We have been doing bulk tests of the OpenNebula 4.8 econe-server.
With just a straight econe-run-instances we can get up to 1000 VM's (the limit 
of our current subnet)
started fairly quickly (about 30 minutes)

But in practice we are using a more complicated sequence of EC2 calls via 
HTCondor.
In particular it is doing a CreateKeyPair call before it launches each VM and 
then
calling the RunInstances method with the --keypair option, a unique keypair for 
each VM.
After the VM exits, it called a DeleteKeyPair call.

IT appears there is a hard limit of the number of key pairs that can be stored 
in
any one user's template and that hard limit is 301.  Any further CreateKeyPair 
calls
return with "connection reset by peer" causing HTCondor to mark the VM as held.
Fortunately it is possible to override this and tell HTCondor to continue, but 
it's a pain.
We do have ways to log into the vm's without the ssh key pair so we wouldn't 
even really need to register
them at all.

Is my analysis correct?  Is there a hard limit of the number of keys that can 
be stored in the user template?
If so, how best to get around this limit?

Steve Timm


_______________________________________________
Users mailing list
[email protected]
http://lists.opennebula.org/listinfo.cgi/users-opennebula.org

Reply via email to