openstack/fuel-docs
Code Issues Proposed changes

Add HA deployment for Networking section

Browse Source 
Add basic content for Networking HA section.
Provide decription of haproxy namespace and VIP
HA failover details.
TODO - provide details for Neutron agents HA, failover,
cleanup and reschedule procedures.

Related bug: #1415398

Change-Id: I4485da25bdbab98c9496e7c13598c2dcd1ea2eb6
Signed-off-by: Bogdan Dobrelya <bdobrelia@mirantis.com>
This commit is contained in:
Bogdan Dobrelya
2015-02-09 18:08:35 +01:00
parent 26551c1f8a
commit 27625571f3
3 changed files with 40 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
pages/reference-architecture/6000-network-architecture.rst
View File

@@ -12,6 +12,7 @@ Network Architecture
.. include:: /pages/reference-architecture/network-concepts/6010-logical-networks.rst
.. include:: /pages/reference-architecture/network-concepts/6011-ha-networking.rst
.. include:: /pages/reference-architecture/network-concepts/6015-mcn.rst
.. include:: /pages/reference-architecture/network-concepts/6020-public-floating-ip.rst
.. include:: /pages/reference-architecture/network-concepts/6030-router.rst

38
pages/reference-architecture/network-concepts/6011-ha-networking.rst Normal file
View File

@@ -0,0 +1,38 @@
.. index:: Reference Architectures: Networking HA Details
.. _Close_look_networking_HA:
HA deployment for Networking
----------------------------
Fuel leverages
`Pacemaker resource agents <http://www.linux-ha.org/wiki/Resource_agents>`_
in order to deploy highly avaiable networking for OpenStack environments.
Virtual IP addresses deployment details
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Starting from the Fuel 5.0 release, HAProxy service and network interfaces
running virtual IP addresses reside in separate `haproxy`
network namespace. Using a separate namespace forces Linux kernel to treat
connections from OpenStack services to HAProxy as remote ones, this ensures
reliable failover of established connections when the management IP address
migrates to another node
(see `LP1285449 <https://bugs.launchpad.net/fuel/+bug/1285449>`_).
In order to achieve this, resource agent scripts for `ocf:heartbeat:haproxy`
and `ocf:heartbeat:IPaddr2` were hardened with network namespaces support.
Successfull failover of public VIP address requires controller nodes
to perform active checking of the public gateway. Fuel configures
the Pacemaker resource `clone_ping_vip__public` that makes public VIP to
migrate in case the controller can't ping its public gateway.
TCP keepalive configuration details
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Failover sometimes ends up with dead connections. The detection
of such connections requires additional assistance from the Linux kernel.
To speed up the detection process from the default of two hours to a more
acceptable 3 minutes, Fuel adjusts kernel parameters for
`net.ipv4.tcp_keepalive_time`, `net.ipv4.tcp_keepalive_intvl`,
`net.ipv4.tcp_keepalive_probes` and `net.ipv4.tcp_retries2`.

1
pages/terminology/n/networking.rst
View File

@@ -14,6 +14,7 @@ communicate with each other over using one of the network topologies:
The following documents provide information:
* For networking HA reference architecture, see :ref:`Close_look_networking_HA`.
* For general information to help you select the best network topology
for your environment, see :ref:`net-topology-plan`.
* For a list of the logical networks used in OpenStack