performance-metrics: Add qcow2 random read micro benchmark

Add micro_bench_qcow_random_read which reads clusters from a
prepopulated qcow2 image in a deterministic pseudo-random order.
Unlike the sequential read benchmark, this exercises L2 cache miss
and eviction behaviour under random access patterns.

Uses Fisher-Yates shuffle with DefaultHasher for reproducible
permutation across runs.

Two TEST_LIST entries: micro_block_qcow_random_read_128_us and
micro_block_qcow_random_read_256_us with 128 and 256 cluster
workloads.

Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>

performance-metrics/src/main.rs

@@ -378,7 +378,7 @@ mod adjuster {
     }
 }
 
-const TEST_LIST: [PerformanceTest; 70] = [
+const TEST_LIST: [PerformanceTest; 72] = [
     PerformanceTest {
         name: "boot_time_ms",
         func_ptr: performance_boot_time,
@@ -1277,6 +1277,30 @@ const TEST_LIST: [PerformanceTest; 70] = [
         },
         unit_adjuster: adjuster::s_to_us,
     },
+    PerformanceTest {
+        name: "micro_block_qcow_random_read_128_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_random_read,
+        control: PerformanceTestControl {
+            test_timeout: 10,
+            test_iterations: 20,
+            warmup_iterations: 5,
+            num_ops: Some(128),
+            ..PerformanceTestControl::default()
+        },
+        unit_adjuster: adjuster::s_to_us,
+    },
+    PerformanceTest {
+        name: "micro_block_qcow_random_read_256_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_random_read,
+        control: PerformanceTestControl {
+            test_timeout: 10,
+            test_iterations: 20,
+            warmup_iterations: 5,
+            num_ops: Some(256),
+            ..PerformanceTestControl::default()
+        },
+        unit_adjuster: adjuster::s_to_us,
+    },
     PerformanceTest {
         name: "micro_block_qcow_write_128_us",
         func_ptr: micro_bench_block::micro_bench_qcow_write,

performance-metrics/src/micro_bench_block.rs

@@ -16,8 +16,8 @@ use block::raw_async_aio::RawFileAsyncAio;
 
 use crate::PerformanceTestControl;
 use crate::util::{
-    self, BLOCK_SIZE, QCOW_CLUSTER_SIZE, drain_completions, read_iovec, submit_reads,
-    submit_writes, write_iovec,
+    self, BLOCK_SIZE, QCOW_CLUSTER_SIZE, deterministic_permutation, drain_completions, read_iovec,
+    submit_reads, submit_writes, write_iovec,
 };
 
 /// Submit num_ops AIO writes, wait for them all to land, then time
@@ -81,6 +81,40 @@ pub fn micro_bench_qcow_read(control: &PerformanceTestControl) -> f64 {
     elapsed
 }
 
+/// Read num_ops clusters from a prepopulated qcow2 image in random order.
+///
+/// Unlike micro_bench_qcow_read which reads sequentially, this shuffles
+/// the cluster indices to exercise L2 cache miss and eviction behaviour
+/// under random access patterns.
+///
+/// Returns the total read wall clock time in seconds.
+pub fn micro_bench_qcow_random_read(control: &PerformanceTestControl) -> f64 {
+    let num_ops = control.num_ops.expect("num_ops required") as usize;
+    let (_tmp, disk) = util::qcow_tempfile(num_ops);
+    let mut async_io = disk.new_async_io(1).expect("new_async_io failed");
+
+    let indices = deterministic_permutation(num_ops);
+
+    let mut buf = vec![0u8; QCOW_CLUSTER_SIZE as usize];
+    let iovec = read_iovec(&mut buf);
+
+    let start = Instant::now();
+    for (seq, &cluster_idx) in indices.iter().enumerate() {
+        async_io
+            .read_vectored(
+                (cluster_idx as u64 * QCOW_CLUSTER_SIZE) as libc::off_t,
+                &[iovec],
+                seq as u64,
+            )
+            .expect("read_vectored failed");
+    }
+    let elapsed = start.elapsed().as_secs_f64();
+
+    drain_completions(async_io.as_mut(), num_ops);
+
+    elapsed
+}
+
 /// Write num_ops clusters into an empty qcow2 image through the
 /// QcowSync async_io path and time the total write_vectored wall clock.
 ///