Age | Commit message (Collapse) | Author | Files | Lines |
|
|
|
MidoNet API needs to be loadbalanced if the midonet environment is
activated.
Change-Id: I6f1ac659297b8cf6671e11ad23284f8f543568b0
|
|
|
|
Change-Id: I3fdb705bbac26b4bc43a18131407a0a86d36a8a5
|
|
|
|
Add Aodh (Ceilometer Alarming) support in TripleO Loadbalancer config.
Change-Id: I891985da9248a88c6ce2df1dd186881f582605ee
|
|
Nova vnc and swift proxy were listening on the same port if SSL is
enabled in the load balancer
Change-Id: Ibf4aa118d6c8e94f8f2a68bf270d5445ebda7593
|
|
|
|
keystone and heat_cfn were listening on the same port if SSL is enabled
inm the load balancer.
Change-Id: I099119198ebf3322a783581f0c6758417e705a2e
|
|
When using websockets in HAProxy, like nova_novncproxy does, we
need to set "timeout tunnel" to avoid disconnections after a short
period without traffic.
Change-Id: I1b66cd9a1d20cbbe35a2ada5782a76a01b14bcd1
Closes-BZ: 1267043
|
|
- s/manila/$manila
Change-Id: I7aaa8f83fe758484ab39af28c914fa3d78464633
Signed-off-by: Gael Chamoulaud <gchamoul@redhat.com>
|
|
|
|
The haproxy configuration for horizon does not have 'mode http' set.
This proxy needs to be in http mode since it is using a cookie for
persistence. The default section has 'mode tcp', which is fine, but
horizon proxy needs to override this setting to get http mode. Without
this, you will likely see an error like this:
[WARNING] 238/115010 (13878) : config : cookie will be ignored for proxy
'horizon' (needs 'mode http').'
Closes BZ-1257687
Change-Id: I397986ea022f47a33a5210696752509f4a2731a5
|
|
Currently the address of the syslog server for HAProxy is hardcoded to
/dev/log without a way to customize this setting.
This commit aims to give a user more flexibility about which syslog
server address to use.
Change-Id: If7f7c8154e544e5d8a49f79f642e1ad01644a66d
|
|
When establishing a connection from the client (Web Browser) to the
novncproxy (loadbalanced by HAproxy), we need to make sure the client
will stick on the same server the time he's connected, because HAproxy
load-balance to another novncproxy node, the client will loose the
connection and timeout like 'Connection Reset By Peer error'.
This patch aims to configure novnc HAproxy configuration to balance
using 'source' mode, so it will make sure the server remains the same
while the connection is established.
Change-Id: Ibbb7162b763f1fd2854a10a92a681910e0683c0a
Closes-BZ: 1257324
|
|
To make sure we don't use the ssl-hello-chk option set by the
puppet-haproxy module we used to redefine the listener options
for all listeners.
With this change a default for the options hash is provided to
the puppet class instead.
This change also configures use of tcpka only where wanted, as
documented by [1], removing it from the haproxy defaults section,
given it wasn't used anyway by the other listeners which were
indeed overriding options.
1. https://github.com/beekhof/osp-ha-deploy/blob/master/pcmk/lb.scenario
Change-Id: Ic8deb77533f561cea7ce7db1d20f6be5e2dc0d33
|
|
Adds bindings to the Manila service for HAProxy.
Change-Id: I175d5b7e35a781d04452fc6aee610e8dca005419
|
|
EC2 API returns 400 for unauthenticated requests, making HAProxy believe
that the service is down. We'll use TCP check instead of HTTP check for
EC2 API.
Change-Id: Ide7f9390603c9893b95cacd51d468461255dcf07
|
|
This updates some of the listener options set by loadbalancer.pp.
Iroinc needs to pass in the option to do a httpchk, otherwise
puppet-haproxy defaults it to doing a ssl-hello-chk, which won't work
against the non-ssl loadbalancer server.
Ceilometer and glance_registry both don't support a httpchk against the
root (/) of their webservers (they return a straight 401) so disable
those checks completely.
Change-Id: Ibfc81175842a748eb077b132b0818c4ea17bbcf6
|
|
The default per frontend maxconn is set to 2000, which can easily
be reached with modern hardware with multiple logic cores; this
change adds a parameter to configure the default maxconn value,
default it to 4096 and also increases the global maxconn to 20480
to preserve the 1:5 ratio.
Change-Id: I3fffc51ecc704ceccb86ca008ecba02578c29eb5
|
|
Backend options for Ceilometer and Ironic are aligned with what we
use for the other OpenStack services.
Listener options for Horizon is updated so that we do cookie
tracking as suggested by refarch doc.
Change-Id: I4640d974a3ab8188919eaae79dde71463234b5ff
|
|
Change-Id: Ic0ae6b743a732ccd2cf7e395b5ab172bf3daaf7d
|
|
When doing a heat stack-show, Heat initially returns a 302 redirect.
With the existing loadbalancer config for SSL, this results in a
redirect to an http:// address pointing at the SSL port, which
naturally doesn't work.
The fix for this is to use the rsprep haproxy option to rewrite the
Location header in responses from the Heat api server. This allows
us to properly handle redirect traffic as https.
Also note that http header rewriting requires "mode http", so that
is added here as well.
Change-Id: I7e5c5b1877e9aa46c4b88dfba45c1fddf61727fc
|
|
Just like any other OpenStack API endpoint.
Change-Id: Iaa45d7bef94c3c42df0988a58f146bb8a530f74e
|
|
|
|
|
|
The glance-registry service is returning 401 to httpchk, which
makes haproxy think it is down. This change switches the check
mode to tcp.
Closes-Bug: https://bugzilla.redhat.com/show_bug.cgi?id=1234637
Closes-Bug: 1468566
Change-Id: Icdd80aa9cd56e5afd3707eb7fa38aaedb8535af6
|
|
In the pacemaker scenario we want to use the clustercheck script
to evict galera nodes which are out of sync. This change adds a
parameter meant to enable use of clustercheck for the mysql service.
Change-Id: I7199c7e5d759a76f58c0f48b40e9d460a3163886
Closes-Bug: 1456701
|
|
We do not want to give users control over the galera_master_node,
this should be gathered using the clustercheck script instead.
Depends-On: I56ebd2d8405ac35c707666d993b396f04aeb683e
Change-Id: Ib6a36e9283b73133251fb9ff3f33e71c50edb3db
Closes-Bug: 1467918
Closes-Bug: https://bugzilla.redhat.com/show_bug.cgi?id=1234817
|
|
This patch optionally creates new virtual IPs for
the storage, storage_mgmt, and internal_api
networks if ip addresses are provided.
Additionally the HAproxy configuration is updated
to use hiera lookups to obtain virtual IPs for
alternate networks. By default the ctlplane
VIP is still used.
Change-Id: I20483574920a1da689374b0eb1b39b0391c3d243
|
|
This patch updates the loadbalancer class so that it
defaults to trying to use the node IP list for each
respective service. This data is provided via Hiera
directly (all-nodes-config provides it via the Heat
templates).
By default the ctlplane IP address list is still used
if no service node IP list is provided.
Change-Id: I34cbdf8bd525e6ab61859fe8b8c18fe613dabbfe
|
|
This patch removes the public VIP for the nova metadata
and glance registry services.
Change-Id: I0878f7b3eeed6e16c5d30bdf76ebca56eb49d042
|
|
There are two methods included: Setting $service_certificate will
enable SSL for all public endpoints with the same cert file, while
service-specific certificate settings allow using a cert file for
just one service.
Change-Id: I6f87ed2ebbea08ff1a0dff981559c8f4fc8b67cc
|
|
The per-instance limit to 150 can easily be reached for the
database when OpenStack services are running on hosts with many
CPUs. The global maxconn is increased as per astapor. See [1].
1. https://bugzilla.redhat.com/show_bug.cgi?id=1218322
Change-Id: Ia9258372ca4f707929f11097193a91c138069725
|
|
We want to let the caller decide upon having the haproxy
service started and enabled or not on boot.
Change-Id: I24a9fd2245a974120892a8887c8b58647c65cba9
|
|
|
|
|
|
|
|
|
|
Currently RabbitMQ is listening on both private vip and public vip.
There is no need for RabbitMQ to listen on the public vip, so we remove
it.
Change-Id: I82ea2e1e18b7710ae391ffe4903439a9330b1461
|
|
Add support for Redis in the loadbalancer setup.
Redis loadbalancing system is particular as it does not provide
clustering capabilities yet, hence this pattern[1] will be applied.
[1] https://github.com/falsecz/haredis
Change-Id: I80a6c284af9eceb6b669a03c5d93256261523331
|
|
Change-Id: I98b9b3dbc48009ce255d964ac580e1a31f279f1e
|
|
Enable access to the HAProxy stats page. The listen directive is bound
to the controller virtual IP address.
Change-Id: Ie0012da77ffdd9bfa8f06341aca2d70991558a28
|
|
Initial logic for the backup option in the HAProxy was wrong and
wouldn't do what expected.
Current logic is implemented as follow
1. User passes an array of Galera IP addresses
2. User passes an array of Galera hostnames
3. User passes a Galera master IP and Galera master hostname
Result :
* Set a backend line with Galera master IP and Galera master hostname
* Remove those data from the two arrays of IP and Hostnames
* Set backend lines for whatever is left on those array with the backup
option on
Change-Id: Idfd72de4fafdce2a9c16945961fee996a98049b7
|
|
Currently since only one node was the backend until this commit[1],
servername was deducted from the $::hostname fact. Since commit[1],
several node can be the backend for a service, so we need to provide
their servername also.
The current situation result with HAProxy refusing to start because
current lint look like
server 192.0.2.2:8776 check fall 5 inter 2000 rise 2
when they really should look like
server MYHOSTNAME 192.0.2.2:8776 check fall 5 inter 2000 rise 2
Resulting in error message : 'server' expects <name> and <addr>[:<port>]
as arguments.
[1] https://review.openstack.org/#/c/168044/
Change-Id: I75424cf02f2d24308f33105f67d82a8d411e372d
|
|
Since we can have many controller_hosts backend in a setup, we use the
plural term to define it.
Change-Id: I2a46c250bc3325eef9c3128cac2ab45c88b1ae75
|
|
This commit allows to enable backup mode for non master galera node.
Change-Id: I8b27f470ae171d77c8c8283797ff1502ef44e17f
|
|
|
|
This should allow puppet to validate the required params.
Change-Id: I16b6ae1a9fbcb388bfe5a2a95022a2fdffbf0cd1
|
|
- This changes the puppet-lint requirement to 1.1.x, so that we can use
puppet-lint plugins. Most of these plugins are for 4.x compat, but some just
catch common errors.
Change-Id: I2660b960b6ef696bd5dc8a6965b4a9aa25409b66
Signed-off-by: Gael Chamoulaud <gchamoul@redhat.com>
|