Eric Blake wrote:
On 08/04/2011 11:35 AM, Jim Fehlig wrote:
Discussed previously:
https://www.redhat.com/archives/libvir-list/2011-August/msg00166.html
The qemu migration speed default is 32MiB/s as defined in migration.c
/* Migration speed throttling */ static int64_t max_throttle = (32<< 20);
Too bad it's not publicly exposed in a header file, in case it ever changes.
There is no reason to throttle migration when targeting a file. For dump and save operations, set migration speed to unlimited prior to migration and restore to default value after migration. Default units is MB for migrate_set_speed monitor command, so (INT64_MAX / (1024 * 1024)) is used for unlimited migration speed.
Tested with both json and text monitors.
+++ b/src/qemu/qemu_migration.c @@ -2676,6 +2676,14 @@ qemuMigrationToFile(struct qemud_driver *driver, virDomainObjPtr vm, virCommandPtr cmd = NULL; int pipeFD[2] = { -1, -1 };
+ /* No need for qemu default of 32MiB/s when migrating to a file. + Default speed unit is MB, so set to unlimited with INT64_MAX / 1M. + Failure to change migration speed is not fatal. */ + if (qemuDomainObjEnterMonitorAsync(driver, vm, asyncJob) == 0) { + qemuMonitorSetMigrationSpeed(priv->mon, INT64_MAX / (1024 * 1024)); + qemuDomainObjExitMonitorWithDriver(driver, vm); + } +
This part is now fine.
if (qemuCapsGet(priv->qemuCaps, QEMU_CAPS_MIGRATE_QEMU_FD)&& (!compressor || pipe(pipeFD) == 0)) { /* All right! We can use fd migration, which means that qemu @@ -2783,6 +2791,12 @@ qemuMigrationToFile(struct qemud_driver *driver, virDomainObjPtr vm, ret = 0;
cleanup: + /* Restore migration speed to 32MiB/s default */ + if (qemuDomainObjEnterMonitorAsync(driver, vm, asyncJob) == 0) { + qemuMonitorSetMigrationSpeed(priv->mon, (32<< 20));
If we keep this hunk, I'd like to see this magic number as a #define earlier in the file, with the same justification as you gave in your commit comment about where it comes from (32Mbps in qemu's migration.c), so it becomes easier to replace the number and/or consistently use it elsewhere in the code if qemu ever changes.
Ok.
+ qemuDomainObjExitMonitorWithDriver(driver, vm); + }
I was first worried that this part means we have more issues. That is, the user can independently call virDomainMigrateSetMaxSpeed, and it seems like we should revert back to that value, rather than to the qemu default. But on further thought - guess what? After you complete migration to a file, the qemu process is (supposed to have) ended! There's no reason to restore migration speed after this point, because there's nothing further you can do with the qemu process; once the domain is later restored from the save file, you have created a new qemu process which is once again back at the default migration speed. That means we can effectively drop this hunk with no change in behavior.
Yeah, I debated about whether to even add this hunk, but was considering 'virsh dump --live ...' where the process still exists and could potentially be migrated to another host later.
Meanwhile, it raises independent issues - why do we have a write-only interface in virDomainMigrateSetMaxSpeed? Shouldn't we also be able to query the speed currently in use, and shouldn't the domain XML track the current migration speed?
AFAICT, qemu doesn't provide a way to query the speed. The monitor would need this addition before we could plumb it in libvirt. How would you like to proceed? Regards, Jim