[Openstack-operators] [OCTAVIA][KOLLA] - Amphora to control plan communication question.

Flint WALRUS gael.therond at gmail.com
Wed Aug 1 04:49:07 UTC 2018


Hi Michael,

Oh ok! That config-drive trick was the missing part! Thanks a lot! Is there
a release target for the API vs config-drive thing? I’ll have a look at an
instance as soon as I’ll be able to log into one of my amphora.

By the way, three sub-questions remains:

1°/ - What is the best place to push some documentation improvement ?
2°/ - Is the amphora-agent an auto-generated file at image build time or do
I need to create one and give it to the diskimage-builder process?
3°/ - The amphora agent source-code is available at
https://github.com/openstack/octavia/tree/master/octavia/amphorae/backends/agent
isn’t?

Sorry for the questions volume, but I prefer to really understand the
underlying mechanisms before we goes live with the solution.

G.

Le mer. 1 août 2018 à 02:36, Michael Johnson <johnsomor at gmail.com> a écrit :

> Hi Flint,
>
> Happy to help.
>
> Right now the list of controller endpoints is pushed at boot time and
> loaded into the amphora via config drive/nova.
> In the future we plan to be able to update this list via the amphora
> API, but it has not been developed yet.
>
> I am pretty sure centos is getting the config file as our gate job
> that runs with centos 7 amphora has been passing. It should be in the
> same /etc/octavia/amphora-agent.conf location as the ubuntu based
> amphora.
>
> Michael
>
>
>
> On Tue, Jul 31, 2018 at 10:05 AM Flint WALRUS <gael.therond at gmail.com>
> wrote:
> >
> > Hi Michael, thanks a lot for that explanation, it’s actually how I
> envisioned the flow.
> >
> > I’ll have to produce a diagram for my peers understanding, I maybe can
> share it with you.
> >
> > There is still one point that seems to be a little bit odd to me.
> >
> > How the amphora agent know where to find out the healthManagers and
> worker services? Is that because the worker is sending the agent some
> catalog informations or because we set that at diskimage-create time?
> >
> > If so, I think the Centos based amphora is missing the agent.conf
> because currently my vms doesn’t have any.
> >
> > Once again thanks for your help!
> > Le mar. 31 juil. 2018 à 18:15, Michael Johnson <johnsomor at gmail.com> a
> écrit :
> >>
> >> Hi Flint,
> >>
> >> We don't have a logical network diagram at this time (it's still on
> >> the to-do list), but I can talk you through it.
> >>
> >> The Octavia worker, health manager, and housekeeping need to be able
> >> to reach the amphora (service VM at this point) over the lb-mgmt-net
> >> on TCP 9443. It knows the amphora IP addresses on the lb-mgmt-net via
> >> the database and the information we save from the compute driver (I.e.
> >> what IP was assigned to the instance).
> >>
> >> The Octavia API process does not need to be connected to the
> >> lb-mgmt-net at this time. It only connects the the messaging bus and
> >> the Octavia database. Provider drivers may have other connectivity
> >> requirements for the Octavia API.
> >>
> >> The amphorae also send UDP packets back to the health manager on port
> >> 5555. This is the heartbeat packet from the amphora. It contains the
> >> health and statistics from that amphora. It know it's list of health
> >> manager endpoints from the configuration file
> >> "controller_ip_port_list"
> >> (
> https://docs.openstack.org/octavia/latest/configuration/configref.html#health_manager.controller_ip_port_list
> ).
> >> Each amphora will rotate through that list of endpoints to reduce the
> >> chance of a network split impacting the heartbeat messages.
> >>
> >> This is the only traffic that passed over this network. All of it is
> >> IP based and can be routed (it does not require L2 connectivity).
> >>
> >> Michael
> >>
> >> On Tue, Jul 31, 2018 at 2:00 AM Flint WALRUS <gael.therond at gmail.com>
> wrote:
> >> >
> >> > Hi Folks,
> >> >
> >> > I'm currently deploying the Octavia component into our testing
> environment which is based on KOLLA.
> >> >
> >> > So far I'm quite enjoying it as it is pretty much straight forward
> (Except for some documentation pitfalls), but I'm now facing a weird and
> hard to debug situation.
> >> >
> >> > I actually have a hard time to understand how Amphora are
> communicating back and forth with the Control Plan components.
> >> >
> >> > From my understanding, as soon as I create a new LB, the Control Plan
> is spawning an instance using the configured Octavia Flavor and Image type,
> attach it to the LB-MGMT-NET and to the user provided subnet.
> >> >
> >> > What I think I'm misunderstanding is the discussion that follows
> between the amphora and the different components such as the
> HealthManager/HouseKeeper, the API and the Worker.
> >> >
> >> > How is the amphora agent able to found my control plan? Is the
> HealthManager or the Octavia Worker initiating the communication to the
> Amphora on port 9443 and so give the agent the API/Control plan internalURL?
> >> >
> >> > If anyone have a diagram of the workflow I would be more than happy ^^
> >> >
> >> > Thanks a lot in advance to anyone willing to help :D
> >> >
> >> > _______________________________________________
> >> > OpenStack-operators mailing list
> >> > OpenStack-operators at lists.openstack.org
> >> >
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20180801/99a3c5d0/attachment-0001.html>


More information about the OpenStack-operators mailing list