performance-metrics: Add QCOW2 compressed read micro benchmark

Add micro_bench_qcow_compressed_read which reads clusters from a
zlib compressed QCOW2 image. Every cluster triggers decompression,
isolating the decompression overhead from the normal allocated cluster
read path.

Workloads: 128 and 256 clusters.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>

performance-metrics/src/main.rs

@@ -378,7 +378,7 @@ mod adjuster {
     }
 }
 
-const TEST_LIST: [PerformanceTest; 76] = [
+const TEST_LIST: [PerformanceTest; 78] = [
     PerformanceTest {
         name: "boot_time_ms",
         func_ptr: performance_boot_time,
@@ -1421,6 +1421,30 @@ const TEST_LIST: [PerformanceTest; 76] = [
         },
         unit_adjuster: adjuster::s_to_us,
     },
+    PerformanceTest {
+        name: "micro_block_qcow_compressed_read_128_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_compressed_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_compressed_read_256_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_compressed_read,
+        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

@@ -247,3 +247,27 @@ pub fn micro_bench_qcow_cow_write(control: &PerformanceTestControl) -> f64 {
 
     elapsed
 }
+
+/// Read num_ops clusters from a zlib compressed QCOW2 image.
+///
+/// Every cluster is stored compressed, so each read triggers
+/// decompression.  This isolates the decompression overhead from
+/// the normal allocated-cluster read path.
+///
+/// Returns the total read wall clock time in seconds.
+pub fn micro_bench_qcow_compressed_read(control: &PerformanceTestControl) -> f64 {
+    let num_ops = control.num_ops.expect("num_ops required") as usize;
+    let (_tmp, disk) = util::compressed_qcow_tempfile(num_ops);
+    let mut async_io = disk.new_async_io(1).expect("new_async_io failed");
+
+    let mut buf = vec![0u8; QCOW_CLUSTER_SIZE as usize];
+    let iovec = read_iovec(&mut buf);
+
+    let start = Instant::now();
+    submit_reads(async_io.as_mut(), num_ops, QCOW_CLUSTER_SIZE, &[iovec]);
+    let elapsed = start.elapsed().as_secs_f64();
+
+    drain_completions(async_io.as_mut(), num_ops);
+
+    elapsed
+}