Opened 7 years ago

Last modified 7 years ago

#9 reopened

Sliver with 6 nodes is created, but only 2 show up in sliver listresources (manifest)

Reported by: lnevers@bbn.com Owned by: somebody
Priority: major Milestone:
Component: AM Version: SPIRAL4
Keywords: Cc:
Dependencies:

Description

Using an rspec with the following topology in the BBN rack:

  • 3 VM nodes on lan1
  • 3 VM nodes on lan2
  • 1 of the nodes from lan1 is connected via lan3 to 1 of the nodes on lan2.

Was able to create the sliver successfully, but the sliver manifest only shows two nodes.

Attached are the rspec (exo-3vm-3lan-c.rspec) and the sliver manifest (exo3-rspec-bbn-hn-exogeni-net-11443-orca.xml) that shows only 2 nodes.

Attachments (6)

exo-3vm-3lan-c.rspec (4.2 KB) - added by lnevers@bbn.com 7 years ago.
exo3-rspec-bbn-hn-exogeni-net-11443-orca.xml (3.5 KB) - added by lnevers@bbn.com 7 years ago.
exo-2vm-1lan.rspec (2.0 KB) - added by lnevers@bbn.com 7 years ago.
2 VM on 1 LAN
exo-3vm-1lan.rspec (2.5 KB) - added by lnevers@bbn.com 7 years ago.
3 VM on 1 LAN
exo-3vm-3lan-2sites.rspec (4.7 KB) - added by lnevers@bbn.com 7 years ago.
lnexx0-rspec-geni-renci-org-11443-orca.xml (5.1 KB) - added by lnevers@bbn.com 7 years ago.

Download all attachments as: .zip

Change History (17)

Changed 7 years ago by lnevers@bbn.com

Attachment: exo-3vm-3lan-c.rspec added

Changed 7 years ago by lnevers@bbn.com

comment:1 Changed 7 years ago by ibaldin@renci.org

This is a converter issue that may a little time to resolve. If you use two instead of 3 vms on each vlan it should work.

Changed 7 years ago by lnevers@bbn.com

Attachment: exo-2vm-1lan.rspec added

2 VM on 1 LAN

Changed 7 years ago by lnevers@bbn.com

Attachment: exo-3vm-1lan.rspec added

3 VM on 1 LAN

comment:2 Changed 7 years ago by lnevers@bbn.com

Modified the original topology:

  • 3 VM nodes on lan1
  • 3 VM nodes on lan2
  • 1 of the nodes from lan1 is connected via lan3 to 1 of the nodes on lan2.

to the suggested 2 Nodes per VLAN, which resulted in the following request rspec:

  • 2 VM nodes on lan1
  • 2 VM nodes on lan2
  • 1 of the nodes from lan1 is connected via lan3 to 1 of the nodes on lan2.

The suggested change worked. Also tried two additional experiments at the ExoSM which also show that the requested nodes are not assigned:

  1. Requested 3 VMs on 1 LAN, got only 1 node ( same as the original problem)
  2. Requested 2 VMs on 1 LAN, got 2 nodes (new symptom?)

Attaching rspec for the experiments 1 and 2.

comment:3 Changed 7 years ago by ibaldin@renci.org

Requesting 2VMs on 1 LAN and getting 2 nodes seems like a correct behavior, no?

comment:4 Changed 7 years ago by lnevers@bbn.com

Yes of course, this is the right behavior. Sorry have been looking at too manifest RSpecs :-) Please ignore the 2 VM experiment comment.

Problems only occur with experiments that use more than two VM on 1 vlan.

comment:5 Changed 7 years ago by ibaldin@renci.org

Two VMs on one VLAN in orca are done using something called Node Groups. No equivalent in RSpec. We are working on a workaround...

comment:6 Changed 7 years ago by lnevers@bbn.com

Resolution: fixed
Status: newclosed

This issue is resolved in the current version.

comment:7 Changed 7 years ago by lnevers@bbn.com

Resolution: fixed
Status: closedreopened

Maybe this issues is not fully resolved. The original ticket, had both node groups in the BBN rack and that is resolved.

I just re-ran the test and modified the rspec to have one node group in the BBN rack and one node group in the RENCI rack, but according to the sliver manifest only 1 VM was assigned in each rack.

Attaching the Request RSpec and the sliver manifest.

Changed 7 years ago by lnevers@bbn.com

Attachment: exo-3vm-3lan-2sites.rspec added

Changed 7 years ago by lnevers@bbn.com

comment:8 Changed 7 years ago by lnevers@bbn.com

Re-ran the last problem scenario which includes:

  • 3 VMs on 1 LAN in BBN rack
  • 3 VMs on 1 LAN in RENCI rack
  • 1 LAN connecting one of the BBN nodes to one of the RENCI nodes.

The rspec used is the file exo-3vm-3lan-2sites.rspec which is already attached, with one modification for the RSpec GENI V3 header.

Created a sliver named 3vm3lan2sites which allocated 5 hosts at BBN and 1 at RENCI, from manifest:

$ egrep hostname 3vm3lan2sites-rspec-geni-renci-org-11443-orca.xml 
                  <login authentication="ssh-keys" hostname="152.54.14.12" port="22" username="root"/>      
                  <login authentication="ssh-keys" hostname="192.1.242.11" port="22" username="root"/>      
                  <login authentication="ssh-keys" hostname="192.1.242.12" port="22" username="root"/>      
                  <login authentication="ssh-keys" hostname="192.1.242.13" port="22" username="root"/>      
                  <login authentication="ssh-keys" hostname="192.1.242.10" port="22" username="root"/>      
                  <login authentication="ssh-keys" hostname="192.1.242.14" port="22" username="root"/>      

Attaching the sliver manifest.

comment:9 Changed 7 years ago by lnevers@bbn.com

On 7/5/12 4:02 PM, Ilia Baldine wrote:

It works for nodegroups in flukes, but then all nodes in a group have to be identical (same size, same image). I can modify the converter to support this case...

I tried because I was asked to check the state of this ticket.

I am adding the response from Ilia because this answer has been shared in emails, but not in this ticket.

comment:10 Changed 7 years ago by lnevers@bbn.com

Just checking ticket status..... Node groups are still not supported via omni:

  Result Summary: Failed CreateSliver for slice lngroup at https://geni.renci.org:11443/orca/xmlrpc.  Error from Aggregate: code 2: ERROR: Exception encountered: java.lang.NullPointerException. 

comment:11 Changed 7 years ago by lnevers@bbn.com

Checking status for unresolved tickets, while things are slow :-)

Topology: Requested sliver from ExoSM that includes:

  • 3 VMs (VM, VM-0, VM-1) on 1 LAN in BBN rack
  • 3 VMs (VM-2, VM-3, VM-4) on 1 LAN in RENCI rack
  • 1 LAN connecting one of the BBN nodes (VM-0) to one of the RENCI nodes (VM-2).

Results:

Behavior has changed since last December when this scenario was last tried.

In December, the create sliver would fail due to the lack of support for node groups. This is no longer the case. Now, the create sliver is successful, the sliver becomes ready, and no error is reported, but there is only one VM at the BBN rack and one VM at RENCI rack.

Note: See TracTickets for help on using tickets.