release-notes/3.12.0.md
RabbitMQ 3.12.0 is a new feature release.
This release goes out of community support on June 1st, 2024.
This release includes several new features, optimizations, and graduates (makes mandatory) a number of feature flags.
The user-facing areas that have seen the biggest improvements in this release are
This release also features many internal API improvements in preparation to 4.0 with Khepri.
See Compatibility Notes below to learn about breaking or potentially breaking changes in this release.
RabbitMQ releases are distributed via GitHub. Debian and RPM packages are available via Cloudsmith mirrors, as well as PackageCloud.
Community Docker image, Chocolatey package, and the Homebrew formula are other installation options. They are updated with a delay (usually a few days).
This release requires Erlang 25.0 or later. This introduces feature parity for x86- and ARM64-based CPUs: Erlang 25 offers the JIT and modern Flame Graph profiling tooling for both of those major CPU architectures.
Provisioning Latest Erlang Releases explains what package repositories and tools can be used to provision latest patch versions of Erlang 25.x.
This is the first release that offers Erlang 26 compatibility. Note that Erlang 26 is very new and is nowhere near as battle tested as Erlang 25.
See the Upgrading guide for documentation on upgrades and RabbitMQ change log for release notes of other releases.
RabbitMQ 3.12.0 will require all feature flags from the 3.11.x release series to be enabled before upgrading, similarly to how 3.11.0 required all feature flags introduced before 3.9.0.
If the feature flags are not enabled, RabbitMQ 3.12 and later nodes will refuse to start.
When upgrading from 3.11.x on Erlang 25 to 3.12.x on Erlang 26, the cluster must be on at least version 3.11.17 for a safe upgrade of quorum queues.
Alternatively, it is possible to upgrade all nodes to 3.12.0 on Erlang 25 from an earlier 3.11.x version, then upgrade the 3.12 cluster to Erlang 26.
RabbitMQ 3.12.0 nodes can run alongside 3.11.x nodes starting with 3.11.17.
3.12.x-specific features can only be made available when all nodes in the cluster
upgrade to 3.12.0 or any other patch release in the new series.
While operating in mixed version mode, some aspects of the system may not behave as expected. The list of known behavior changes is covered below. Once all nodes are upgraded to 3.12.0, these irregularities will go away.
Mixed version clusters are a mechanism that allows rolling upgrade and are not meant to be run for extended periods of time (no more than a few hours).
RabbitMQ 3.12.0 will require all feature flags from the 3.11.x release series to be enabled before upgrading.
If the feature flags are not enabled, RabbitMQ 3.12 and later nodes will refuse to start.
Some metrics displayed by the management UI are no longer available for MQTT clients. They are computed for AMQP 0-9-1 channels, which MQTT plugin no longer uses. They may be reintroduced in future releases when more protocol implementations (e.g. AMQP 1.0) are refactored the same way the MQTT plugin was for this release.
Connection-level metrics such as ingress and egress data flows are still available, and so are the metrics of queues used by MQTT subscribers.
Starting with this release, RabbitMQ requires Erlang 25.0 or later versions. Nodes will fail to start on older Erlang releases.
Erlang 25 as our new baseline means much improved performance on ARM64 architectures, profiling with flame graphs across all architectures, and the most recent TLS 1.3 implementation available to all RabbitMQ 3.11 users.
Client libraries that were compatible with RabbitMQ 3.11.x will be compatible with 3.12.0.
Any questions about this release, upgrades or RabbitMQ in general are welcome on the RabbitMQ mailing list.
Release notes are kept under rabbitmq-server/release-notes.
Quorum queues now sustain higher throughput with large backlogs. They also offer higher throughput on average, and in particular when Single Active Consumer is used.
GitHub issue: #7553
Reduced memory footprint, improved memory use predictability and throughput of classic queues. This particularly benefits classic queues with longer backlogs.
Reduced peak memory footprint of quorum queues.
GitHub issue: #7175
Improved stream leader distribution efficiency. Mostly relevant to environments with lots of streams.
GitHub issue: #6440
All feature flags introduced before 3.11.0 are now required to be enabled. Make sure all feature flags are enabled before upgrading to 3.12.0.
GitHub issue: #7219
Plugin and core API: all schema data store operations are now available via the rabbit_db_* modules.
Plugin maintainers should switch to them for an easier transition to a future 4.0 release.
Plugin and core API: node data directory now should be accessed using rabbit:data_dir/0.
GitHub issue: #6462
Initial (incomplete) Erlang 26 compatibility
There is now a way to pre-configure users and their permissions for newly created virtual hosts:
default_users.qa_user.vhost_pattern = qa.*
default_users.qa_user.tags = policymaker,monitoring
default_users.qa_user.password = fd237824441a78cd922410af4b83f0888186a8d7
default_users.qa_user.read = .*
default_users.qa_user.write = .*
default_users.qa_user.configure = .*
This is primarily useful in environments where RabbitMQ is provided as a service but customers (clients) have control over virtual hosts.
Contributed by @illotum (AWS).
GitHub issue: #7208.
Policies can now be defined to only apply to specific queue types.
For example, it is now possible to have two policies matching all queue names ('.*') but one applicable only to one queue type, and the other only to different queue type, without conflicts.
For example, this policy will only apply to quorum queues and not streams or classic queues:
rabbitmqctl set_policy at-least-once-dead-lettering ".*" '{"dead-letter-strategy": "at-least-once"}' --apply-to quorum_queues
Nodes with a lot of classic queues now shut down up to 10 times faster.
GitHub issue: #7745
Reduced metric collection-induced memory use spikes every two minutes in environments with a lot of mostly idle queues.
Delivery acknowledgement timeout now can be configured per queue (using a policy) or per consumer (using optional consumer arguments).
GitHub issue: #5437
It is now possible to limit the maximum number of virtual hosts that can be created in the cluster.
Contributed by @SimonUnge (AWS).
GitHub issue: #7777
Nodes with a lot of (non-mirrored) classic queues with v2 storage enabled
by default (using classic_queue.default_version = 2) now boot faster.
GitHub issue: #7676
It is now possible to limit how many shovels or federation links can run on a node
using rabbitmq.conf:
runtime_parameters.limits.shovel = 10
runtime_parameters.limits.federation = 10
Contributed by @illotum (AWS).
GitHub issue: #7917
Quorum queues: The x-delivery-count header will no longer be added to messages the first time they are delivered (x-delivery-count=0)
as it has a potential performance impact on consumer deliveries.
GitHub issue: #7732
Quorum queues will now log if they could not apply policy changes, for example, because there was no quorum of replicas online, or the queue was going through a leader election.
GitHub issue: #7853
This release includes all bug fixes shipped in the 3.11.x series.
It is now possible to use at-most-once dead lettering with streams (dead letter messages to a stream).
GitHub issue: #7846
It is now possible to use streams in combination with the event exchange.
GitHub issue: #7846
Superstream could fail to elect a single active consumer (SAC) in certain consumer churn conditions.
GitHub issue: #7743
Quorum queues now have a mechanism of shrinking down to just one replica. This is meant to be used exclusively for disaster recovery when a majority of nodes hosting replicas of a queue were permanently lost.
GitHub issue: #8322
Classic mirrored queues that had a length limit defined on them handled the overflow of messages differently from the current elected leader, eventually causing exceptions and replica restarts.
GitHub issue: #7579
When a queue was deleted and instantly re-declared with a different type,
polling consumers (those that use basic.get) ran into an exception.
GitHub issue: #4976
rabbitmq-queues rebalance now supports streams in addition to classic and quorum queues.
GitHub issue: #6440
rabbitmq-streams restart_stream is a new command that can be used to kick off
a new leader election for a stream. This is another option that can be used
to rebalance stream leaders across cluster nodes.
GitHub issue: #6440
Classic queue metrics are now available via rabbitmq-diagnostics observer.
GitHub issue: #3314
rabbitmqctl update_vhost_metadata is a new command that can be used to update
the description, default queue type, or tags of a virtual host:
rabbitmqctl update_vhost_metadata vh1 --tags qa,quorum,team3,project2
rabbitmqctl update_vhost_metadata vh1 --description "QA env 1 for issue 37483"
rabbitmqctl update_vhost_metadata vh1 --description "QQs all the way" --default-queue-type "quorum"
rabbitmqctl update_vhost_metadata vh1 --description "streaming my brain out" --default-queue-type "stream"
GitHub issue: #7914
rabbitmqctl set_permissions_globally is a new command that sets up user permissions in all existing virtual hosts.
GitHub issue: #1000
rabbitmq-diagnostics cluster_status now lists how many CPU cores are available to individual nodes, plus a total.
GitHub issue: #7135
rabbitmq-stream stream_status failed with an exception if one of the replicas was offline.
Contributed by @gomoripeti (CloudAMQP)
GitHub issue: #8270
Significant (up to more than 90% with some workloads) reduction in memory footprint of MQTT connections, and a double digit % reduction in memory footprint of MQTT-over-WebSockets (Web MQTT) connections.
The x-mqtt-dup header will no longer be present for consumer deliveries as it wasn't used correctly.
Fixed a quorum queue-specific issue that could prevent queue federation links from making any progress.
GitHub issue: #8328
Support for Identity Provider-initiated Login (IDP).
Potentially breaking change: deprecated UAA-specific configuration keys were removed:
management.enable_uaa, management.uaa_*.
GitHub issue: #7887
Classic queues now list their storage implementation version on the details page.
Contributed by @gomoripeti (CloudAMQP)
GitHub issue: #8244
Default queue type key used when add a virtual host via the HTTP API has changed from
defaultqueuetype to default_queue_type. default_queue_type will be supported
for a period of time for backwards compatibility.
GitHub issue: #7738
It was impossible to return to a tab that had a filter expression that was not a valid regular expressions.
Now such expressions will be used as regular text filters.
GitHub issue: #8008
It is now possible to configure (or strip) the scope prefix used by this plugin when translating token scopes to RabbitMQ permissions.
To do so, override auth_oauth2.scope_prefix in rabbitmq.conf:
# "rabbitmq." is the default prefix
auth_oauth2.scope_prefix = rabbitmq.
GitHub issue: #8001
Several variables ({username}, {vhost} and JWT claims that are single string values)
now can be used (expanded) in topic operation authorization.
GitHub issue: #7178
The authorization backend could run into an exception when used in combination with other backends.
GitHub issue: #7864
Consul peer discovery now correctly determines if the node uses long or short names.
Contributed by @SimonUnge (AWS).
GitHub issue: #4230
To obtain source code of the entire distribution, please download the archive named rabbitmq-server-3.12.0.tar.xz
instead of the source tarball produced by GitHub.