obs/ash: add BenchmarkSetWorkState microbenchmark

Add a microbenchmark for the SetWorkState + clearWorkState hot path.
This complements the existing end-to-end BenchmarkASH in pkg/bench by
isolating the activeWorkStates sync.Map contention that degrades on
high-core machines. Varying GOMAXPROCS via -test.cpu shows how
per-operation latency scales with core count due to cache coherence
traffic on the map's internal mutex.

Fixes: #164683

Release note: None

Co-Authored-By: roachdev-claude <roachdev-claude-bot@cockroachlabs.com>

Author: alyshanjahani-crl

pkg/obs/ash/BUILD.bazel

@@ -42,6 +42,7 @@ go_test(
     name = "ash_test",
     srcs = [
         "aggregate_test.go",
+        "bench_test.go",
         "buffer_test.go",
         "report_test.go",
         "sampler_test.go",

pkg/obs/ash/bench_test.go

@@ -0,0 +1,49 @@
+// Copyright 2026 The Cockroach Authors.
+//
+// Use of this software is governed by the CockroachDB Software License
+// included in the /LICENSE file.
+
+package ash
+
+import (
+	"testing"
+
+	"github.com/cockroachdb/cockroach/pkg/roachpb"
+)
+
+// BenchmarkSetWorkState measures the per-operation cost of a
+// SetWorkState + clearWorkState cycle — the pair of calls every
+// registered goroutine makes on the ASH hot path.
+//
+// activeWorkStates is a global sync.Map. On high-core machines,
+// concurrent Stores from different cores cause cache-line
+// invalidations on the map's internal mutex and dirty-map pointer,
+// degrading throughput as core count grows. This benchmark
+// quantifies that degradation by varying GOMAXPROCS via -test.cpu.
+// RunParallel spawns one goroutine per core, which is the right
+// model: contention scales with cores executing simultaneously,
+// not goroutines queued in the scheduler.
+//
+// See #168289 for the sharded-map proposal to address this.
+//
+// Example:
+//
+//	./dev bench pkg/obs/ash -f BenchmarkSetWorkState --test-args='-test.cpu=1,4,16,64'
+func BenchmarkSetWorkState(b *testing.B) {
+	enabled.Store(true)
+	defer enabled.Store(false)
+
+	tenantID := roachpb.MustMakeTenantID(5)
+	info := WorkloadInfo{WorkloadID: 12345}
+
+	b.ReportAllocs()
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			clear := SetWorkState(tenantID, info, WorkCPU, "BenchmarkOp")
+			clear()
+		}
+	})
+
+	// Drain retired states so subsequent benchmarks start clean.
+	reclaimRetiredWorkStates()
+}