1 files changed, 94 insertions, 0 deletions
diff --git a/vespalib/src/tests/cpu_usage/cpu_usage_test.cpp b/vespalib/src/tests/cpu_usage/cpu_usage_test.cpp
new file mode 100644
index 00000000000..c8835d82cd8
--- /dev/null
+++ b/vespalib/src/tests/cpu_usage/cpu_usage_test.cpp
@@ -0,0 +1,94 @@
+// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+#include <vespa/vespalib/util/cpu_usage.h>
+#include <vespa/vespalib/util/benchmark_timer.h>
+#include <vespa/vespalib/testkit/test_kit.h>
+
+#include <thread>
+
+using namespace vespalib;
+
+bool verbose = false;
+size_t loop_cnt = 10;
+double budget = 0.25;
+
+using Sampler = vespalib::cpu_usage::ThreadSampler;
+
+//-----------------------------------------------------------------------------
+
+void be_busy(duration d) {
+    if (d > 0ms) {
+        volatile int tmp = 123;
+        auto t0 = steady_clock::now();
+        while ((steady_clock::now() - t0) < d) {
+            for (int i = 0; i < 1000; ++i) {
+                tmp = (tmp + i);
+                tmp = (tmp - i);
+            }
+        }
+    }
+}
+
+std::vector<duration> sample(const std::vector<Sampler*> &list) {
+    std::vector<duration> result;
+    result.reserve(list.size());
+    for (Sampler *sampler: list) {
+        result.push_back(sampler->sample());
+    }
+    return result;
+}
+
+//-----------------------------------------------------------------------------
+
+TEST_MT_F("require that external thread-based CPU usage sampling works", 5, std::vector<Sampler*>(4, nullptr)) {
+    if (thread_id == 0) {
+        TEST_BARRIER(); // #1
+        auto t0 = steady_clock::now();
+        std::vector<duration> pre_usage = sample(f1);
+        TEST_BARRIER(); // #2
+        TEST_BARRIER(); // #3
+        auto t1 = steady_clock::now();
+        std::vector<duration> post_usage = sample(f1);
+        TEST_BARRIER(); // #4
+        double wall = to_s(t1 - t0);
+        std::vector<double> load(4, 0.0);
+        for (size_t i = 0; i < 4; ++i) {
+            load[i] = to_s(post_usage[i] - pre_usage[i]) / wall;
+        }
+        EXPECT_GREATER(load[3], load[0]);
+        fprintf(stderr, "loads: { %.2f, %.2f, %.2f, %.2f }\n", load[0], load[1], load[2], load[3]);
+    } else {
+        int idx = (thread_id - 1);
+        Sampler sampler;
+        f1[idx] = &sampler;
+        TEST_BARRIER(); // #1
+        TEST_BARRIER(); // #2
+        for (size_t i = 0; i < loop_cnt; ++i) {
+            be_busy(std::chrono::milliseconds(idx));
+        }
+        TEST_BARRIER(); // #3
+        TEST_BARRIER(); // #4
+    }
+}
+
+//-----------------------------------------------------------------------------
+
+TEST("measure thread CPU clock overhead") {
+    Sampler sampler;
+    duration d;
+    double min_time_us = BenchmarkTimer::benchmark([&d, &sampler]() noexcept { d = sampler.sample(); }, budget) * 1000000.0;
+    fprintf(stderr, "approx overhead per sample (thread CPU clock): %f us\n", min_time_us);
+}
+
+//-----------------------------------------------------------------------------
+
+int main(int argc, char **argv) {
+    TEST_MASTER.init(__FILE__);
+    if ((argc == 2) && (argv[1] == std::string("verbose"))) {
+        verbose = true;
+        loop_cnt = 1000;
+        budget = 5.0;
+    }
+    TEST_RUN_ALL();
+    return (TEST_MASTER.fini() ? 0 : 1);
+}