performance-metrics: Add QCOW2 async write micro benchmark

Add micro_bench_qcow_async_write which writes clusters into an
empty QCOW2 image through the QcowAsync io_uring path. Writes
in QcowAsync are synchronous due to COW metadata allocation, so
this measures the write path overhead through the async code path.

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

performance-metrics/src/main.rs

@@ -378,7 +378,7 @@ mod adjuster {
     }
 }
 
-const TEST_LIST: [PerformanceTest; 94] = [
+const TEST_LIST: [PerformanceTest; 96] = [
     PerformanceTest {
         name: "boot_time_ms",
         func_ptr: performance_boot_time,
@@ -1637,6 +1637,30 @@ const TEST_LIST: [PerformanceTest; 94] = [
         },
         unit_adjuster: adjuster::s_to_us,
     },
+    PerformanceTest {
+        name: "micro_block_qcow_async_write_128_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_async_write,
+        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_async_write_256_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_async_write,
+        control: PerformanceTestControl {
+            test_timeout: 10,
+            test_iterations: 20,
+            warmup_iterations: 5,
+            num_ops: Some(256),
+            ..PerformanceTestControl::default()
+        },
+        unit_adjuster: adjuster::s_to_us,
+    },
 ];
 
 fn run_test_with_timeout(

performance-metrics/src/micro_bench_block.rs

@@ -502,3 +502,28 @@ pub fn micro_bench_qcow_async_compressed_read(control: &PerformanceTestControl)
     drain_async_completions(async_io.as_mut(), num_ops);
     start.elapsed().as_secs_f64()
 }
+
+/// Write num_ops clusters into an empty QCOW2 image through the
+/// QcowAsync io_uring path.
+///
+/// Writes in QcowAsync are synchronous (COW metadata allocation must
+/// complete before the host offset is known), so this measures the
+/// write path overhead through the async code path.
+///
+/// Returns the total write wall clock time in seconds.
+pub fn micro_bench_qcow_async_write(control: &PerformanceTestControl) -> f64 {
+    let num_ops = control.num_ops.expect("num_ops required") as usize;
+    let (_tmp, disk) = util::empty_qcow_async_tempfile(num_ops);
+    let mut async_io = disk
+        .new_async_io(num_ops as u32)
+        .expect("new_async_io failed");
+
+    let buf = vec![0xA5u8; QCOW_CLUSTER_SIZE as usize];
+    let iovec = write_iovec(&buf);
+
+    let start = Instant::now();
+    submit_writes(async_io.as_mut(), num_ops, QCOW_CLUSTER_SIZE, &[iovec]);
+
+    drain_async_completions(async_io.as_mut(), num_ops);
+    start.elapsed().as_secs_f64()
+}