Apache
/
activemq-artemis
mirror of https://github.com/apache/activemq-artemis.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
activemq-artemis/docs/user-manual/en/address-model.md

540 lines
18 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# Address Model
Every messaging protocol and API that Apache ActiveMQ Artemis supports defines
a different set of messaging resources.
- JMS uses _queues_ and _topics_
- STOMP uses generic _destinations_
- MQTT uses _topics_
- AMQP uses generic _nodes_
In order to deal the the unique semantics and use-cases for each of these the
broker has a flexible and powerful address model based on the following _core_
set of resources:
- **address**
- **queue**
- **routing type**
### Address
Messages are *sent* to an address. An address is given a unique name, a routing
type, and zero or more queues.
### Queue
Messages are *consumed* from a queue. A queue is bound to an address. It is
given a unique name and a routing type. There can be zero or more queues bound
to one address. When a message is sent to an address it is routed to one or
more of its queues based on the configured routing type.
The name of the queue must be _globally_ unique. For example, you can't have a
queue named `q1` on address `a1` and also a queue named `q1` address `a2`.
### Routing Type
A routing type determines how messages are routed from an address to the
queue(s) bound to that address. Two different routing types are supported,
**anycast** and **multicast**.
If you want your messages routed to...|Use this routing type...
---|---
a single anycast queue on the address.|anycast
every multicast queue on the address.|multicast
> **Note:**
>
> It is possible to define queues with a different routing type for the same
> address, but this typically results in an anti-pattern and is therefore not
> recommended.
## Automatic Configuration
By default Apache ActiveMQ Artemis will automatically create addresses and
queues to support the semantics of whatever protocol you're using. The broker
understands how to support each protocol's functionality with the core
resources so that in most cases no manual configuration is required. This saves
you from having to preconfigure each address and queue before a client can
connect to it.
The broker can optionally be configured to automatically delete addresses and
queues when they are no longer in use.
Automatic creation and deletion is configured on a per address basis and is
controlled by the following [`address-setting`](address-settings.md) elements:
- `auto-create-addresses`
- `auto-delete-addresses`
- `default-address-routing-type`
- `auto-create-queues`
- `auto-delete-queues`
- `default-queue-routing-type`
See [the documentation on address settings](address-settings.md) for more
details on these elements.
Of course, automatic configuration can be disabled and everything can be
configured manually. Read on for more details about manual configuration.
## Basic Manual Configuration
The following examples show how to configure resources for basic anycast and
multicast use-cases.
> **Note:**
>
> Many of the details of these use-cases are protocol agnostic. The goal here
> is to demonstrate and explain the basic configuration elements and how the
> address model works fundamentally.
### Anycast
The most common use-case for anycast semantics, sometimes referred to as
[point-to-point](messaging-concepts.md#point-to-point), involves applications
following a "competing consumer" pattern to receive messages from a shared
queue. The more consumers receiving messages the greater the overall message
throughput. Multiple Java applications sharing a JMS queue is a classic example
of this use-case.
In this use-case the broker is configured, for example, with an address,
`anycast.foo` using the `anycast` routing type with just one queue, `q1`. When
a producer sends a message to `address.foo` it is then routed to `q1` and
finally dispatched to one of the consumers.
![Anycast](images/anycast.png)
Figure 1. Anycast
This is what the configuration for this use-case would look like in
`etc/broker.xml`:
```xml
<addresses>
<address name="address.foo">
<anycast>
<queue name="q1"/>
</anycast>
</address>
</addresses>
```
For most protocols and APIs which support this kind of use-case (e.g. JMS,
AMQP, etc.) it is customary to use the _same name_ when sending and consuming
messages. In that case you'd use a configuration like this:
```xml
<addresses>
<address name="orderQueue">
<anycast>
<queue name="orderQueue"/>
</anycast>
</address>
</addresses>
```
### Multicast
The most common use-case for multicast semantics, sometimes referred to as
[publish/subscribe](messaging-concepts.md#publish-subscribe) or "pub/sub",
involves each application receiving every message sent to an address. Multiple
applications consuming from a JMS topic is a classic example of this use-case.
MQTT subscriptions is another supported example of multicast semantics.
In this use-case the broker is configured with an address, `address.foo` using
the `multicast` routing type with two queues, `q1` & `q2`. When a producer
sends a message to `address.foo` it is then routed to *both* `q1` & `q2` so
that ultimately both consumers receive the same messages.
![Multicast](images/multicast.png)
Figure 2. Multicast
This is what the configuration for this use-case would look like in
`etc/broker.xml`:
```xml
<addresses>
<address name="address.foo">
<multicast>
<queue name="q1"/>
<queue name="q2"/>
</multicast>
</address>
</addresses>
```
This basic configuration is simple and straight-forward, but there's a
problem. In a normal pub/sub use-case like with a JMS topic or with MQTT the
number of subscribers *isn't known ahead of time*. In that case, this is the
recommended configuration:
```xml
<addresses>
<address name="address.foo">
<multicast/>
</address>
</addresses>
```
Define `<multicast/>` with no queues and the broker will automatically create
queues for each subscription when the consumers connect to `address.foo`. Then
when a message is sent to `address.foo` it will be routed to each queue for
each subscriber and therefore each subscriber will get every message. These
queues are often referred to as _subscription queues_ for obvious reasons.
These subscription queues are typically named based on the semantics of the
protocol used to create them. For example, JMS supports durable and non-durable
subscriptions. The queue for a non-durable subscription is named with a UUID,
but the queue used for a durable subscription is named according to the JMS
"client ID" and "subscription name." Similar conventions are used for AMQP,
MQTT, STOMP, etc.
## Advanced Manual Configuration
### Fully Qualified Queue Names
In most cases it’s not necessary or desirable to statically configure the
aforementioned subscription queues. However, there are scenarios where a user
may want to statically configure a subscription queue and later connect to that
queue directly using a **Fully Qualified Queue Name** (FQQN).
An FQQN uses a special syntax to specify *both* the address and the queue so
that applications using protocols and APIs which don't natively understand the
address/queue separation (e.g. AMQP, JMS, etc.) can send messages or subscribe
*directly* to a queue rather than being limited to the address. Applications
simply need to use the address name and the queue name separated by `::` (e.g.
`address::queue`).
In this example, the address `a1` is configured with two queues: `q1`, `q2` as
shown in the configuration below.
```xml
<addresses>
<address name="a1">
<multicast>
<queue name="q1" />
<queue name="q2" />
</multicast>
</address>
</addresses>
```
Here's a snippet of Java code using JMS which demonstrates the FQQN syntax:
```java
Queue q1 session.createQueue("a1::q1");
MessageConsumer consumer = session.createConsumer(q1);
```
> **Note**
>
> The string `::` should only be used for FQQN and not in any other context in
> address or queue names.
The examples below show how to use broker side configuration to statically
configure a queue with publish subscribe behavior for shared, non-shared,
durable and non-durable subscription behavior.
#### Shared, Durable Subscription Queue using `max-consumers`
The default behavior for queues is to not limit the number connected queue
consumers. The `max-consumers` parameter of the queue element can be used to
limit the number of connected consumers allowed at any one time.
Open the file `etc/broker.xml` for editing.
```xml
<addresses>
<address name="durable.foo">
<multicast>
<!-- pre-configured shared durable subscription queue -->
<queue name="q1" max-consumers="10">
<durable>true</durable>
</queue>
</multicast>
</address>
</addresses>
```
#### Non-shared, Durable Subscription Queue
The broker can be configured to prevent more than one consumer from connecting
to a queue at any one time. The subscriptions to queues configured this way are
therefore "non-shared". To do this simply set the `max-consumers` parameter to
`1`:
```xml
<addresses>
<address name="durable.foo">
<multicast>
<!-- pre-configured non shared durable subscription queue -->
<queue name="q1" max-consumers="1">
<durable>true</durable>
</queue>
</multicast>
</address>
</addresses>
```
#### Non-durable Subscription Queue
Non-durable subscriptions are again usually managed by the relevant protocol
manager, by creating and deleting temporary queues.
If a user requires to pre-create a queue that behaves like a non-durable
subscription queue the `purge-on-no-consumers` flag can be enabled on the
queue. When `purge-on-no-consumers` is set to `true`. The queue will not
start receiving messages until a consumer is attached. When the last consumer
is detached from the queue. The queue is purged (its messages are removed)
and will not receive any more messages until a new consumer is attached.
Open the file `etc/broker.xml` for editing.
```xml
<addresses>
<address name="non.shared.durable.foo">
<multicast>
<queue name="orders1" purge-on-no-consumers="true"/>
</multicast>
</address>
</addresses>
```
### Disabled Queue
If a user requires to statically configure a queue and disable routing to it,
for example where a queue needs to be defined so a consumer can bind, but you
want to disable message routing to it for the time being.
Or you need to stop message flow to the queue to allow investigation keeping
the consumer bound, but don't wish to have further messages routed to the queue
to avoid message build up.
When `enabled` is set to `true` the queue will have messages routed to it.
(default)
When `enabled` is set to `false` the queue will NOT have messages routed to it.
Open the file `etc/broker.xml` for editing.
```xml
<addresses>
<address name="foo.bar">
<multicast>
<queue name="orders1" enabled="false"/>
</multicast>
</address>
</addresses>
```
> **Warning:**
>
> Disabling all the queues on an address means that any message sent to that
> address will be silently dropped.
### Temporary Queues
For some protocols and APIs which only support monolithic "destinations"
without the address/queue separation (e.g. AMQP, JMS, etc.) temporary queues
are created by the broker using a UUID (i.e universally unique identifier) as
the name for both the address and the queue. Because the name is a UUID it is
impossible to create an `address-setting` for it whose `match` is anything but
`#`.
To solve this problem one can specify the `temporary-queue-namespace` in
`broker.xml` and then create an `address-setting` whose `match` value
corresponds to the configured `temporary-queue-namespace`. When the
`temporary-queue-namespace` is set and a temporary queue is created then the
broker will prepend the `temporary-queue-namespace` value along with the
`delimiter` value configured in `wildcard-addresses` (defaults to `.`) to the
address name and use that to lookup the associated `address-setting` values.
Here's a simple example configuration:
```xml
<temporary-queue-namespace>temp</temporary-queue-namespace>
<address-settings>
<address-setting match="temp.#">
<enable-metrics>false</enable-metrics>
</address-setting>
</address-settings>
```
Using this configuration any temporary queue will have metrics disabled.
> **Note:**
>
> This setting does *not* change the actual name of the temporary queue. It
> only changes the name used to *lookup* the address-settings.
### Other Advanced Configurations
Each of the following advanced configurations have their own chapter so their
details are not repeated here:
- [Exclusive queues](exclusive-queues.md)
- [Last Value queues](last-value-queues.md)
- [Non-Destructive queues](non-destructive-queues.md)
- [Ring queues](ring-queues.md)
- [Retroactive addresses](retroactive-addresses.md)
## How to filter messages
Apache ActiveMQ Artemis supports the ability to filter messages using [Filter
Expressions](filter-expressions.md).
Filters can be applied in two places - on a queue and on a consumer.
Filtering messages on a queue increases performance vs. filtering on the
consumer because the messages don't need to be scanned. However, a queue filter
is often not as flexible.
### Queue Filter
When a filter is applied to a queue, messages are filtered _before_ they are
routed to the queue. To add a filter use the `filter` element when configuring
a queue, e.g.:
```xml
<addresses>
<address name="filter">
<anycast>
<queue name="filter">
<filter string="color='red'"/>
</queue>
</anycast>
</address>
</addresses>
```
The filter defined above ensures that only messages with an attribute
`"color='red'"` is sent to this queue.
### Consumer Filters
Consumer filters are applied _after_ messages have routed to the queue and are
defined using the appropriate client APIs. The following JMS example shows how
consumer filters work.
Define an address with a single queue, with no filter applied in
`etc/broker.xml`.
```xml
<addresses>
<address name="filter">
<anycast>
<queue name="filter"/>
</anycast>
</address>
</addresses>
```
Then send some messages to the queue.
```java
...
// Send some messages
for (int i = 0; i < 3; i ++) {
TextMessage redMessage = senderSession.createTextMessage("Red");
redMessage.setStringProperty("color", "red");
producer.send(redMessage)
TextMessage greenMessage = senderSession.createTextMessage("Green");
greenMessage.setStringProperty("color", "green");
producer.send(greenMessage)
}
```
At this point the queue would have 6 messages: red, green, red, green, red,
green.
Create a consumer with the filter `color='red'`.
```java
MessageConsumer redConsumer = redSession.createConsumer(queue, "color='red'");
```
The `redConsumer` has a filter that only matches "red" messages. The
`redConsumer` will receive 3 messages.
```
red, red, red
```
The resulting queue would now be
```
green, green, green
```
## Alternate Ways to Determine Routing Type
Typically the routing type is determined either by the static XML configuration
or by the `default-address-routing-type` and `default-queue-routing-type`
`address-setting` elements used for [automatic address and queue
creation](#automatic-address-queue-management). However, there are two other
ways to specify routing type:
- a configurable prefix which client applications can use when sending
messages or creating consumers
- a property client applications can set on the messages they send
### Using a Prefix to Determine Routing Type
These prefixes are configured using the `anycastPrefix` and `multicastPrefix`
parameters within the URL of the `acceptor` which the client is using. When
multiple values are needed, these can be separated by a comma.
#### Configuring an Anycast Prefix
In `etc/broker.xml`, add the `anycastPrefix` to the URL of the desired
`acceptor`. In the example below, the acceptor is configured to use `queue/`
for the `anycastPrefix`. Client code can specify `queue/foo/` if the client
wants anycast routing.
```xml
<acceptor name="artemis">tcp://0.0.0.0:61616?protocols=AMQP;anycastPrefix=queue/</acceptor>
```
Consider, for example, a STOMP client that wants to send a message using
anycast semantics to a queue that doesn't exist. Consider also that the broker
is configured to auto-create addresses and queues, but the
`default-address-routing-type` and `default-queue-routing-type` are both
`MULTICAST`. Since the `anycastPrefix` is `queue/` it can just send a message
to `queue/foo` and the broker will automatically create an address named `foo`
with an anycast queue also named `foo`.
#### Configuring a Multicast Prefix
In `etc/broker.xml`, add the `multicastPrefix` to the URL of the desired
`acceptor`. In the example below, the acceptor is configured to use `topic/`
for the `multicastPrefix`. Client code can specify `topic/foo/` if the client
wants multicast routing.
```xml
<acceptor name="artemis">tcp://0.0.0.0:61616?protocols=AMQP;multicastPrefix=topic/</acceptor>
```
Consider, for example, a STOMP client that wants to create a subscription with
multicast semantics on an address that doesn't exist. Consider also that the
broker is configured to auto-create addresses and queues, but the
`default-address-routing-type` and `default-queue-routing-type` are both
`ANYCAST`. Since the `multicastPrefix` is `topic/` it can just subscribe to
`topic/foo` and the broker will automatically create an address named `foo`
with a multicast queue for the subscription. Any messages sent to `foo` will
then be routed to the subscription queue.
### Using a Message Property to Determine Routing Type
The `_AMQ_ROUTING_TYPE` property represents a `byte` value which will be used
by the broker to determine the routing type when a message is _sent_. Use `0`
for anycast routing or `1` for multicast routing.
> **Note:**
>
> A message will **only** be routed to queues which match its
> `_AMQ_ROUTING_TYPE` property value (if any). For example, if a message with
> an `_AMQ_ROUTING_TYPE` value of `1` (i.e. multicast) is sent to an address
> that only has anycast queues then the message won't actually be routed to any
> of the queues since the routing types don't match. If no `_AMQ_ROUTING_TYPE`
> is set then the message will be routed to all the queues on the address
> according to the queues' routing semantics.
## Configuring Addresses and Queues via Address Settings
This content has been relocated to its [own chapter](address-settings.md).
Reference in New Issue View Git Blame Copy Permalink