Description of problem:

The current implementation of the qemuMigrationWaitForCompletion() function will sleep for 50 ms, even when a migration has already finished. This sleep is unnecessary and only adds up to the guest pause time. We would like to remove this, to reduce the VM pause time. 


Version-Release number of selected component (if applicable):

The latest version of libvirt from git has this bug, with the most recent commit at a45ef3a9cdfe7a1f847da68ab3d7313595c0bc05. 


How reproducible:

It is within the source code. 

Steps to Reproduce:
1.
2.
3.

Actual results:

An additional sleep of 50 ms is done even when a snapshot completes. 

Expected results:

We should avoid another sleep of 50 ms when a snapshot completes. 

Additional info:

Here is my patch to fix it. 

diff --git a/src/qemu/qemu_migration.c b/src/qemu/qemu_migration.c
index 29f5173..af44b69 100644
--- a/src/qemu/qemu_migration.c
+++ b/src/qemu/qemu_migration.c
@@ -2389,6 +2389,12 @@ qemuMigrationWaitForCompletion(virQEMUDriverPtr driver,
         /* Poll every 50ms for progress & to allow cancellation */
         struct timespec ts = { .tv_sec = 0, .tv_nsec = 50 * 1000 * 1000ull };

+        virObjectUnlock(vm);
+
+        nanosleep(&ts, NULL);
+
+        virObjectLock(vm);
+
         if (qemuMigrationUpdateJobStatus(driver, vm, job, asyncJob) == -1)
             break;

@@ -2407,11 +2413,6 @@ qemuMigrationWaitForCompletion(virQEMUDriverPtr driver,
             break;
         }

-        virObjectUnlock(vm);
-
-        nanosleep(&ts, NULL);
-
-        virObjectLock(vm);
     }

     if (jobInfo->type == VIR_DOMAIN_JOB_COMPLETED) {