Chiradeep,

I have made the change in my feature branch.  I will submit it as part of 
larger patch, S3 secondary storage, that I plan submit later this week or early 
next.

Thanks,
-John

On Oct 29, 2012, at 4:31 PM, Chiradeep Vittal <chiradeep.vit...@citrix.com> 
wrote:

> Yes, 0. 
> I think that the discrepancy is because sysctl.conf is modified too late
> in the game.
> 0. SSVM boots for first time
> 1. cloud-early-config figures out that scripts need to be patched
> 2. Scripts are patched and reboot is initiated. Sysctl is not modified yet
> 3. SSVM boots, steps 1-2 do not take place, figures out it is an SSVM
> 4. cloud-early-config modifies /etc/sysctl.conf, but DOES NOT execute
> sysctl -w
> 
> Hence the runtime value of rp_filter remains 1 while the config file says
> "0".
> --
> Chiradeep
> 
> On 10/29/12 1:00 PM, "John Burwell" <jburw...@basho.com> wrote:
> 
>> Chiradeep,
>> 
>> Currently, net.ipv4.conf.default.rp_filer is set to 1 in
>> systemvm/debian/config/etc/sysctl.conf.  Should it be modified to be 0?
>> 
>> Thanks,
>> -John
>> 
>> On Oct 4, 2012, at 6:09 PM, Chiradeep Vittal
>> <chiradeep.vit...@citrix.com> wrote:
>> 
>>> It is disabled in sysctl.conf, not sure how it gets re-enabled. See
>>> patches/systemvm/debian/config/etc/init.d/cloud-early-config (function
>>> disable_rpfilter).
>>> Perhaps it is interface-specific rather than "all".
>>> 
>>> On 10/4/12 2:39 PM, "John Burwell" <jburw...@basho.com> wrote:
>>> 
>>>> Ahmad,
>>>> 
>>>> You were correct on the rp_filter issue.  Once disabled, the SSVM was
>>>> able to connect outbound to S3, as well as, any host reachable from
>>>> devcloud.  I noticed that rp_filter is disabled in sysctl.conf yet it
>>>> is
>>>> somehow being enabled at runtime.  Is this behavior intended?
>>>> 
>>>> Thanks,
>>>> -John
>>>> 
>>>> On Oct 4, 2012, at 1:07 PM, Ahmad Emneina <ahmad.emne...@citrix.com>
>>>> wrote:
>>>> 
>>>>> On 10/4/12 9:16 AM, "John Burwell" <jburw...@basho.com> wrote:
>>>>> 
>>>>>> Kelcey,
>>>>>> 
>>>>>> I am a bit confused about how secstorage.allowed.internal.sites is
>>>>>> used
>>>>>> which stems to lack of knowledge regarding the devcloud network
>>>>>> configuration.  Also, is there documentation available for setting up
>>>>>> such a NAT? 
>>>>>> 
>>>>>> As a point of clarification to my original question, I am working in
>>>>>> the
>>>>>> devcloud environment (using the OVA downloaded from the wiki) where I
>>>>>> need to get the SSVM to connect to S3 or to a local VirtualBox VM
>>>>>> running
>>>>>> an S3-compatible object store.  Thus far, I have been unable to get
>>>>>> devcloud to bring up a second NIC on a host-only network.  I have
>>>>>> attempted to setup an advanced network configuration as follows:
>>>>>> 
>>>>>> Physical Network with VLAN isolation method
>>>>>> Management Server: 10.0.2.15 -> Gateway: 10.0.2.2
>>>>>> Storage Network: 10.0.2.50-10.0.2.59 -> Gateway 10.0.2.2 on VLAN0
>>>>>> Management Network: 10.0.2.200-10.0.2.220 -> Gateway 10.0.2.2
>>>>>> Public Network: 10.0.2.100-10.0.2.199 -> VLAN0
>>>>> 
>>>>> The issue that gets created here is you get system vm's that are
>>>>> multi-homed. Your system vm's get a nic (leg) on each network... But
>>>>> that
>>>>> network is one and the same. Why this is an issue is rp_filter is
>>>>> enabled
>>>>> by default on the system vm's, message comes in on one of those nics,
>>>>> but
>>>>> it's default route out is another nic... Thus blocking the response.
>>>>> 
>>>>> Ideally you'd use a basic zone for this kind of configuration, or else
>>>>> you'll end up having to log into the system vm's every time a new one
>>>>> is
>>>>> spawned and disabling rp_filter for the nics. You might want to test
>>>>> this,
>>>>> by logging in and disabling rp_filter on the nics and see if things
>>>>> start
>>>>> working as expected.
>>>>> 
>>>>> 
>>>>>> 
>>>>>> Obviously, my network configuration is incorrect, but I have the
>>>>>> reached
>>>>>> the limits of my CloudStack and Xen knowledge to identify the
>>>>>> problem(s).
>>>>>> 
>>>>>> Given this information, what is the best way to give the SSVM access
>>>>>> to
>>>>>> the Internet and/or a VirtualBox host-only network?
>>>>>> 
>>>>>> Thank you for your help,
>>>>>> -John  
>>>>>> 
>>>>>> On Oct 3, 2012, at 10:39 PM, "Kelceydamage@bbits" <kel...@bbits.ca>
>>>>>> wrote:
>>>>>> 
>>>>>>> The the secondary storage VM can be NATed to from any network
>>>>>>> router,
>>>>>>> however the console proxy does not work over NAT.
>>>>>>> 
>>>>>>> Sent from my iPhone
>>>>>>> 
>>>>>>> On Oct 3, 2012, at 7:32 PM, Edison Su <edison...@citrix.com> wrote:
>>>>>>> 
>>>>>>>> System vm will have 4 nics, eth2 is on the public network, eth1 is
>>>>>>>> the
>>>>>>>> private(mgt) network.
>>>>>>>> The IP address of eth2 is got from pod configuration: in one of IP
>>>>>>>> address range ["startip", "endip"] in createPod API.
>>>>>>>> The IP address of eth1 is got from guest network, if it's basic
>>>>>>>> network mode, this IP range is configured by createVlanIpRanges API
>>>>>>>> SSVM will connect to mgt server through eth1(mgt server's ip
>>>>>>>> address
>>>>>>>> is configured to route through eth1), and download template from
>>>>>>>> eth2.
>>>>>>>> What's your specific issue about network configuration?
>>>>>>>> 
>>>>>>>>> -----Original Message-----
>>>>>>>>> From: John Burwell [mailto:jburw...@basho.com]
>>>>>>>>> Sent: Wednesday, October 03, 2012 7:11 PM
>>>>>>>>> To: cloudstack-dev@incubator.apache.org
>>>>>>>>> Subject: SSVM Network Configuration
>>>>>>>>> 
>>>>>>>>> All,
>>>>>>>>> 
>>>>>>>>> How do you configure networking to permit the SSVM to connect to
>>>>>>>>> the
>>>>>>>>> public Internet or another internal network?  I have been trying
>>>>>>>>> to
>>>>>>>>> understand the network configuration from the documentation, but
>>>>>>>>> am
>>>>>>>>> missing something in my configuration attempt.
>>>>>>>>> 
>>>>>>>>> Thank you for your assistance,
>>>>>>>>> -John
>>>>>> 
>>>>>> 
>>>>> 
>>>>> 
>>>>> -- 
>>>>> Æ
>>>>> 
>>>>> 
>>>>> 
>>>> 
>>> 
>> 
> 

Reply via email to