[libvirt] [RFC] driver's persistent connection closed event

Nikolay Shirokovskiy nshirokovskiy at parallels.com
Sat Jun 6 10:57:57 UTC 2015


Hello.                                                                                                              
                                                                                                                    
I have a question concerning hypervisor driver "persistent connection                                               
closed" event mechanics.                                                                                            
                                                                                                                    
Currently the only driver that have such a connection (or at least                                                  
notify of such event) is the remote driver. Accordingly we don't                                                    
have interface for this event at driver level.                                                                      
Parallels driver actually have a persistent connection. Imagine                                                     
we have a libvirt connection to parallels driver through remote driver,                                             
if parallels internal connection is closed we can catch it in daemon                                                
but the only option to notify remote driver of this event would                                                     
be closing of remote driver to daemon connection which will                                                         
in turn trigger connection closed event deliver to client. Thus                                                     
event will be delivered but at cost of forcedly closed connection                                                   
to daemon.                                                                                                          
I suspect the described solution have architectural flaws so                                                        
i suggest move interface of "persistent connection closed" event                                                    
to driver level which will give us a chance to deliver this                                                         
event without side effects. We don't need to implement this                                                         
interface for most drivers and clients could handle this                                                            
well as in this case "unimplemented" would be a synonym                                                             
of "have not persistent connection". Alternatively 
we could silently do nothing if driver does not support
this interface so clients would not break.




More information about the libvir-list mailing list