RE: [MERGE, ACS41] Merge branch cloud-agent-with-openvswitch

[Sudha Ponnaganti]

Hi Hugo,

Would it be possible to have test cases and test results to be posted to 
cwiki/QA for the following stories. Seems like you have done some validation. 
If there is anyone available from your org that can do some sanity validation 
and submit results on ACS Master that would be great. 

Also do you have any thoughts around regression for future changes that you 
would be making. I would like to call on community to see if anyone can take up 
QA for these features.  I am a little concern as this is lot of code and Wido 
indicates that he would not be able to actually test the logic. 

All stories are related to Nicira integration

CLOUDSTACK-934, 
CLOUDSTACK - 726, 
CLOUDSTACK-727, 
CLOUDSTACK-101

Thanks
/sudha

-----Original Message-----
From: Wido den Hollander [mailto:w...@widodh.nl] 
Sent: Thursday, January 17, 2013 3:47 AM
To: cloudstack-dev@incubator.apache.org
Subject: Re: [MERGE, ACS41] Merge branch cloud-agent-with-openvswitch

On 01/15/2013 10:32 AM, Hugo Trippaers wrote:
> Hey all,
>
> I would like to merge my branch with master. This branch deals with the items 
> reported in tickets CLOUDSTACK-934, CLOUDSTACK-727, CLOUDSTACK-101. Basically 
> the patch introduces support for dealing with bridges and interfaces on KVM 
> hypervisors running openvswitch. Support is included for both vlan and Nicira 
> NVP based networks.
>
> During development some bugs were discovered, following the commits 
> for those fixes
> - 1ceecc92c88c84b99c99547f5c086657ec26f8d7 Fix logic error in 
> ConfigurationServerImpl.java
> - 87fe64695305d6a2c505f757ab7607b765c313b7 NPE in 
> KVMStoragePoolManager
>
> Testing done
> - mvn clean test
> - setup clean CloudStack installation with KVM server using vlan based 
> isolation. Deploy systemvms and several instances on isolated networks and 
> test connectivity.
> - reuse existing CloudStack installation with Nicira NVP. Add cluster with 
> KVM hosts. Deploy several instances and networks and test connectivity.
>
> Testing not done
> - basic networking based setup, I figured that if advanced networks work, 
> basic networks should work as well.
>
> Risk
> - minimal risk, anyone willing to use this feature needs to explicitly enable 
> this by making changes to agent.properties for the cloud-agent. By default 
> the current bridge drivers will be used.
>
> Database changes
> - none
>
> Documentation tracked in a separate ticket: CLOUDSTACK-977
>
> Cheers,
>

I've been reviewing this, but I haven't been able to test the code since I 
don't have anything for this running.

I did find some things which I report in random order :) I basically went from 
commit to commit in the branch and checked it.

Globally I've found some indentation things, like:

         switch (_bridgeType) {
         case OPENVSWITCH:
             getOvsPifs();
             break;
         case NATIVE:
         default:
             getPifs();
             break;
         }

Isn't the indentation missing there for "case"?

Further in for example OvsVifDriver there are some blank lines with spaces or 
trailing spaces.

You also seem to write if/else-contitions in various ways, like:

if (....) {

}
else {

}

I thought our code convention said we should use:

if (....) {

} else {

}

Not a very big deal, but just to prevent other people from taking over the 
"habbits".

In OvsVifDriver there is a method called: "deleteExitingLinkLocalRoutTable"

Isn't that a typo? Shouldn't 'Rout' be 'Route'?

What I don't get in LibvirtVMDef is this:

if (_virtualPortType != null) {
   netBuilder.append("<virtualport type='" + _virtualPortType + "'>\n");
   if (_virtualPortInterfaceId != null) {
     netBuilder.append("<parameters interfaceid='" + _virtualPortInterfaceId + 
"'/>\n");
   }
   netBuilder.append("</virtualport>\n");
}

Why should a virtualport be added if the interfaceId is null? Should the first 
if already check if both the type AND id are set? Or can the ID be set after 
the VM is running? (In OvsVifDriver?)

If so, couldn't that be an issue if the VM is migrated and this XML definition 
isn't properly transferred to the other host?

When setting the VLAN tag in the XML definition you do this check:

if (_vlanTag != -1) {
   netBuilder.append("<vlan trunk='no'>\n<tag id='" + _vlanTag + 
"'/>\n</vlan>"); }

It's my understanding that VLAN ID 0 and 4095 are reserved VLAN IDs, so 
shouldn't it be:

if (_vlanTag > 0) {
...
}

This is what I found while doing my first review. Like I said, I can't test the 
real logic now.

Wido

> Hugo
>