Add benchmarks for basic prebuilt MeterFilter implementations

benchmarks/benchmarks-core/src/jmh/java/io/micrometer/benchmark/BenchmarkSupport.java

@@ -0,0 +1,298 @@
+/*
+ * Copyright 2025 VMware, Inc.
+ *
+ * Licensed 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
+ *
+ * https://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 io.micrometer.benchmark;
+
+import io.micrometer.common.lang.Nullable;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.profile.GCProfiler;
+import org.openjdk.jmh.profile.LinuxPerfAsmProfiler;
+import org.openjdk.jmh.profile.LinuxPerfNormProfiler;
+import org.openjdk.jmh.runner.Runner;
+import org.openjdk.jmh.runner.RunnerException;
+import org.openjdk.jmh.runner.options.ChainedOptionsBuilder;
+import org.openjdk.jmh.runner.options.Options;
+import org.openjdk.jmh.runner.options.OptionsBuilder;
+import org.openjdk.jmh.runner.options.TimeValue;
+
+import java.util.Arrays;
+import java.util.BitSet;
+import java.util.Locale;
+import java.util.Random;
+import java.util.concurrent.TimeUnit;
+
+public class BenchmarkSupport {
+
+    // Making things deterministic
+    private static final Random RANDOM = new Random(0x5EAL);
+
+    // A generic number of test instances to use. Making sure it's small
+    // enough for L2 even for moderately sized objects, yet big enough
+    // to confuse JIT.
+    public static final int DEFAULT_POOL_SIZE = 1 << 12;
+
+    // A bitmask that would prevent an incremented number escaping
+    // DEFAULT_POOL_SIZE by a simple AND operation.
+    public static final int DEFAULT_MASK = DEFAULT_POOL_SIZE - 1;
+
+    // To prevent any kind of JIT assumption that a specific instance
+    // corresponds to a specific sample (shouldn't happen, but anyway),
+    // we need to walk them at a different rate. 2 is a bad choice,
+    // because it will discard half of the samples, but for every
+    // n = 2^k the value of 3 will be a divisor of either n + 1 or
+    // 2n + 1.
+    public static final int SAMPLE_STEP = 3;
+
+    private BenchmarkSupport() {
+    }
+
+    /**
+     * When you need to break existing patterns in an array.
+     */
+    public static <T> T[] shuffle(T[] values) {
+        for (int i = 0; i < values.length; i++) {
+            int target = RANDOM.nextInt(values.length);
+            T buffer = values[target];
+            values[target] = values[i];
+            values[i] = buffer;
+        }
+
+        return values;
+    }
+
+    /**
+     * Create a bit mask with [count] of [options] bits set. A simple method to select a
+     * combination of unique items using each bit as the presence mark for every item.
+     * @param count Number of bits to be set.
+     * @param options Number of all possible options.
+     */
+    public static BitSet selection(int count, int options) {
+        if (count > options) {
+            throw new IllegalArgumentException(
+                    "Requested count " + count + " is bigger than list of available options (" + options + ")");
+        }
+
+        if (count < 0) {
+            throw new IllegalArgumentException(
+                    "Number of selected options must be non-negative, " + count + " provided");
+        }
+
+        BitSet reply = new BitSet(options);
+
+        if (count == options) {
+            for (int i = 0; i < count; i++) {
+                reply.set(i);
+            }
+            return reply;
+        }
+
+        for (int flag = 0; flag < count; flag++) {
+            // with [remaining] positions left, we select a _disabled_
+            // bit N to install, meaning we have to pretend that all
+            // _enabled_ bits don't exist at all.
+
+            int remaining = options - flag;
+            int next = RANDOM.nextInt(remaining);
+            int skipped = 0;
+
+            for (int bit = 0; bit < options; bit++) {
+                if (reply.get(bit)) {
+                    skipped++;
+                    continue;
+                }
+
+                if (bit == skipped + next) {
+                    reply.set(bit);
+                    break;
+                }
+            }
+
+            // bit can be as much as options - 1 in the previous loop, thus >=
+            if (skipped + next >= options) {
+                String message = String.format(
+                        "Failed to set the bit: while looking to set option #%d out of total %d, disabled bit #%d should have been enabled, however, %d bits were skipped in %s",
+                        flag, options, next, skipped, reply);
+                throw new IllegalStateException(message);
+            }
+        }
+
+        return reply;
+    }
+
+    public static BitSet selection(int minimum, int maximum, int options) {
+        int count = minimum + RANDOM.nextInt(maximum - minimum + 1);
+        return selection(count, options);
+    }
+
+    public static BitSet selection(int options) {
+        return selection(RANDOM.nextInt(options + 1), options);
+    }
+
+    public static class ModeUniformDistribution {
+
+        private ModeUniformDistribution() {
+        }
+
+        /**
+         * <p>
+         * A silly mock distribution to test the reaction of the code to different input
+         * sizes. With the specified probability, the [mode] value will be returned,
+         * otherwise a value between [minimum] and [maximum] (inclusively, but without the
+         * mode) will be selected uniformly.
+         * </p>
+         *
+         * <p>
+         * The reason for not taking any nonsilly distribution is the generic case modeled
+         * here, a case with a distinct mode, but without the probability of other values
+         * falling off sharply with the distance from the mode. This allows both to see
+         * how code reacts to the position of that mode and to confuse the JIT regarding
+         * any assumptions (beyond mode value) and machinery like predictors. Using just a
+         * bell-like distribution here would either shrink the mode or reduce output to
+         * 3-5 values that would totally dominate everything else, so it requires a tight
+         * bell mixed with a uniform distribution to blend in all possibilities. This is
+         * exactly what this method does, with replacing the distribution by explicit
+         * boost for a specific value to keep things less error-prone (no one ever will
+         * plot these values or look at them in the debugger again).
+         * </p>
+         * @param minimum The minimum amount that can be selected, inclusive.
+         * @param maximum The maximum amount that can be selected, inclusive.
+         * @param mode The value that appears with increased probability.
+         * @param probability The probability of selecting the value instead of using
+         * uniform distribution.
+         * @return Value that differs from uniform distribution by a more often selected
+         * mode, according to passed probability.
+         */
+        public static int sample(int minimum, int maximum, int mode, double probability) {
+            if (mode < minimum || mode > maximum) {
+                throw new IllegalArgumentException(
+                        "Provided mode " + mode + " isn't in the min/max interval [" + minimum + ", " + maximum + "]");
+            }
+
+            if (RANDOM.nextDouble() < probability) {
+                return mode;
+            }
+
+            // excluding mode here, so upper bound is maximum, not maximum + 1
+            int selection = minimum + RANDOM.nextInt(maximum - minimum);
+
+            // compensating for absent mode
+            if (selection >= mode) {
+                return selection + 1;
+            }
+
+            return selection;
+        }
+
+    }
+
+    public static ChainedOptionsBuilder defaults() {
+        ChainedOptionsBuilder options = new OptionsBuilder().forks(1)
+            // 0.5m warmups
+            .warmupIterations(6)
+            .warmupTime(new TimeValue(5L, TimeUnit.SECONDS))
+            // 9.5m benchmarks
+            .measurementIterations(57)
+            .measurementTime(new TimeValue(5L, TimeUnit.SECONDS))
+            .mode(Mode.AverageTime)
+            .timeUnit(TimeUnit.NANOSECONDS)
+            .shouldDoGC(true)
+            .addProfiler(GCProfiler.class);
+
+        // Please forgive me
+        if (System.getProperty("os.name", "_fallback_").toLowerCase(Locale.ROOT).contains("linux")) {
+            options.addProfiler(LinuxPerfAsmProfiler.class).addProfiler(LinuxPerfNormProfiler.class);
+        }
+
+        return options;
+    }
+
+    /**
+     * Runs benchmarks in specified classes with defaults (10s iterations, 0.5m warmup,
+     * 9.5m benchmark, GC & perf profilers). As usual, be aware that defaults might not
+     * suit your case, use overrides when necessary.
+     */
+    public static void run(String[] patterns, @Nullable ChainedOptionsBuilder override) throws RunnerException {
+        ChainedOptionsBuilder options = defaults();
+
+        for (String pattern : patterns) {
+            options.include(pattern);
+        }
+
+        Options defaults = options.build();
+
+        new Runner(override == null ? defaults : override.parent(defaults).build()).run();
+    }
+
+    /**
+     * @see #run(String[], ChainedOptionsBuilder)
+     */
+    public static void run(String[] patterns) throws RunnerException {
+        run(patterns, null);
+    }
+
+    /**
+     * @see #run(String[], ChainedOptionsBuilder)
+     */
+    public static void run(Class<?>[] sources, @Nullable ChainedOptionsBuilder override) throws RunnerException {
+        String[] patterns = Arrays.stream(sources).map(Class::getCanonicalName).toArray(String[]::new);
+
+        run(patterns, override);
+    }
+
+    /**
+     * @see #run(String[], ChainedOptionsBuilder)
+     */
+    public static void run(Class<?> source, @Nullable ChainedOptionsBuilder override) throws RunnerException {
+        run(new Class<?>[] { source }, override);
+    }
+
+    /**
+     * @see #run(String[], ChainedOptionsBuilder)
+     */
+    public static void run(Class<?>... sources) throws RunnerException {
+        run(sources, null);
+    }
+
+    /**
+     * @see #run(String[], ChainedOptionsBuilder)
+     */
+    public static void run(Class<?> source) throws RunnerException {
+        run(source, null);
+    }
+
+    public static void main(String[] includes) throws RunnerException {
+        ChainedOptionsBuilder builder = defaults()
+            // Only 2m benchmarks, we're launching everything at once here
+            .measurementIterations(24)
+            .measurementTime(new TimeValue(5L, TimeUnit.SECONDS));
+
+        for (String pattern : includes) {
+            builder.include(pattern);
+        }
+
+        Options options = builder.build();
+
+        Runner runner = new Runner(options);
+
+        if (includes.length == 0) {
+            System.out.println("Specify benchmark patterns as CLI arguments");
+            runner.list();
+        }
+        else {
+            runner.run();
+        }
+    }
+
+}

benchmarks/benchmarks-core/src/jmh/java/io/micrometer/benchmark/core/instrument/MeterBenchmarks.java

@@ -0,0 +1,72 @@
+/*
+ * Copyright 2025 VMware, Inc.
+ *
+ * Licensed 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
+ *
+ * https://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 io.micrometer.benchmark.core.instrument;
+
+import io.micrometer.benchmark.BenchmarkSupport;
+import io.micrometer.benchmark.core.instrument.config.filter.FilterBenchmarkSupport;
+import io.micrometer.core.instrument.Meter;
+import io.micrometer.core.instrument.Tag;
+import org.openjdk.jmh.annotations.*;
+import org.openjdk.jmh.runner.RunnerException;
+
+import java.util.List;
+import java.util.concurrent.TimeUnit;
+
+public class MeterBenchmarks {
+
+    private MeterBenchmarks() {
+    }
+
+    @Fork(value = 1)
+    @Warmup(iterations = 6, time = 10, timeUnit = TimeUnit.SECONDS)
+    @Measurement(iterations = 54, time = 10, timeUnit = TimeUnit.SECONDS)
+    @BenchmarkMode(Mode.AverageTime)
+    @OutputTimeUnit(TimeUnit.NANOSECONDS)
+    @State(Scope.Benchmark)
+    public static class GetTags {
+
+        private static final int COUNT = BenchmarkSupport.DEFAULT_POOL_SIZE;
+
+        private static final int MASK = BenchmarkSupport.DEFAULT_MASK;
+
+        @Param({ "0", "1", "2", "4", "8", "16", "32", "64" })
+        public int mode;
+
+        private Meter.Id[] identifiers;
+
+        private int iteration;
+
+        @Setup
+        public void setUp() {
+            identifiers = FilterBenchmarkSupport.distributed(mode).limit(COUNT).toArray(Meter.Id[]::new);
+        }
+
+        @Benchmark
+        public List<Tag> baseline() {
+            return identifiers[iteration++ & MASK].getTags();
+        }
+
+        public static void main(String[] args) throws RunnerException {
+            BenchmarkSupport.run(GetTags.class);
+        }
+
+    }
+
+    public static void main(String[] args) throws RunnerException {
+        BenchmarkSupport.run(MeterBenchmarks.class);
+    }
+
+}