perf: More Context benchmarks (#2707)

Author: bantonsson

opentelemetry-sdk/benches/context.rs

@@ -1,4 +1,4 @@
-use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion, Throughput};
+use criterion::{black_box, criterion_group, criterion_main, BenchmarkId, Criterion, Throughput};
 use opentelemetry::{
     global::BoxedTracer,
     trace::{
@@ -30,27 +30,57 @@ fn criterion_benchmark(c: &mut Criterion) {
             group.bench_function(
                 BenchmarkId::new("has_active_span", param.clone()),
                 |b| match api {
-                    Api::Alt => b.iter(|| Context::map_current(TraceContextExt::has_active_span)),
-                    Api::Spec => b.iter(|| Context::current().has_active_span()),
+                    Api::Alt => b.iter(has_active_span_alt),
+                    Api::Spec => b.iter(has_active_span_spec),
                 },
             );
             group.bench_function(
                 BenchmarkId::new("is_sampled", param.clone()),
                 |b| match api {
-                    Api::Alt => {
-                        b.iter(|| Context::map_current(|cx| cx.span().span_context().is_sampled()))
-                    }
-                    Api::Spec => b.iter(|| Context::current().span().span_context().is_sampled()),
+                    Api::Alt => b.iter(is_sampled_alt),
+                    Api::Spec => b.iter(is_sampled_spec),
                 },
             );
             group.bench_function(BenchmarkId::new("is_recording", param), |b| match api {
-                Api::Alt => b.iter(|| Context::map_current(|cx| cx.span().is_recording())),
-                Api::Spec => b.iter(|| Context::current().span().is_recording()),
+                Api::Alt => b.iter(is_recording_alt),
+                Api::Spec => b.iter(is_recording_spec),
             });
         }
     }
 }
 
+#[inline(never)]
+fn has_active_span_alt() {
+    let _ = black_box(Context::map_current(TraceContextExt::has_active_span));
+}
+
+#[inline(never)]
+fn has_active_span_spec() {
+    let _ = black_box(Context::current().has_active_span());
+}
+
+#[inline(never)]
+fn is_sampled_alt() {
+    let _ = black_box(Context::map_current(|cx| {
+        cx.span().span_context().is_sampled()
+    }));
+}
+
+#[inline(never)]
+fn is_sampled_spec() {
+    let _ = black_box(Context::current().span().span_context().is_sampled());
+}
+
+#[inline(never)]
+fn is_recording_alt() {
+    let _ = black_box(Context::map_current(|cx| cx.span().is_recording()));
+}
+
+#[inline(never)]
+fn is_recording_spec() {
+    let _ = black_box(Context::current().span().is_recording());
+}
+
 #[derive(Copy, Clone)]
 enum Api {
     /// An alternative way which may be faster than what the spec recommends.

opentelemetry/Cargo.toml

@@ -57,6 +57,11 @@ harness = false
 name = "anyvalue"
 harness = false
 
+[[bench]]
+name = "context_attach"
+harness = false
+required-features = ["tracing"]
+
 [lib]
 bench = false
 

opentelemetry/benches/context_attach.rs

@@ -0,0 +1,122 @@
+use criterion::{
+    black_box, criterion_group, criterion_main, measurement::WallTime, BenchmarkGroup, BenchmarkId,
+    Criterion, Throughput,
+};
+use opentelemetry::{
+    trace::{SpanContext, TraceContextExt},
+    Context,
+};
+
+// Run this benchmark with:
+// cargo bench --bench current_context
+
+fn criterion_benchmark(c: &mut Criterion) {
+    let span_context = Context::new().with_remote_span_context(SpanContext::empty_context());
+    let contexts = vec![
+        ("empty_cx", Context::new()),
+        ("single_value_cx", Context::new().with_value(Value(4711))),
+        ("span_cx", span_context),
+    ];
+    for (name, cx) in contexts {
+        single_cx_scope(&mut group(c), name, &cx);
+        nested_cx_scope(&mut group(c), name, &cx);
+        overlapping_cx_scope(&mut group(c), name, &cx);
+    }
+}
+
+fn single_cx_scope(
+    group: &mut BenchmarkGroup<'_, WallTime>,
+    context_type: &str,
+    context: &Context,
+) {
+    let _restore = Context::current().attach();
+    group.throughput(Throughput::Elements(1)).bench_function(
+        BenchmarkId::new("single_cx_scope", context_type),
+        |b| {
+            b.iter_batched(
+                || context.clone(),
+                |cx| {
+                    single_cx(cx);
+                },
+                criterion::BatchSize::SmallInput,
+            );
+        },
+    );
+}
+
+#[inline(never)]
+fn single_cx(cx: Context) {
+    let cx = black_box(cx.attach());
+    let _ = black_box(dummy_work());
+    drop(cx);
+}
+
+fn nested_cx_scope(group: &mut BenchmarkGroup<'_, WallTime>, cx_type: &str, context: &Context) {
+    let _restore = Context::current().attach();
+    group.throughput(Throughput::Elements(1)).bench_function(
+        BenchmarkId::new("nested_cx_scope", cx_type),
+        |b| {
+            b.iter_batched(
+                || (context.clone(), context.clone()),
+                |(cx1, cx2)| {
+                    nested_cx(cx1, cx2);
+                },
+                criterion::BatchSize::SmallInput,
+            );
+        },
+    );
+}
+
+#[inline(never)]
+fn nested_cx(cx1: Context, cx2: Context) {
+    let outer = black_box(cx1.attach());
+    let inner = black_box(cx2.attach());
+    let _ = black_box(dummy_work());
+    drop(inner);
+    drop(outer);
+}
+
+fn overlapping_cx_scope(
+    group: &mut BenchmarkGroup<'_, WallTime>,
+    cx_type: &str,
+    context: &Context,
+) {
+    let _restore = Context::current().attach();
+    group.throughput(Throughput::Elements(1)).bench_function(
+        BenchmarkId::new("overlapping_cx_scope", cx_type),
+        |b| {
+            b.iter_batched(
+                || (context.clone(), context.clone()),
+                |(cx1, cx2)| {
+                    overlapping_cx(cx1, cx2);
+                },
+                criterion::BatchSize::SmallInput,
+            );
+        },
+    );
+}
+
+#[inline(never)]
+fn overlapping_cx(cx1: Context, cx2: Context) {
+    let outer = cx1.attach();
+    let inner = cx2.attach();
+    let _ = black_box(dummy_work());
+    drop(outer);
+    drop(inner);
+}
+
+#[inline(never)]
+fn dummy_work() -> i32 {
+    black_box(1 + 1)
+}
+
+fn group(c: &mut Criterion) -> BenchmarkGroup<WallTime> {
+    c.benchmark_group("context_attach")
+}
+
+#[derive(Debug, PartialEq)]
+struct Value(i32);
+
+criterion_group!(benches, criterion_benchmark);
+
+criterion_main!(benches);