Skip to main content

Classic Queues

What is a Classic Queue​

A RabbitMQ classic queue (the original queue type) is a versatile queue type suitable for use cases where data safety is not a priority because the data stored in classic queues is not replicated. Classic queues uses the non-replicated FIFO queue implementation.

If data safety is a priority, the recommendation is to use quorum queues and streams instead of classic queues.

Classic queues are the default queue type as long as the default queue type is not overriden for the virtual host.

There are two versions (implementations) of classic queue message storage and indexing. The version only impacts how the data is stored on and read from disk: all features are available in both versions.

Classic Queue Features​

Classic queues fully support queue exclusivity, queue and message TTL (Time-To-Live), queue length limits, message priority, consumer priority and adhere to settings controlled using policies.

Classic queues support dead letter exchanges with the exception of at-least-once dead-lettering.

Classic queues do not support poison message handling, unlike quorum queues. Classic queues also do not support at-least-once dead-lettering, suported by quorum queues.

Per-consumer QoS prefetch should be preferred over global QoS prefetch, even though classic queues support both options. Global QoS prefetch is a deprecated feature that will be removed in RabbitMQ 4.0.

While classic queues can be declared as transient, this makes queue removal difficult to reason about in case of node restarts during upgrades and such, so the use of transient queues is discouraged. Support for transient queues is deprecated and will be removed in RabbitMQ 4.0.

Classic queues can be mirrored across multiple nodes in the cluster. This functionality is deprecated and will be removed in RabbitMQ 4.0. Quorum queues and streams provide a better alternative when high availability and data safety is required.

Until RabbitMQ 3.12, classic queues could operate in lazy mode. As of RabbitMQ 3.12 the queue mode is ignored and classic queues behave in a similar manner to lazy queues. They only keep a small number of messages in memory based on the consumption rate; all other messages are written to disk directly.

Persistence (Durable Storage) in Classic Queues​

Classic queues use an on-disk index for storing message locations on disk as well as a message store for persisting messages.

Both persistent and transient messages are always persisted to disk except when:

  • the queue is declared as transient or messages are transient
  • messages are smaller than the embedding threshold (defaults to 4096 bytes)
  • for RabbitMQ 3.12 and later versions: the queue is short (queues may keep up to 2048 messages in memory at most, depending on the consumer delivery rate)

In general messages are not kept in memory unless the rate of consumption of messages is high enough that the messages that are in memory are expected to be consumed within the next second. Classic queues keep up to 2048 messages in memory, depending on the consumer delivery rate. Larger messages are not read into memory until the moment when they have to be sent to consumers.

Persisted messages may be embedded in the queue or sent to a shared message store. The decision to store messages in the queue or in the shared message store is based on the size of the message, including headers. The shared message store is more efficient at handling larger messages, particularly when those messages are sent to multiple queues.

The message location is written in the queue's index. Each queue has one index. The queue is responsible for tracking messages location as well as their position in the queue, and it persists this information in the index.

Embedded messages are written in its queue index when using classic queues version 1; and in its per-queue message store when using classic queues version 2.

Larger messages are written to a shared message store. Each vhost has two such stores: one for persistent messages and one for transient messages, but they are usually considered together as the shared message store. All queues in the vhost use the same message store.

Classic Queue Storage Implementation Versions​

There are currently two classic queue versions (implementation). Depending on the version, classic queues will use a different index for messages, as well as operate differently regarding the embedding of small messages in the index.

Classic Queue Implementation Version 1​

Version 1 is the default and the original implementation of classic queues. The index in version 1 uses a combination of a journal and segment files. When reading messages from segment files it loads an entire segment file in memory, which can lead to memory issues but does reduce the amount of I/O performed. Version 1 embeds small messages in the index, further worsening memory issues.

Classic Queue Implementation Version 2​

Version 2 takes advantage of the improved performance of modern storage devices and is the recommended version. The index in version 2 only uses segment files and only loads messages from disk when necessary. It will load more messages based on the current consumption rate. Version 2 does not embed messages in its index, instead a per-queue message store is used.

Version 2 was added in RabbitMQ 3.10.0 and was significantly improved in RabbitMQ 3.12.0. It is currently possible to switch back and forth between version 1 and version 2. In the future, version 1 will be removed and the migration to version 2 will be performed automatically on node startup after the upgrade.

The version can be changed using the queue-version policy key. When setting a new version via policy the queue will immediately convert its data on disk. It is possible to upgrade to version 2 or downgrade to version 1. Note that for large queues the conversion may take some time and results in the queue being unavailable while the conversion is running. As a point of reference, on our test machine, the migration takes:

  • 2 seconds to migrate 1000 queues with 1000 100-byte messages each
  • 9 seconds to migrate a queue with 1 million 100-byte messages
  • 3 seconds to migrate a queue with 1 million 5000-byte messages (with the default embedding size of 4096 bytes, 5000-byte messages are in the message store so there is less data to migrate)

Given the numbers above, unless there is a lot of queues with a lot of messages, the migration should complete in a matter of seconds.

The default version can be set through configuration by setting classic_queue.default_version in rabbitmq.conf. Changing the default version only affects newly declared queues. Pre-existing queues will remain on version 1, until explicitly migrated or deleted and redeclared.

Resource Use with Classic Queues​

Classic queues aim to provide reasonably good throughput in the majority of situations without configuration. However, some configuration is sometimes useful. This section covers a few configurable values that affect stability, throughput, latency and I/O characteristics of a node. Consider getting accustomed to benchmarking with PerfTest in addition to get the most out of your queues.

Some related information includes:

File Handle Usage with Classic Queues​

The RabbitMQ server is limited in the number of file handles it can open. Every running network connection requires one file handle, and the rest are available for queues to use.

Classic queues behave differently with regard to file handle usage depending on the version used.

Version 1 tries to avoid running out of file descriptors. If there are more disk-accessing queues than file handles after network connections have been taken into account, then the disk-accessing queues will share the file handles among themselves; each gets to use a file handle for a while before it is taken back and given to another queue.

This prevents the server from crashing due to there being too many disk-accessing queues, but it can become expensive. The management plugin can show I/O statistics for each node in the cluster; as well as showing rates of reads, writes, seeks and so on it will also show a rate of file handle churn — the rate at which file handles are recycled in this way. A busy server with too few file handles might be doing hundreds of reopens per second - in which case its performance is likely to increase notably if given more file handles.

Version 2 does not try to accomodate for low numbers of file descriptors anymore. It expects servers to have a large file descriptor limit configured and to always be able to open a new file handle when necessary. The index keeps up to 4 file handles open at any time, and the per-queue store keeps 1 file handle open but may open another one when flushing data to disk. This means that each queue needs up to 6 file descriptors available to properly function, in theory. In practice only busy queues will need that many; other queues will function just fine with 3 or 4 file handles.

As a result of not using the file handle management subsystem, version 2 does not track as many I/O statistics; only the numbers of reads and writes. Other metrics can be obtained at the OS level.

Memory Footprint with Classic Queues​

Classic queues may keep up to 2048 messages in memory, depending on the consume rate. Classic queues will, however, avoid reading larger messages from disk too early. In RabbitMQ 3.12 this means messages larger than the embedded threshold (by default, 4096 bytes).

The index in version 1 has to read entire segment files in order to access the messages inside. This can lead to memory usage spikes, especially when many messages are embedded. This also leads to spikes in CPU usage due to increased garbage collection. In addition, the index will buffer writes and keep data up to 1MB in memory before flushing to disk.

The index and per-queue store in version 2 will buffer entries. This is typically not a concern as far as the index is concerned since it only tracks metadata. The per-queue store will however use up to 1MB of memory by default (512KB in the write buffer and 512KB in a cache). When flushing to disk the store will first clear the cache then move the messages in the write buffer to the cache, effectively replacing the data in the cache with the data in the write buffer. The size of the write buffer and the cache are therefore linked. It can be configured using the advanced config via rabbit's classic_queue_store_v2_max_cache_size parameter.

Idle queues will reduce their memory usage. This can sometimes result in surprising spikes when performing operations that affect many queues, such as defining new policies. In that case the queues will need to allocate more memory again. The more queues, the bigger the spike should be expected.

The shared message store needs an index. The default message store index uses a small amount of memory for every message in the store.

Alternate Message Store Index Implementations​

As mentioned above, each message which is written to the message store uses a small amount of memory for its index entry. The message store index is pluggable in RabbitMQ, and other implementations are available as plugins which can remove this limitation.

The reason they are not shipped with the RabbitMQ distribution is that they all use native code. Note that such plugins typically make the message store run more slowly.