This patch adds loadbalancer-config role
which is "wrapper" around haproxy-config
and proxysql-config role which will be added
in follow-up patches.
Change-Id: I64d41507317081e1860a94b9481a85c8d400797d
this adds back the ability to configure
the rabbitmq/erlang kernel network interface
which was removed in https://review.opendev.org/#/c/584427/
seemingly by accident.
Closes-Bug: 1900160
Change-Id: I6f00396495853e117429c17fadfafe809e322a31
Follow up to I91d0e23b22319cf3fdb7603f5401d24e3b76a56e, which fixes a
conditional corner case when removing the ha-all policy.
Change-Id: Iea75551bc6d0da7dd10515dd8bd28c014eed7a5e
When OpenStack is deployed with Kolla-Ansible, by default there
are no durable queues or exchanges created by the OpenStack
services in RabbitMQ. In Rabbit terminology, not being durable
is referred to as `transient`, and this means that the queue
is generally held in memory.
Whether OpenStack services create durable or transient queues is
traditionally controlled by the Oslo Notification config option:
`amqp_durable_queues`. In Kolla-Ansible, this remains set to
the default of `False` in all services. The only `durable`
objects are the `amq*` exchanges which are internal to RabbitMQ.
More recently, Oslo Notification has introduced support for
Quorum queues [7]. These are a successor to durable classic
queues, however it isn't yet clear if they are a good fit for
OpenStack in general [8].
For clustered RabbitMQ deployments, Kolla-Ansible configures all
queues as `replicated` [1]. Replication occurs over all nodes
in the cluster. RabbitMQ refers to this as 'mirroring of classic
queues'.
In summary, this means that a multi-node Kolla-Ansible deployment
will end up with a large number of transient, mirrored queues
and exchanges. However, the RabbitMQ documentation warns against
this, stating that 'For replicated queues, the only reasonable
option is to use durable queues: [2]`. This is discussed
further in the following bug report: [3].
Whilst we could try enabling the `amqp_durable_queues` option
for each service (this is suggested in [4]), there are
a number of complexities with this approach, not limited to:
1) RabbitMQ is planning to remove classic queue mirroring in
favor of 'Quorum queues' in a forthcoming release [5].
2) Durable queues will be written to disk, which may cause
performance problems at scale. Note that this includes
Quorum queues which are always durable.
3) Potential for race conditions and other complexity
discussed recently on the mailing list under:
`[ops] [kolla] RabbitMQ High Availability`
The remaining option, proposed here, is to use classic
non-mirrored queues everywhere, and rely on services to recover
if the node hosting a queue or exchange they are using fails.
There is some discussion of this approach in [6]. The downside
of potential message loss needs to be weighed against the real
upsides of increasing the performance of RabbitMQ, and moving
to a configuration which is officially supported and hopefully
more stable. In the future, we can then consider promoting
specific queues to quorum queues, in cases where message loss
can result in failure states which are hard to recover from.
[1] https://www.rabbitmq.com/ha.html
[2] https://www.rabbitmq.com/queues.html
[3] https://github.com/rabbitmq/rabbitmq-server/issues/2045
[4] https://wiki.openstack.org/wiki/Large_Scale_Configuration_Rabbit
[5] https://blog.rabbitmq.com/posts/2021/08/4.0-deprecation-announcements/
[6] https://fuel-ccp.readthedocs.io/en/latest/design/ref_arch_1000_nodes.html#replication
[7] https://bugs.launchpad.net/oslo.messaging/+bug/1942933
[8] https://www.rabbitmq.com/quorum-queues.html#use-cases
Partial-Bug: #1954925
Change-Id: I91d0e23b22319cf3fdb7603f5401d24e3b76a56e
rabbitmq starting from 3.8.0, built-in Prometheus support,
prometheus plugins are enabled by default, when the environment is
"enable_prometheus is no", rabbitmq role will disable prometheus plugins
Closes-Bug: #1885106
Change-Id: I4d694d6224c813285d228d6bc7eece5731db1078
Role vars have a higher precedence than role defaults. This allows to
import default vars from another role via vars_files without overriding
project_name (see related bug for details).
Change-Id: I3d919736e53d6f3e1a70d1267cf42c8d2c0ad221
Related-Bug: #1951785
We get a nice optimisation by using a filtered loop instead
of task skipping per service with 'when'.
Partially-Implements: blueprint performance-improvements
Change-Id: I8f68100870ab90cb2d6b68a66a4c97df9ea4ff52
As mentioned in the Iced014acee7e590c10848e73feca166f48b622dc
commit message, in Ussuri+ we can use ``+sbwtdcpu none
+sbwtdio none`` as well. This is due to relying on RMQ-provided
erlang in version 23.x.
This change adds the extra arguments by default.
It should be backported down to Ussuri before we do a release with
Iced014acee7e590c10848e73feca166f48b622dc.
Change-Id: I32e247a6cb34d7f6763b544f247fd408dce2b3a2
By default, Ansible injects a variable for every fact, prefixed with
ansible_. This can result in a large number of variables for each host,
which at scale can incur a performance penalty. Ansible provides a
configuration option [0] that can be set to False to prevent this
injection of facts. In this case, facts should be referenced via
ansible_facts.<fact>.
This change updates all references to Ansible facts within Kolla Ansible
from using individual fact variables to using the items in the
ansible_facts dictionary. This allows users to disable fact variable
injection in their Ansible configuration, which may provide some
performance improvement.
This change disables fact variable injection in the ansible
configuration used in CI, to catch any attempts to use the injected
variables.
[0] https://docs.ansible.com/ansible/latest/reference_appendices/config.html#inject-facts-as-vars
Change-Id: I7e9d5c9b8b9164d4aee3abb4e37c8f28d98ff5d1
Partially-Implements: blueprint performance-improvements
The host list order seen during Ansible handlers may differ to the usual
play host list order, due to race conditions in notifying handlers. This
means that restart_services.yml for RabbitMQ may be included in a
different order than the rabbitmq group, resulting in a node other than
the 'first' being restarted first. This can cause some nodes to fail to
join the cluster. The include_tasks loop was introduced in [1].
This change fixes the issue by splitting the handler into two tasks, and
restarting the first node before all others.
[1] https://review.opendev.org/c/openstack/kolla-ansible/+/763137
Change-Id: I1823301d5889589bfd48326ed7de03c6061ea5ba
Closes-Bug: #1930293
On machines with many cores, we were seeing excessive CPU load on systems
that were not very busy. With the following Erlang VM argument we saw
RabbitMQ CPU usage drop from about 150% to around 20%, on a system with
40 hyperthreads.
+S 2:2
By default RabbitMQ starts N schedulers where N is the number of CPU
cores, including hyper-threaded cores. This is fine when you assume all
your CPUs are dedicated to RabbitMQ. Its not a good idea in a typical
Kolla Ansible setup. Here we go for two scheduler threads.
More details can be found here:
https://www.rabbitmq.com/runtime.html#scheduling
and here:
https://erlang.org/doc/man/erl.html#emulator-flags
+sbwt none
This stops busy waiting of the scheduler, for more details see:
https://www.rabbitmq.com/runtime.html#busy-waiting
Newer versions of rabbit may need additional flags:
"+sbwt none +sbwtdcpu none +sbwtdio none"
But this patch should be back portable to older versions of RabbitMQ
used in Train and Stein.
Note that information on this tuning was found by looking at data from:
rabbitmq-diagnostics runtime_thread_stats
More details on that can be found here:
https://www.rabbitmq.com/runtime.html#thread-stats
Related-Bug: #1846467
Change-Id: Iced014acee7e590c10848e73feca166f48b622dc
This change enables the use of Docker healthchecks for rabbitmq services.
Implements: blueprint container-health-check
Depends-On: https://review.opendev.org/c/openstack/kolla/+/784562
Change-Id: I23a2c2efab858b9ed39c6ce0ec4a82df10e7f93d
This reverts commit 9cae59be51e8d2d798830042a5fd448a4aa5e7dc.
Reason for revert: This patch was found to introduce issues with fluentd customisation. The underlying issue is not currently fully understood, but could be a sign of other obscure issues.
Change-Id: Ia4859c23d85699621a3b734d6cedb70225576dfc
Closes-Bug: #1906288
Main plays are action-redirect-stubs, ideal for import_tasks.
This avoids 'include' penalty and makes logs/ara look nicer.
Fixes haproxy and rabbitmq not to check the host group as well.
Change-Id: I46136fc40b815e341befff80b54a91ef431eabc0
Partially-Implements: blueprint performance-improvements
Config plays do not need to check containers. This avoids skipping
tasks during the genconfig action.
Ironic and Glance rolling upgrades are handled specially.
Swift and Bifrost do not use the handlers at all.
Partially-Implements: blueprint performance-improvements
Change-Id: I140bf71d62e8f0932c96270d1f08940a5ba4542a
This change adds support for encryption of communication between
OpenStack services and RabbitMQ. Server certificates are supported, but
currently client certificates are not.
The kolla-ansible certificates command has been updated to support
generating certificates for RabbitMQ for development and testing.
RabbitMQ TLS is enabled in the all-in-one source CI jobs, or when
The Zuul 'tls_enabled' variable is true.
Change-Id: I4f1d04150fb2b5af085b762890092f87ae6076b5
Implements: blueprint message-queue-ssl-support
Including tasks has a performance penalty when compared with importing
tasks. If the include has a condition associated with it, then the
overhead of the include may be lower than the overhead of skipping all
imported tasks. For unconditionally included tasks, switching to
import_tasks provides a clear benefit.
Benchmarking of include vs. import is available at [1].
This change switches from include_tasks to import_tasks where there is
no condition applied to the include.
[1] https://github.com/stackhpc/ansible-scaling/blob/master/doc/include-and-import.md#task-include-and-import
Partially-Implements: blueprint performance-improvements
Change-Id: Ia45af4a198e422773d9f009c7f7b2e32ce9e3b97
Previously we mounted /etc/timezone if the kolla_base_distro is debian
or ubuntu. This would fail prechecks if debian or ubuntu images were
deployed on CentOS. While this is not a supported combination, for
correctness we should fix the condition to reference the host OS rather
than the container OS, since that is where the /etc/timezone file is
located.
Change-Id: Ifc252ae793e6974356fcdca810b373f362d24ba5
Closes-Bug: #1882553
Including tasks has a performance penalty when compared with importing
tasks. If the include has a condition associated with it, then the
overhead of the include may be lower than the overhead of skipping all
imported tasks. In the case of the check-containers.yml include, the
included file only has a single task, so the overhead of skipping this
task will not be greater than the overhead of the task import. It
therefore makes sense to switch to use import_tasks there.
Partially-Implements: blueprint performance-improvements
Change-Id: I65d911670649960708b9f6a4c110d1a7df1ad8f7
The common role was previously added as a dependency to all other roles.
It would set a fact after running on a host to avoid running twice. This
had the nice effect that deploying any service would automatically pull
in the common services for that host. When using tags, any services with
matching tags would also run the common role. This could be both
surprising and sometimes useful.
When using Ansible at large scale, there is a penalty associated with
executing a task against a large number of hosts, even if it is skipped.
The common role introduces some overhead, just in determining that it
has already run.
This change extracts the common role into a separate play, and removes
the dependency on it from all other roles. New groups have been added
for cron, fluentd, and kolla-toolbox, similar to other services. This
changes the behaviour in the following ways:
* The common role is now run for all hosts at the beginning, rather than
prior to their first enabled service
* Hosts must be in the necessary group for each of the common services
in order to have that service deployed. This is mostly to avoid
deploying on localhost or the deployment host
* If tags are specified for another service e.g. nova, the common role
will *not* automatically run for matching hosts. The common tag must
be specified explicitly
The last of these is probably the largest behaviour change. While it
would be possible to determine which hosts should automatically run the
common role, it would be quite complex, and would introduce some
overhead that would probably negate the benefit of splitting out the
common role.
Partially-Implements: blueprint performance-improvements
Change-Id: I6a4676bf6efeebc61383ec7a406db07c7a868b2a
Some services look for /etc/timezone on Debian/Ubuntu, so we should
introduce it to the containers.
In addition, added prechecks for /etc/localtime and /etc/timezone.
Closes-Bug: #1821592
Change-Id: I9fef14643d1bcc7eee9547eb87fa1fb436d8a6b3
Both include_role and import_role expect role's name to be given
via "name" param instead of "role".
This worked but caused errors with ansible-lint.
See: https://review.opendev.org/694779
Change-Id: I388d4ae27111e430d38df1abcb6c6127d90a06e0
We assume that all groups are present in the inventory, and quite obtuse
errors can result if any are not.
This change adds a precheck that checks for the presence of all expected
groups in the inventory for each service. It also introduces a common
service-precheck role that we can use for other common prechecks.
Change-Id: Ia0af1e7df4fff7f07cd6530e5b017db8fba530b3
Partially-Implements: blueprint improve-prechecks
Make it require uniqueness of resolution as well to avoid later
issues with RabbitMQ going crazy.
Change-Id: I000ba6c62ab44eac0abdf8d5d1f069adfbc6552f
Closes-bug: #1863363
