[
https://issues.apache.org/jira/browse/CLOUDSTACK-9114?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16467324#comment-16467324
]
ASF GitHub Bot commented on CLOUDSTACK-9114:
--------------------------------------------
rhtyd commented on a change in pull request #2508: CLOUDSTACK-9114: Reduce VR
downtime during network restart
URL: https://github.com/apache/cloudstack/pull/2508#discussion_r186706446
##########
File path: server/src/com/cloud/network/vpc/VpcManagerImpl.java
##########
@@ -2435,4 +2444,43 @@ public boolean isSrcNatIpRequired(long vpcOfferingId) {
final Map<Network.Service, Set<Network.Provider>>
vpcOffSvcProvidersMap = getVpcOffSvcProvidersMap(vpcOfferingId);
return
vpcOffSvcProvidersMap.get(Network.Service.SourceNat).contains(Network.Provider.VPCVirtualRouter);
}
+
+ private boolean rollingRestartVpc(final Vpc vpc, final ReservationContext
context) throws ResourceUnavailableException, ConcurrentOperationException,
InsufficientCapacityException {
+ s_logger.debug("Performing rolling restart of routers of VPC " + vpc);
+ _ntwkMgr.destroyExpendableRouters(_routerDao.listByVpcId(vpc.getId()),
context);
+
+ final DeployDestination dest = new
DeployDestination(_dcDao.findById(vpc.getZoneId()), null, null, null);
+ final List<DomainRouterVO> oldRouters =
_routerDao.listByVpcId(vpc.getId());
+
+ // Create a new router
+ if (oldRouters.size() > 0) {
+ vpc.setRollingRestart(true);
+ }
+ startVpc(vpc, dest, context);
+ if (oldRouters.size() > 0) {
Review comment:
@rafaelweingartner if there are any existing old routers, we do rolling
restart which is to deploy a new VR and then destroy old VRs. The
`setRollingRestart` is a way to tell lower facade to treat deployment like for
redundant routers, i.e. expect upto 2 VRs at a given point which is seen in
changes in `VirtualRouterElement.java`. The startVpc will call the facade to
deploy suitable number of routers. If there is only one old VPC VR, and if
rolling restart is `true` this will cause the lower layers to deploy a new VPC
VR. After that, we set the rolling restart to false which works great both in
case of rVR (expected VRs is 2) and non-rVRs (expected VRs becomes 1), and
finally we destroy old VR. If the VPC has rVRs, a new backup router is created
when `startVpc` is called otherwise this will simply reconfigure existing VRs.
----------------------------------------------------------------
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:
us...@infra.apache.org
> restartnetwork with cleanup should not update/restart both routers at once
> --------------------------------------------------------------------------
>
> Key: CLOUDSTACK-9114
> URL: https://issues.apache.org/jira/browse/CLOUDSTACK-9114
> Project: CloudStack
> Issue Type: Improvement
> Security Level: Public(Anyone can view this level - this is the
> default.)
> Reporter: Wei Zhou
> Assignee: Wei Zhou
> Priority: Major
>
> for now, restartnetwork with cleanup will stop both RVRs at first, then start
> two new RVRs.
> to reduce the downtime of network, we'd better restart the RVRs one by one.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
