s %10s %8s%n",
|
||
"CUBE(K)", "G (flat)", "S (sources)", "Defect S×G", "Fix G+S");
|
||
System.out.printf("%-8s %10s %12s %10s %8s%n",
|
||
"-------", "--------", "-----------", "-----------", "-------");
|
||
|
||
boolean allRatiosOk = true;
|
||
for (int k : new int[]{3, 5, 7, 8, 10}) {
|
||
List<String> gsets = generateCubeGroupingSets(k);
|
||
List<String> srcs = generateSourceExpressions(k, k); // K grouping + K aggregates
|
||
int G = gsets.size();
|
||
int S = srcs.size();
|
||
long defectCount = (long) S * G;
|
||
long fixCount = G + S;
|
||
double ratio = (double) defectCount / fixCount;
|
||
System.out.printf("CUBE(%d) %10d %12d %10d %8d [%.1fx]%n",
|
||
k, G, S, defectCount, fixCount, ratio);
|
||
if (k >= 7 && ratio < 10.0) {
|
||
System.out.printf(" FAIL: expected ratio >= 10x for CUBE(%d)%n", k);
|
||
allRatiosOk = false;
|
||
}
|
||
}
|
||
System.out.println();
|
||
|
||
// Performance benchmarks
|
||
System.out.println("--- Performance benchmarks ---");
|
||
System.out.printf("%-10s %15s %12s %10s%n", "Scenario", "Defect (ns/q)", "Fix (ns/q)", "Speedup");
|
||
System.out.printf("%-10s %15s %12s %10s%n", "--------", "-------------", "-----------", "-------");
|
||
|
||
int iters = 50_000;
|
||
for (int k : new int[]{5, 7, 8}) {
|
||
List<String> gsets = generateCubeGroupingSets(k);
|
||
List<String> srcs = generateSourceExpressions(k, 10);
|
||
|
||
// Warmup
|
||
for (int w = 0; w < 500; w++) {
|
||
benchmarkDefect(gsets, srcs, 1);
|
||
benchmarkFixed(gsets, srcs, 1);
|
||
}
|
||
|
||
long dt = benchmarkDefect(gsets, srcs, iters);
|
||
long ft = benchmarkFixed(gsets, srcs, iters);
|
||
double speedup = (double) dt / ft;
|
||
|
||
System.out.printf("CUBE(%d) %15.0f %12.0f %9.1fx%n",
|
||
k,
|
||
(double) dt / iters,
|
||
(double) ft / iters,
|
||
speedup);
|
||
}
|
||
|
||
System.out.println();
|
||
System.out.printf("RESULT: %s%n", (correct && allRatiosOk) ? "PASS" : "FAIL");
|
||
if (!correct || !allRatiosOk) System.exit(1);
|
||
}
|
||
}
|