Merge "Add bgp speaker and route advertisement doc"

This commit is contained in:
Jenkins 2016-07-19 15:33:18 +00:00 committed by Gerrit Code Review
commit 58e3e42894
2 changed files with 229 additions and 2 deletions

View File

@ -24,5 +24,144 @@
BGP Speaker
===========
BGP Speaker acts as a route server using BGP routing protocol. It advertises
routes to the BGP peers which are added to the BGP Speaker. Now there is a
framework that allows different `BGP drivers <../design/drivers.rst>`_
to be plugged into a `dynamic routing agent <./dynamic-routing-agent.rst>`_.
TODO: Coming Soon
Currently, BGP Speaker only advertises routes for a network to which it is associated.
A BGP Speaker requires association with a "gateway" network to determine eligible routes.
In Neutron, a "gateway" network connects Neutron routers to the upstream routers. An
external network is best for being used as a gateway network. The association builds a
list of all virtual routers with gateways on provider and self-service networks within
the same address scope. Hence, the BGP speaker advertises self-service network prefixes
with the corresponding router as the next-hop IP address.
For details refer to `Route advertisement <./route-advertisement.rst>`_.
Address Scopes
--------------
`Address scopes <https://git.openstack.org/cgit/openstack/neutron/tree/doc/source/devref/address_scopes.rst>`_
provide flexible control as well as decoupling of address overlap from tenancy,
so this kind control can provide a routable domain, the domain has itself route
and no overlap address, it means an address scope define "a L3 routing domain".
BGP Speaker will associate the external networks and advertise the tenant's
networks routes. Those networks should reside in the same address scope.
Neutron can route the tenant network directly without NAT. Then Neutron can
host globally routable IPv4 and IPv6 tenant networks. For determining which
tenant networks prefixes should be advertised, Neutron will identify all routers
with gateway ports on the network which had been bounded with BGP Speaker,
check the address scope of the subnets on all connected networks, then begin
advertising nexthops for all tenant networks to routers on the bound network.
BGP Peer
--------
BGP peer defined in Neutron represents real BGP infrastructure such as
routers, route reflectors and route servers. When a BGP peer is defined and
associated with a BGP Speaker, Neutron will attempt to open a BGP peering
session with the mentioned remote peer. It is this session, using which Neutron
announces it's routes.
How to configure a remote peer
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
A remote peer can be real or virtual e.g. vRouters or real routers.
The remote peer should be configured to handle peering with Neutron in passive
mode. The peer needs to waits for the Neutron dynamic routing agent to
initiate the peering session. Also, the remote peer can be configured in active
mode, but it still can speak BGP until the complete initialization of BGP Speaker
running on Neutron dynamic routing agent.
Configuring BGP Speaker:
One needs to ensure below points for setting a BGP connection.
* Host running Neutron dynamic agent MUST connect to the external router.
* BGP configuration on the router should be proper.
``bgp router-id XX.XX.XX.XX``
This must be an IP address, the unique identifier of BGP routers actually
and can be virtual. If one doesn't configure the router-id, it will be selected
automatically as the highest IP address configured for the local interfaces.
Just a suggestion, please make sure that it is the same as the ``peer_ip``
which you configure in Neutron for distinguishing easily.
``local_as``
Autonomous System number can be same or different from the AS_id of external
BGP router. AS_id will be same for iBGP and different for eBGP sessions.
Setting BGP peer:
::
neighbor A.B.C.D remote-as AS_ID
A.B.C.D is the host IP which run Neutron dynamic routing agent.
A Sample Quagga router configuration file forming BGP peering with Neutron:
::
!
password zebra
log file /var/log/quagga/bgpd.log
!
debug bgp events
debug bgp keepalives
debug bgp updates
debug bgp fsm
debug bgp filters
!
bgp multiple-instance
!
router bgp <BgpPeer remote_as> view test-as
bgp router-id <quagga router IP address>
neighbor <dr_agent IP address> remote-as <BgpSpeaker local_as>
neighbor <dr_agent IP address> passive
!
line vty
!
BGP Speaker Architecture
------------------------
Dynamic routing project saves BGP Speaker configuration as per the defined
`_data model<https://git.openstack.org/cgit/openstack/neutron-dynamic-routing/tree/neutron_dynamic_routing/db/bgp_db.py#n85>`_.
and pass on the configuration request to the dynamic routing agent for further processing.
The implementation of a BGP Speaker is driver specific. During the driver interface
initialization process, needed configurations are read from the configuration file
and BGP Speaker object instance is created. For details refer to
`BGP drivers <../design/drivers.rst>`_.
BGP Speaker Life Cycle
~~~~~~~~~~~~~~~~~~~~~~
Now we support RyuBgpDriver, BGP Speaker will be processed by Dragent. When
associating a BGP Speaker with an active Dragent, the plugin will send an RPC
message to the agent for calling driver in order to create a BGP Speaker instance.
In RyuBgpDriver, the created instance ``BGP Speaker`` will setup by router-id
and ASN, then Ryu will setup new context with speaker configuration and listeners
which monitor whether the related peers are alive.
Then the following operation could be done.
* Add peers to BGP Speaker
When BGP Speaker is not associated with an active Dragent, there is no real speaker
instance, so it will be still the db operation until the speaker is associated with
dragent, and all the peers connection before will be setup by ``BGP Speaker``
creation. If add peers into speaker which is running, Dragent will call driver
to add peer dynamically. For RyuBgpDriver, it will register a new neighbor
based on your peer configuration and try to establish a session with the peer.
* Delete peers from BGP Speaker
The same logic with below, but it is reverse.
If you don't want use the specific BGP Speaker anymore, you can use CLI:
``neutron bgp-speaker-delete <SPEAKER NAME/ID>``
BGP Plugin will find all the associated Dragent and send RPC ``bgp_speaker_remove_end``
to make the Dragents to clean the ``BGP Speaker`` instances. This is the same
with CLI:
``neutron bgp-dragent-speaker-remove <DRAGENT ID> <SPEAKER NAME/ID>``
BGP Plugin just send rpc ``bgp_speaker_remove_end`` to the specific Dragent.
Advertisement
~~~~~~~~~~~~~
For details refer to `Route Advertisement <./route-advertisement.rst>`_.
How to work
-----------
For details refer to `Testing <../others/testing.rst>`_.

View File

@ -24,5 +24,93 @@
Route Advertisement
===================
BGP
---
TODO: Coming Soon
This page discusses the behavior of BGP dynamic routing about how to advertise
routes and show the routes details in the project.
BGP dynamic routing could advertise 3 classes of routes:
* Host routes for floating IP addresses hosted on non-DVR routers, as floatingip
address set on the router namespace, it knows how to route the message to the
correct way, so the next-hop should be the IP address of router gateway port.
* Host routes for floating IP addresses hosted on DVR routers. With DVR-enabled
routers, the floating IP can be reached directly on the compute node hosting
a given instance. As such, host routes for the floating IP address should
advertise the FIP agent gateway on the compute node as the next-hop instead of
the centralized router. This will keep inbound floating IP traffic from
encountering the bottleneck of the centralized router.
* Prefix routes for directly routable tenant networks with address scopes, the
nexthop is the centralized router, the same for DVR and CVR. BGP dynamic
routing could advertise tenant network prefixes to physical network
devices(routers which support BGP protocol), called this
``Prefixes advertisement``.
When distributed virtual routing (DVR) is enabled on a router, next-hops for
floating IP's and fixed IP's are not advertised as being at the centralized
router. Host routes with the next-hop set to the appropriate compute node
are advertised.
Logical Model
~~~~~~~~~~~~~
::
+--------+ 1 N +---------------------+
| Router |---------| BgpAdvertisedRoute |
+--------+ +---------------------+
| N
|
| 1
+---------+ N N +------------+ N N +---------+
| BgpPeer |-----------| BgpSpeaker |-----------| Network |
+---------+ +------------+ +---------+
| N
|
| 1
+--------------+
| AddressScope |
+--------------+
.. note::
A BGP Speaker only supports one address family to speak BGP. A dual-stack IPv4
and IPv6 network needs two BGP Speakers to advertise the routes with BGP, one
for IPv4 and the other for IPv6. So A network can have N number of BGP
Speakers bound to it.
BgpAdvertisedRoute represents derived data. As the number of
BgpAdvertisedRoutes can be quite large, storing in a database table is not
feasible. BgpAdvertisedRoute information can be derived by joining data
already available in the Neutron database. And now BGP dynamic routing project
process the Bgpadvertiseroutes which should be advertised to external Router is
basing on the exist Neutron DB tables.
Neutron looks on each of the gateway network for any routers with a gateway port
on that network. For each router identified, Neutron locates each floating IP
and tenant network accessible through the router gateway port. Neutron then
advertises each floating IP and tenant network with the IP address of the router
gateway port as the next hop.
When BGP Plugin is started, it will register callbacks. All callbacks are used for
processing Floating IP, Router Interface and Router Gateway creation or update, this
functions listen the events of these resources for calling Dragent to change the
advertisement routes.
Now we just focus on the resources which may cause route change, the following
callbacks does this work.
* floatingip_update_callback
This function listens to the Floating IP's AFTER_UPDATE event, it judges whether
the associated router is changed, and changes the advertisement routes and nexthop
based on that.
* router_interface_callback
This function listens to the tenants' network routes change, it listens to AFTER_CREATE
and AFTER_DELETE events of Router Interface resource. It calls Dragent to advertise
or stop the prefix routes after a interface attach into a router.
* router_gateway_callback
This function listens to the router gateway port creation or deletion. It also focuses
on tenants' network routes change.
You could get the advertisement routes of specific BGP Speaker like:
``neutron bgp-speaker-advertiseroute-list <created-bgp-speaker>``
It does a complicated db query to generate the list of advertised routes.
For more details refer to `route advertisement db lookup <https://git.openstack.org/cgit/openstack/neutron-dynamic-routing/tree/neutron_dynamic_routing/db/bgp_db.py#n462>`_