java-topology/defects/victoria-metrics/unit/VictoriaMetricsStreamAggrAlgorithm.java

package unit;

import java.util.*;

/**
 * victoria-metrics-0001 — streamaggr: slices.Contains O(L×K) inside Push() hot loop
 *
 * Models lib/streamaggr/streamaggr.go:
 *   func (a *aggregator) Push(tss []prompb.TimeSeries, ...) {
 *       for _, ts := range tss {
 *           labels.Labels = dropSeriesLabels(labels.Labels, ts.Labels, dropLabels)
 *           getInputOutputLabels(..., a.by, a.without)
 *       }
 *   }
 *
 * dropSeriesLabels — for each label, calls slices.Contains(labelNames, label.Name) → O(K)
 * getInputOutputLabels — for each label, calls slices.Contains(without/by, label.Name) → O(K)
 *
 * Defective: O(N × L × K) total comparisons
 * Fixed:     O(N × L) using pre-built map[string]struct{} (built once, O(K))
 *
 * Ratio >= 10× at N=1000, L=20, K=20
 *
 * No JUnit. Uses assert. Prints N/N PASS.
 *
 * Compile: javac -d . VictoriaMetricsStreamAggrAlgorithm.java
 * Run:     java -ea -cp . unit.VictoriaMetricsStreamAggrAlgorithm
 */
public class VictoriaMetricsStreamAggrAlgorithm {

    static int passed = 0;
    static int total  = 0;

    static void check(String desc, boolean cond) {
        total++;
        if (cond) {
            passed++;
            System.out.println("PASS: " + desc);
        } else {
            System.out.println("FAIL: " + desc);
            throw new AssertionError("FAIL: " + desc);
        }
    }

    // -----------------------------------------------------------------------
    // Slow: models dropSeriesLabels + getInputOutputLabels with slices.Contains
    // For each of N series, for each of L labels, scans K-element filter list
    // Returns total comparison operations performed.
    // -----------------------------------------------------------------------
    static long pushSlow(int N, int L, List<String> filterList) {
        long ops = 0;
        int K = filterList.size();
        // Simulate N time series, each with L labels named "label_0".."label_{L-1}"
        // filterList contains the "by" or "without" set (some subset of label names)
        for (int s = 0; s < N; s++) {
            for (int l = 0; l < L; l++) {
                String labelName = "label_" + l;
                // slices.Contains: O(K) linear scan
                for (int k = 0; k < K; k++) {
                    ops++;
                    if (filterList.get(k).equals(labelName)) {
                        break; // found — early exit (worst case: not found, full K ops)
                    }
                }
            }
        }
        return ops;
    }

    // -----------------------------------------------------------------------
    // Fast: pre-build map[string]struct{} once, then O(1) per label per series
    // Returns total comparison operations performed.
    // -----------------------------------------------------------------------
    static long pushFast(int N, int L, List<String> filterList) {
        long ops = 0;
        // Build set once — O(K)
        Set<String> filterSet = new HashSet<>(filterList);
        ops += filterList.size(); // cost of building the set

        // Now O(1) per label lookup
        for (int s = 0; s < N; s++) {
            for (int l = 0; l < L; l++) {
                String labelName = "label_" + l;
                ops++; // O(1) hash lookup
                filterSet.contains(labelName);
            }
        }
        return ops;
    }

    // -----------------------------------------------------------------------
    // Correctness: both methods produce the same classification of labels
    // -----------------------------------------------------------------------
    static List<String> classifyLabelsSlice(List<String> labels, List<String> byList) {
        List<String> output = new ArrayList<>();
        for (String label : labels) {
            boolean inBy = false;
            for (String b : byList) {
                if (b.equals(label)) { inBy = true; break; }
            }
            if (inBy) output.add(label); // output = "by" labels
        }
        return output;
    }

    static List<String> classifyLabelsMap(List<String> labels, List<String> byList) {
        Set<String> bySet = new HashSet<>(byList);
        List<String> output = new ArrayList<>();
        for (String label : labels) {
            if (bySet.contains(label)) output.add(label);
        }
        return output;
    }

    public static void main(String[] args) {
        System.out.println("victoria-metrics-0001: streamaggr label filter O(N×L×K) → O(N×L)");
        System.out.println();

        // --- Correctness ---
        List<String> labels  = Arrays.asList("label_0","label_1","label_2","label_3","label_4");
        List<String> byList  = Arrays.asList("label_1","label_3");

        List<String> slowResult = classifyLabelsSlice(labels, byList);
        List<String> fastResult = classifyLabelsMap(labels, byList);

        check("correctness: same labels classified", slowResult.equals(fastResult));
        check("correctness: 2 labels in 'by' set", slowResult.size() == 2);
        check("correctness: label_1 in output", slowResult.contains("label_1"));
        check("correctness: label_3 in output", slowResult.contains("label_3"));
        check("correctness: label_0 not in output", !slowResult.contains("label_0"));

        // --- Performance ---
        // N=1000 series, L=20 labels each, K=20 filter entries (worst case: none match → full scan)
        int N = 1000, L = 20, K = 20;

        // Build a filter list of 20 entries NOT matching any of our L labels
        // (worst case — label not found means full K comparisons every time)
        List<String> filterList = new ArrayList<>();
        for (int k = 0; k < K; k++) filterList.add("filter_" + k);

        long slowOps = pushSlow(N, L, filterList);
        long fastOps = pushFast(N, L, filterList);

        double ratio = (double) slowOps / fastOps;
        System.out.println("  N=" + N + " series, L=" + L + " labels, K=" + K + " filters");
        System.out.println("  Slow ops: " + slowOps);
        System.out.println("  Fast ops: " + fastOps);
        System.out.printf ("  Ratio:    %.1fx%n", ratio);
        System.out.println();

        check("ops: slow is O(N*L*K) = " + (N*L*K),       slowOps == (long) N * L * K);
        check("ops: fast is ~O(N*L+K) < 2*N*L",           fastOps <= (long) 2 * N * L + K);
        check("speedup >= 10x at N=1000/L=20/K=20",       ratio >= 10.0);

        // Scale: larger cluster
        N = 10000; L = 30; K = 20;
        long slowOps2 = pushSlow(N, L, filterList); // reuse same filterList (K=20 entries)
        List<String> filterList2 = new ArrayList<>();
        for (int k = 0; k < K; k++) filterList2.add("filter_" + k);
        long fastOps2 = pushFast(N, L, filterList2);
        double ratio2 = (double) slowOps2 / fastOps2;
        System.out.println("  Scale test: N=" + N + " series, L=" + L + " labels, K=" + K);
        System.out.println("  Slow ops: " + slowOps2);
        System.out.println("  Fast ops: " + fastOps2);
        System.out.printf ("  Ratio:    %.1fx%n", ratio2);
        System.out.println();
        check("scale: speedup >= 10x at N=10000/L=30/K=20", ratio2 >= 10.0);

        System.out.println(passed + "/" + total + " PASS");
    }
}