[ovs-discuss] Question on using TCP transport keep alive for ovsdb connection when possible.

[Hexin Wang]

Got it. Networking-Ovn can be fixed to follow the same logic in ovn-controller(8) to configure “inactivity_probe” interval. 

Meanwhile, I still think having Transport level keepavlie is another good option for applications that don’t want to use openflow keep alive. 


Thanks.

Hexin 



On 3/6/17, 2:11 PM, "Ben Pfaff" <blp at ovn.org> wrote:

>I don't know about the plugin, someone else will have to help with that.
>
>On Mon, Mar 06, 2017 at 09:59:06PM +0000, Hexin Wang wrote:
>> How about from neutron networking-ovn plugin connecting to both northbound and southbound db?
>> 
>> Thanks.
>> 
>> Hexin
>> 
>> 
>> 
>> On 3/6/17, 12:56 PM, "Ben Pfaff" <blp at ovn.org> wrote:
>> 
>> >It already is, see ovn-controller(8).  Note that ovn-controller only
>> >connects to the southbound database.
>> >
>> >On Mon, Mar 06, 2017 at 07:16:51PM +0000, Hexin Wang wrote:
>> >> Fair comment. In that case, is there any plan in making "inactivity probe" interval configurable in ovn north and south dub connection?
>> >> 
>> >> Thanks,
>> >> 
>> >> Hexin
>> >> 
>> >> > On Mar 6, 2017, at 10:59 AM, Ben Pfaff <blp at ovn.org> wrote:
>> >> > 
>> >> > If the application is extremely slow, then the connection is effectively
>> >> > dead, and we might as well drop it anyway.
>> >> > 
>> >> >> On Mon, Mar 06, 2017 at 06:53:51PM +0000, Hexin Wang wrote:
>> >> >> Hi Ben,
>> >> >> 
>> >> >> TCP keep alive gives a good alternative to use cases where there is no firewall or proxy concern. Transport level keep alive is good enough for ovsdb to determine if connection is alive or not, without worrying if application is slow (or extremely slow) in open flow echo reply. 
>> >> >> 
>> >> >> Can we working on pushing this into ovs release?
>> >> >> 
>> >> >> Thanks.
>> >> >> 
>> >> >> Hexin
>> >> >> 
>> >> >> 
>> >> >> 
>> >> >>> On 3/2/17, 3:57 PM, "Ben Pfaff" <blp at ovn.org> wrote:
>> >> >>> 
>> >> >>> We used application-level echo request and replies instead, because they
>> >> >>> are reliable even if a TCP connection passes through a firewall or proxy
>> >> >>> that does not properly pass through TCP keepalives.
>> >> >>> 
>> >> >>>> On Thu, Mar 02, 2017 at 11:48:04PM +0000, Hexin Wang wrote:
>> >> >>>> Hi Ben,
>> >> >>>> 
>> >> >>>> What is the reason that it is not getting to the main release? This seems the right keep alive mechanism for neutron to talk to ovn database, if they are not running on the same host/container and would have to use tcp as the transport.
>> >> >>>> 
>> >> >>>> Thanks.
>> >> >>>> 
>> >> >>>> Hexin
>> >> >>>> 
>> >> >>>> 
>> >> >>>> 
>> >> >>>> 
>> >> >>>>> On 3/2/17, 3:45 PM, "Ben Pfaff" <blp at ovn.org> wrote:
>> >> >>>>> 
>> >> >>>>>> On Thu, Mar 02, 2017 at 11:43:01PM +0000, Hexin Wang wrote:
>> >> >>>>>> I have a question on the following patch that use TCP keep alive for ovsdb connection.
>> >> >>>>>> 
>> >> >>>>>> https://mail.openvswitch.org/pipermail/ovs-dev/2011-April/251891.html
>> >> >>>>>> 
>> >> >>>>>> Did this patch go into ovs main release?
>> >> >>>>> 
>> >> >>>>> No.