openstack/openstack-ansible-rabbitmq_server - openstack-ansible-rabbitmq_server - OpenDev: Free Software Needs Free Tools

Current code de-facto ignores rabbitmq_erlang_install_method variable
value for debian-based distros.
We need to use it for erlang repo installation, otherwise it may lead
to role failure.
Change-Id: I826fec7febe3bda5748a0e90fa80fdb8586d63ac
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

After removal of content from ppa1.rabbitmq.com there
are no other sources for the erlang, as deb1 does not contain
trixie packages for erlang so far.
We should revert this change as soon as packages appear in deb1 [1]
[1] https://github.com/rabbitmq/erlang-debian-package/issues/45
Change-Id: I8a23fdf2693c2d8cb3a03563e912151f1c69cb49
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

Old repo (ppa1.rabbitmq.com) was put down in favor
of deb1.rabbitmq.com, so we need to switch to it
for erlang installation as well.
Change-Id: I1bbb4b5a685cc6526e33fde429085d8ec8198c5d
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

There are 2 mirrors available for erlang/rabbitmq - yum1 and yum2.
Ansible's yum_repository module baseurl [1] can accept a list, so worth
passing both mirrors for ability to failover.
[1] https://docs.ansible.com/ansible/latest/collections/ansible/builtin/yum_repository_module.html#parameter-baseurl
Change-Id: If42695aa347cb478b7f795726975724afcbecc22
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

RabbitMQ team has reverted their support policy, and made
4.1 series supported for 18 month again.
Change-Id: Ifb2336b8b2d1b3b3118934a8a0727059e017df59
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

RabbitMQ has changed their docs to suggest using deb1.rabbitmq.com
instead of ppa1.rabbitmq.com.
For Debian Trixie ppa1.rabbitmq.com contains only RabbitMQ 3.12,
while deb1.rabbitmq.com contains all of 3.13, 4.0 and 4.1
We leave erlang as is for now, as it's not available for Trixie yet.
Change-Id: I98dfeb1706701787e7a7e912e37910108cec3a10
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

RabbitMQ supports Erlang 27 since release 4.0.4 [1]
With that Debian Trixie package is available only for Erlang 27
[1] https://www.rabbitmq.com/docs/which-erlang
Depends-On: https://review.opendev.org/c/openstack/openstack-ansible-plugins/+/965798
Change-Id: I055941dd6829c4055377fd46ec6cec8f9351546e
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

Change-Id: Ia5940434b593dc993bbd9fef11b483475f8a96db
Signed-off-by: Ivan Anfimov <lazekteam@gmail.com>

use full broker name.
trivial fix
Change-Id: I976d120e91ae0ecbd391d261c579d328eb60bf43
Signed-off-by: Ivan Anfimov <lazekteam@gmail.com>

In case of defining different IP address for SSH and Management
network in metal scenario, as described in [1], you get SSH address
defined as `ansible_host`, which results in HAProxy configuring
backends on SSH address rather then management one, where
service is actually listening.
We use `ansible_host` as a fallback for non-OSA usage of the role.
[1] https://docs.openstack.org/openstack-ansible/latest/reference/inventory/configure-inventory.html#having-ssh-network-different-from-openstack-management-network
Change-Id: I85a164b181b41025f8f872dbdf5426bf58f6f36e
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

The module has quite massive output due to included GPG key
into the structure, which makes it barely readable.
Using loop label on the module will limit output to necessary set
of arguments.
Change-Id: I4d050312410951847413c728165a74ce26fd77a0
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

During migration to Quorum queues we have renamed vhosts
to not contain leading slash for them. However, some services
at the same time were configured not to have RPC messaging,
but only notifications whenever needed - like keystone, glance, etc.
This left such vhosts still containing HA policy on them, leading
to RabbitMQ upgrade failures with:
```
rabbitmq-server[2934794]: Application rabbit exited with reason:
{{failed_to_deny_deprecated_features,[classic_queue_mirroring]},{rabbit,start,[normal,[]]}}
```
This patch implements a check of deprecated feature flags and
disables all HA policies across the board before upgrade happens.
As we don't want to force removal when running RabbitMQ 3.13,
the removal is enforced only during upgrade to 4.0.
Conditions are maintained on a task level rather then a block
to make it extendable with other deprecated flags in the future.
Change-Id: I372421e0347687a1d861b548fd20ee92b150c301
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

su/runuser is required for some rabbitmqctl commands, thus
let's ensure they're present for all cases.
This intends to fix CentOS 10 Stream molecule jobs at the first place.
Change-Id: I4fe224b471eff483b442d0256da49f32440549ec
Depends-On: https://review.opendev.org/c/openstack/openstack-ansible/+/957859
Signed-off-by: Dmitriy Rabotyagov <dmitriy.rabotyagov@cleura.com>

Change-Id: I167743ef596b1041acd8e3820c2d981fde90ec50

With no specific repository for EL10, packages built for EL9 are
installable and seems working at first glance.
Once an appropriate repository is published, this commit can be reversed
Change-Id: Ic80b316b69dc6b8086d90cfff46ceb318072ac0a

Change-Id: Ic5c9a6d0a13538b7e31851f9cbbfafb8e654e4fe

The tags framework has been discontinued for a long time.
https://governance.openstack.org/tc/reference/tags/
https://governance.openstack.org/tc/resolutions/20211224-tags-framework-removal.html
Change-Id: Iba8f67aaa663747da22843435f0d38ae54b0965b

Current implementation of rabbitmq-upgrade post_upgrade [1] performs
re-balance of queues between nodes, once upgrade is completed.
Which might be a very reasonable task to execute after any upgrade.
[1] https://www.rabbitmq.com/docs/man/rabbitmq-upgrade.8#post_upgrade
Change-Id: I54b8e22d3c8a3ea797724611f542d9e17982d92c

In case it is unsafe to shutdown current node due to queues going
below their minimal quorum size, we attempt to fix them by growing
affected queues to rest of the cluster.
This step is vital one in case of OS upgrades, when operating system
on controller is being re-installed, as quorum/stream queues
must be manually stretched after re-install.
Change-Id: I87e2ae7b43dbcf77c9f166656686d25dbd8937fd

Ansible's fail will fail only current host in case of met conditions
while continuing play for others.
In order to prevent further execution for all hosts in case of unmet
conditions, which may further break the cluster, `any_errors_fatal`
was added to fail tasks.
This will end the play for current batch and cancel further batches.
Change-Id: I92f24c77b5fb8a3dd9ed17915d7c725d15f9f21e

At the moment we simply skip the upgrade preparation tasks
in case cluster is not healthy or meeting expectations for the role
to proceed with the upgrade. This may lead to unexpected behaviour
and unobvious failures later on in the role.
Instead, we add another failure before node upgrade preparation is
run which will display all noticed inconsistencies and human readable
message for each occurance.
Change-Id: I0f252d08b2bcb94bca142dce5e376b5c3bc06417

Change-Id: I3c2751da86492c8659252cf5d7b691ce2cc111cf

Since ansible-core 2.10 it is recommended to use modules via FQCN
In order to align with recommendation, we perform migration
by applying suggestions made by `ansible-lint --fix=fqcn`
Change-Id: Id8a34727251ade1a1b2d5a881e5ee5e808b495dc

In order to reduce divergance with ansible-lint rules, we apply
auto-fixing of violations.
In current patch we replace all kind of truthy variables with
`true` or `false` values to align with recommendations along with
alignment of used quotes.
Change-Id: Ie295a52b403df4d404bf984fe0296d5be110f532

Depends-On: https://review.opendev.org/c/openstack/openstack-ansible/+/939824
Change-Id: Iba67d3e321c6ebc256bb63de27ce61c26db10ab0

The service is unconditionally stopped during the upgrade
prep tasks. There are several conditions that will notify
the handler to restart the service, such as updating the
config file or installing a new version of the rabbitmq
package. It is possible that none of these things happen
resulting in the service not being restarted during an upgrade.
This patch ensures that the service is always restarted
during an upgrade by notifying the restart handler when
the service is sucessfully stopped in the upgrade prep tasks.
Change-Id: I9fbe7806b3531921b90ac91c17b6a07deca2a4d2

There is currently a mismatch between the apt pin priorities for the
erlang and server rabbitmq repos. This patch sets the pinning to be
the same.
This should not affect normal package installation as recent changes
to this role require installation of a specific version and will
fail "safe" should that be unavailable. The intention with this patch
is to ensure that the behaviour is the same for both the erlang
and server repos in an abnormal situation where the installation
might fail.
Change-Id: I0d988572a102ca148059d198379d2a3115ecb795

With ansible-core 2.16 a breaking changes landed [1] to some filters
making their result returned in arbitrary order. With that, we were
relying on them to always return exactly same ordered lists.
With that we need to ensure that we still have determenistic behaviour
where this is important.
[1] https://github.com/ansible/ansible/issues/82554
Change-Id: I9ca1c66f88e9beea0e2c5add30a356fb717cfcfa

Having rabbitmq_package_state latest by default is not always expected
as all other roles do respect defined `package_state` value.
Change-Id: I5bcc550419cfe6d51dfb73eb935a93b93e460068

We are aligned on used RabbitMQ/Erlang versions for quite a while
now so it makes sense to move version defenition to defaults
instead of carrying them separately per distro family.
Change-Id: I5ca75bf25d099260bc446a6b4f7d9be7a2bdde58

Depends-On: https://review.opendev.org/c/openstack/openstack-ansible/+/934042
Change-Id: I231ea65b5340c43bffb541ce23d3a8e600015c2d

This makes us possible to leverage version comparison instead of search.
This also might come handy in the future when we need to be more
specific of expected versions for upgrade process.
Change-Id: I7a132145bc781db999f82ecb192e6e1ecfc9c0ad

Due to historical reasons RabbitMQ operations, like start/stop/restart
were perfromed in tasks and triggered on changes.
As of today there seems to be no need in such approach and we can use
handlers instead.
It will also move some tasks, which intend to run when service is
running, to their own file. This also allows to run them once on cluster
rather then for each host.
We also merge set_cookie with post_install as there is no reason to
maintain the task in it's own file.
Change-Id: I222b9de3ca531724b05c7fa2cf032a3271055d64

We can simplify logic and re-use rabbitmq_cluster_state tasks to
fetch the current state of the cluster in a dictionary, which allows
to avoid unnecessary today parsing of rabbitmq output.
Change-Id: Ib99fcb1a3e0092bf96143435a0c876ab6f286f36

At the moment rabbitmq_upgrade will temporary stop all nodes in the
cluster, shortly disrupting operations.
This behavior is now changed to perform a regular rolling upgrade [1]
Community right now relying on the rabbitmq_upgrade variable to
"recover" the cluster state by basically re-bootstrapping the cluster.
Such behaviour should be implemented separately in a follow-up patch.
[1] https://www.rabbitmq.com/docs/upgrade#rolling-upgrade
Depends-On: https://review.opendev.org/c/openstack/openstack-ansible-plugins/+/931801
Change-Id: Ibb3fae0d85dc31237ce169bb4cde9551dd5cabb6