[
https://issues.apache.org/jira/browse/CLOUDSTACK-10305?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16376953#comment-16376953
]
ASF GitHub Bot commented on CLOUDSTACK-10305:
---------------------------------------------
rhtyd closed pull request #2466: CLOUDSTACK-10305: Rare race condition in KVM
migration
URL: https://github.com/apache/cloudstack/pull/2466
This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:
As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):
diff --git
a/plugins/hypervisors/kvm/src/com/cloud/hypervisor/kvm/resource/wrapper/LibvirtMigrateCommandWrapper.java
b/plugins/hypervisors/kvm/src/com/cloud/hypervisor/kvm/resource/wrapper/LibvirtMigrateCommandWrapper.java
index 30f0e20ddd9..ad32759a517 100644
---
a/plugins/hypervisors/kvm/src/com/cloud/hypervisor/kvm/resource/wrapper/LibvirtMigrateCommandWrapper.java
+++
b/plugins/hypervisors/kvm/src/com/cloud/hypervisor/kvm/resource/wrapper/LibvirtMigrateCommandWrapper.java
@@ -172,13 +172,21 @@ Use VIR_DOMAIN_XML_SECURE (value = 1) prior to v1.0.0.
// pause vm if we meet the vm.migrate.pauseafter threshold and
not already paused
final int migratePauseAfter =
libvirtComputingResource.getMigratePauseAfter();
- if (migratePauseAfter > 0 && sleeptime > migratePauseAfter &&
dm.getInfo().state == DomainState.VIR_DOMAIN_RUNNING ) {
- s_logger.info("Pausing VM " + vmName + " due to property
vm.migrate.pauseafter setting to " + migratePauseAfter+ "ms to complete
migration");
+ if (migratePauseAfter > 0 && sleeptime > migratePauseAfter) {
+ DomainState state = null;
try {
- dm.suspend();
+ state = dm.getInfo().state;
} catch (final LibvirtException e) {
- // pause could be racy if it attempts to pause right
when vm is finished, simply warn
- s_logger.info("Failed to pause vm " + vmName + " : " +
e.getMessage());
+ s_logger.info("Couldn't get VM domain state after " +
sleeptime + "ms: " + e.getMessage());
+ }
+ if (state != null && state ==
DomainState.VIR_DOMAIN_RUNNING) {
+ try {
+ s_logger.info("Pausing VM " + vmName + " due to
property vm.migrate.pauseafter setting to " + migratePauseAfter + "ms to
complete migration");
+ dm.suspend();
+ } catch (final LibvirtException e) {
+ // pause could be racy if it attempts to pause
right when vm is finished, simply warn
+ s_logger.info("Failed to pause vm " + vmName + " :
" + e.getMessage());
+ }
}
}
}
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> Rare race condition in KVM migration
> ------------------------------------
>
> Key: CLOUDSTACK-10305
> URL: https://issues.apache.org/jira/browse/CLOUDSTACK-10305
> Project: CloudStack
> Issue Type: Bug
> Security Level: Public(Anyone can view this level - this is the
> default.)
> Reporter: Nicolas Vazquez
> Priority: Major
> Fix For: 4.11.1.0
>
>
> There is a race condition in the monitoring of the migration process on KVM.
> If the monitor wakes up in the tight window after the migration succeeds, but
> before the migration thread terminates, the monitor will get a
> LibvirtException “Domain not found: no domain with matching uuid” when
> checking on the migration status. This in turn causes CloudStack to sync the
> VM state to stop, in which it issues a defensive StopCommand to ensure it is
> correctly synced.
> This is issue is reported under these conditions:
> * VM migration has physically finished
> * Migration thread executor is still not marked as terminated
> From the migration monitoring process, as VM has been migrated, that domain
> cannot be found (we should get the destination domain from the migration
> thread later) and the LibvirtException is thrown: {{LibvirtException “Domain
> not found: no domain with matching uuid”}}.
> So basically the idea is just logging the issue but not suspending the VM as
> the migration has already been performed, but the migration thread is not
> marked as finished. Once the migrarion thread is marked as finished, the
> migration monitoring process (while loop) ends and the migration command
> wrapper would continue its execution.
> Regarding your last question, we won't need to suspend the VM in this case as
> the migration has already been performed. VM suspension will take place
> whenever the migration is in progress and those conditions are met
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
