Cross-Platform AMQP Messaging in Java JMS, .NET, C++, and Python Apache Qpid is a reliable, asynchronous messaging system that supports the AMQP messaging protocol in several common programming languages. Qpid is supported on most common platforms. On the Java platform, Qpid uses the established Java JMS API. On the .NET platform, Qpid defines a WCF binding. For Python and C++, Qpid defines its own messaging API, the Qpid Messaging API, which is conceptually similar in each supported language. Support for this API in Ruby will be added soon (Ruby currently uses an API that is closely tied to the AMQP version). The Qpid Messaging API is quite simple, consisting of only a handful of core classes. A message consists of a standard set of fields (e.g. subject, reply-to), an application-defined set of properties, and message content (the main body of the message). A connection represents a network connection to a remote endpoint. A session provides a sequentially ordered context for sending and receiving messages. A session is obtained from a connection. A sender sends messages to a target using the sender.send method. A sender is obtained from a session for a given target address. A receiver receives messages from a source using the receiver.fetch method. A receiver is obtained from a session for a given source address. The following sections show how to use these classes in a simple messaging program.

The following C++ program shows how to create a connection, create a session, send messages using a sender, and receive messages using a receiver. #include #include #include #include #include using namespace qpid::messaging; int main(int argc, char** argv) { std::string broker = argc > 1 ? argv[1] : "localhost:5672"; std::string address = argc > 2 ? argv[2] : "amq.topic"; Connection connection(broker); try { connection.open(); Session session = connection.createSession(); Receiver receiver = session.createReceiver(address); Sender sender = session.createSender(address); sender.send(Message("Hello world!")); Message message = receiver.fetch(Duration::SECOND * 1); std::cout << message.getContent() << std::endl; session.acknowledge(); connection.close(); return 0; } catch(const std::exception& error) { std::cerr << error.what() << std::endl; connection.close(); return 1; } }]]>

The following Python program shows how to create a connection, create a session, send messages using a sender, and receive messages using a receiver.

An address is the name of a message target or message source. In the programs we have just seen, we used the address amq.topic (which is the name of an exchange on an AMQP 0-10 messaging broker). The methods that create senders and receivers require an address. The details of sending to a particular target or receiving from a particular source are then handled by the sender or receiver. A different target or source can be used simply by using a different address. An address resolves to a node. The Qpid Messaging API recognises two kinds of nodes, queues and topics The terms queue and topic here were chosen to align with their meaning in JMS. These two addressing 'patterns', queue and topic, are sometimes refered as point-to-point and publish-subscribe. AMQP 0-10 has an exchange type called a topic exchange. When the term topic occurs alone, it refers to a Messaging API topic, not the topic exchange.. A queue stores each message until it has been received and acknowledged, and only one receiver can receive a given message There are exceptions to this rule; for instance, a receiver can use browse mode, which leaves messages on the queue for other receivers to read. A topic immediately delivers a message to all eligible receivers; if there are no eligible receivers, it discards the message. In the AMQP 0-10 implementation of the API, The AMQP 0-10 implementation is the only one that currently exists. queues map to AMQP queues, and topics map to AMQP exchanges. In AMQP 0-10, messages are sent to exchanges, and read from queues. The Messaging API also allows a sender to send messages to a queue; internally, Qpid implements this by sending the message to the default exchange, with the name of the queue as the routing key. The Messaging API also allows a receiver to receive messages from a topic; internally, Qpid implements this by setting up a private subscription queue for the receiver and binding the subscription queue to the exchange that corresponds to the topic. In the rest of this tutorial, we present many examples using two programs that take an address as a command line parameter. spout sends messages to the target address, drain receives messages from the source address. The source code is available in both C++ and Python, and can be found in the examples directory for each language. These programs can use any address string as a source or a destination, and have many command line options to configure behavior—use the -h option for documentation on these options. Currently, the Python and C++ implementations of drain and spout have slightly different options. This tutorial uses the C++ implementation. The options will be reconciled in the near future. The examples in this tutorial also use the qpid-config utility to configure AMQP 0-10 queues and exchanges on a Qpid broker. Create a queue with qpid-config, send a message using spout, and read it using drain: $ qpid-config add queue hello-world $ ./spout hello-world $ ./drain hello-world Message(properties={spout-id:c877e622-d57b-4df2-bf3e-6014c68da0ea:0}, content='') The queue stored the message sent by spout and delivered it to drain when requested. Once the message has been delivered and and acknowledged by drain, it is no longer available on the queue. If we run drain one more time, no messages will be retrieved. $ ./drain hello-world $ This example is similar to the previous example, but it uses a topic instead of a queue. First, use qpid-config to remove the queue and create an exchange with the same name: $ qpid-config del queue hello-world $ qpid-config add exchange topic hello-world Now run drain and spout the same way we did in the previous example: $ ./spout hello-world $ ./drain hello-world $ Topics deliver messages immediately to any interested receiver, and do not store messages. Because there were no receivers at the time spout sent the message, it was simply discarded. When we ran drain, there were no messages to receive. Now let's run drain first, using the -t option to specify a timeout in seconds. While drain is waiting for messages, run spout in another window. First Window: $ ./drain -t 30 hello-word Second Window: $ ./spout hello-word Once spout has sent a message, return to the first window to see the output from drain: Message(properties={spout-id:7da2d27d-93e6-4803-8a61-536d87b8d93f:0}, content='') You can run drain in several separate windows; each creates a subscription for the exchange, and each receives all messages sent to the exchange.

So far, our examples have used address strings that contain only the name of a node. An address string can also contain a subject and options. The syntax for an address string is: [ / ] [ ; ] options ::= { : , ... } ]]> Addresses, subjects, and keys are strings. Values can be numbers, strings (with optional single or double quotes), maps, or lists. A complete BNF for address strings appears in . So far, the address strings in this tutorial have used only addresses. The following sections show how to use subjects and options.

Every message has a property called subject, which is analogous to the subject on an email message. If no subject is specified, the message's subject is null. For convenience, address strings also allow a subject. If a sender's address contains a subject, it is used as the default subject for the messages it sends. If a receiver's address contains a subject, it is used to select only messages that match the subject—the matching algorithm depends on the message source. In AMQP 0-10, each exchange type has its own matching algorithm, and queues do not provide filtering. This is discussed in . Currently, a receiver bound to a queue ignores subjects, receiving messages from the queue without filtering. In the future, if a receiver is bound to a queue, and its address contains a subject, the subject will be used as a selector to filter messages. In this example we show how subjects affect message flow. First, let's use qpid-config to create a topic exchange. $ qpid-config add exchange topic news-service Now we use drain to receive messages from news-service that match the subject sports. First Window: $ ./drain -t 30 news-service/sports In a second window, let's send messages to news-service using two different subjects: Second Window: $ ./spout news-service/sports $ ./spout news-service/news Now look at the first window, the message with the subject sports has been received, but not the message with the subject news: Message(properties={qpid.subject:sports, spout-id:9441674e-a157-4780-a78e-f7ccea998291:0}, content='') If you run drain in multiple windows using the same subject, all instances of drain receive the messages for that subject. The AMQP exchange type we are using here, amq.topic, can also do more sophisticated matching. A sender's subject can contain multiple words separated by a . delimiter. For instance, in a news application, the sender might use subjects like usa.news, usa.weather, europe.news, or europe.weather. The receiver's subject can include wildcard characters— # matches one or more words in the message's subject, * matches a single word. For instance, if the subject in the source address is *.news, it matches messages with the subject europe.news or usa.news; if it is europe.#, it matches messages with subjects like europe.news or europe.pseudo.news. This example uses drain and spout to demonstrate the use of subjects with two-word keys. Let's use drain with the subject *.news to listen for messages in which the second word of the key is news. First Window: $ ./drain -t 30 news-service/*.news Now let's send messages using several different two-word keys: Second Window: $ ./spout news-service/usa.news $ ./spout news-service/usa.sports $ ./spout news-service/europe.sports $ ./spout news-service/europe.news In the first window, the messages with news in the second word of the key have been received: Message(properties={qpid.subject:usa.news, spout-id:73fc8058-5af6-407c-9166-b49a9076097a:0}, content='') Message(properties={qpid.subject:europe.news, spout-id:f72815aa-7be4-4944-99fd-c64c9747a876:0}, content='') Next, let's use drain with the subject #.news to match any sequence of words that ends with news. First Window: $ ./drain -t 30 news-service/#.news In the second window, let's send messages using a variety of different multi-word keys: Second Window: $ ./spout news-service/news $ ./spout news-service/sports $ ./spout news-service/usa.news $ ./spout news-service/usa.sports $ ./spout news-service/usa.faux.news $ ./spout news-service/usa.faux.sports In the first window, messages with news in the last word of the key have been received: Message(properties={qpid.subject:news, spout-id:cbd42b0f-c87b-4088-8206-26d7627c9640:0}, content='') Message(properties={qpid.subject:usa.news, spout-id:234a78d7-daeb-4826-90e1-1c6540781eac:0}, content='') Message(properties={qpid.subject:usa.faux.news, spout-id:6029430a-cfcb-4700-8e9b-cbe4a81fca5f:0}, content='')

The options in an address string contain additional information for the senders or receivers created for it, including: Policies for assertions about the node to which an address refers. For instance, in the address string my-queue; {assert: always, node:{ type: queue }}, the node named my-queue must be a queue; if not, the address does not resolve to a node, and an exception is raised. Policies for automatically creating or deleting the node to which an address refers. For instance, in the address string xoxox ; {create: always}, the queue xoxox is created, if it does not exist, before the address is resolved. Extension points that can be used for sender/receiver configuration. For instance, if the address for a receiver is my-queue; {mode: browse}, the receiver works in browse mode, leaving messages on the queue so other receivers can receive them. Let's use some examples to show how these different kinds of address string options affect the behavior of senders and receives. First, let's use the assert option to ensure that the address resolves to a node of the required type. Let's use qpid-config to create a queue and a topic. $ qpid-config add queue my-queue $ qpid-config add exchange topic my-topic We can now use the address specified to drain to assert that it is of a particular type: $ ./drain 'my-queue; {assert: always, node:{ type: queue }}' $ ./drain 'my-queue; {assert: always, node:{ type: topic }}' 2010-04-20 17:30:46 warning Exception received from broker: not-found: not-found: Exchange not found: my-queue (../../src/qpid/broker/ExchangeRegistry.cpp:92) [caused by 2 \x07:\x01] Exchange my-queue does not exist The first attempt passed without error as my-queue is indeed a queue. The second attempt however failed; my-queue is not a topic. We can do the same thing for my-topic: $ ./drain 'my-topic; {assert: always, node:{ type: topic }}' $ ./drain 'my-topic; {assert: always, node:{ type: queue }}' 2010-04-20 17:31:01 warning Exception received from broker: not-found: not-found: Queue not found: my-topic (../../src/qpid/broker/SessionAdapter.cpp:754) [caused by 1 \x08:\x01] Queue my-topic does not exist Now let's use the create option to create the queue xoxox if it does not already exist: First Window: $ ./drain -t 30 "xoxox ; {create: always}" In previous examples, we created the queue before listening for messages on it. Using create: always, the queue is automatically created if it does not exist. Now we can send messages to this queue: Second Window: $ ./spout "xoxox ; {create: always}" Returning to the first window, we see that drain has received this message: Message(properties={spout-id:1a1a3842-1a8b-4f88-8940-b4096e615a7d:0}, content='') Other options specify message transfer semantics; for instance, they may state whether messages should be consumed or read in browsing mode, or specify reliability characteristics. The following example uses the browse option to receive messages without removing them from a queue. Let's use the browse mode to receive messages without removing them from the queue. First we send three messages to the queue: $ ./spout my-queue --content one $ ./spout my-queue --content two $ ./spout my-queue --content three Now we use drain to get those messages, using the browse option: $ ./drain 'my-queue; {mode: browse}' Message(properties={spout-id:fbb93f30-0e82-4b6d-8c1d-be60eb132530:0}, content='one') Message(properties={spout-id:ab9e7c31-19b0-4455-8976-34abe83edc5f:0}, content='two') Message(properties={spout-id:ea75d64d-ea37-47f9-96a9-d38e01c97925:0}, content='three') We can confirm the messages are still on the queue by repeating the drain: $ ./drain 'my-queue; {mode: browse}' Message(properties={spout-id:fbb93f30-0e82-4b6d-8c1d-be60eb132530:0}, content='one') Message(properties={spout-id:ab9e7c31-19b0-4455-8976-34abe83edc5f:0}, content='two') Message(properties={spout-id:ea75d64d-ea37-47f9-96a9-d38e01c97925:0}, content='three') option value semantics assert one of: always, never, sender or receiver Asserts that the properties specified in the node option match whatever the address resolves to. If they do not, resolution fails and an exception is raised. create one of: always, never, sender or receiver Creates the node to which an address refers if it does not exist. No error is raised if the node does exist. The details of the node may be specified in the node option. delete one of: always, never, sender or receiver Delete the node when the sender or receiver is closed. node A nested map containing the entries shown in . Specifies properties of the node to which the address refers. These are used in conjunction with the assert or create options. link A nested map containing the entries shown in . Used to control the establishment of a conceptual link from the client application to or from the target/source address. mode one of: browse, consume This option is only of relevance for source addresses that resolve to a queue. If browse is specified the messages delivered to the receiver are left on the queue rather than being removed. If consume is specified the normal behaviour applies; messages are removed from teh queue once the client acknoweldges their receipt. property value semantics type topic, queue durable True, False Indicates whether the node survives a loss of volatile storage e.g. if the broker is restarted. x-declare A nested map whose values correspond to the valid fields on an AMQP 0-10 queue-declare or exchange-declare command. These values are used to fine tune the creation or assertion process. Note however that they are protocol specific. x-bindings A nested list in which each binding is represented by a map. The entries of the map for a binding contain the fields that describe an AMQP 0-10 binding. Here is the format for x-bindings: , queue: , key: , arguments: { : , ..., : } }, ... ] ]]> In conjunction with the create option, each of these bindings is established as the address is resolved. In conjunction with the assert option, the existence of each of these bindings is verified during resolution. Again, these are protocol specific. option value semantics reliability one of: unreliable, at-least-once, at-most-once, exactly-once durable True, False Indicates whether the link survives a loss of volatile storage e.g. if the broker is restarted. x-declare A nested map whose values correspond to the valid fields of an AMQP 0-10 queue-declare command. These values can be used to customise the subscription queue in the case of receiving from an exchange. Note however that they are protocol specific. x-subscribe A nested map whose values correspond to the valid fields of an AMQP 0-10 message-subscribe command. These values can be used to customise the subscription. x-bindings A nested list each of whose entries is a map that may contain fields (queue, exchange, key and arguments) describing an AMQP 0-10 binding. These bindings are established during resolution independent of the create option. They are considered logically part of the linking process rather than of node creation.

This section provides a formal grammar for address strings. The following regular expressions define the tokens used to parse address strings: The formal grammar for addresses is given below: The address string options map supports the following parameters: [ / ] ; { create: always | sender | receiver | never, delete: always | sender | receiver | never, assert: always | sender | receiver | never, mode: browse | consume, node: { type: queue | topic, durable: True | False, x-declare: { ... ... }, x-bindings: [, ... ] }, link: { name: , durable: True | False, reliability: unreliable | at-most-once | at-least-once | exactly-once, x-declare: { ... ... }, x-bindings: [, ... ], x-subscribe: { ... ... } } } ]]> The create, delete, and assert policies specify who should perfom the associated action: always: the action is performed by any messaging client sender: the action is only performed by a sender receiver: the action is only performed by a receiver never: the action is never performed (this is the default) The node-type is one of: topic: in the AMQP 0-10 mapping, a topic node defaults to the topic exchange, x-declare may be used to specify other exchange types queue: this is the default node-type

The Qpidd broker and C++ clients can both use environment variables to enable logging. Use QPID_LOG_ENABLE to set the level of logging you are interested in (trace, debug, info, notice, warning, error, or critical): $ export QPID_LOG_ENABLE="warning+" The Qpidd broker and C++ clients use QPID_LOG_OUTPUT to determine where logging output should be sent. This is either a file name or the special values stderr, stdout, or syslog: export QPID_LOG_TO_FILE="/tmp/myclient.out"

The Python client library supports logging using the standard Python logging module. The easiest way to do logging is to use the basicConfig(), which reports all warnings and errors: from logging import basicConfig basicConfig() Qpidd also provides a convenience method that makes it easy to specify the level of logging desired. For instance, the following code enables logging at the DEBUG level: from qpid.log import enable, DEBUG enable("qpid.messaging.io", DEBUG) For more information on Python logging, see http://docs.python.org/lib/node425.html. For more information on Qpid logging, use $ pydoc qpid.log.

Connections in the Qpid Messaging API support automatic reconnect if a connection is lost. This is done using connection options. The following example shows how to use connection options in C++ and Python. In C++, these options are set using Connection::setOption(): In Python, these options are set using named arguments in the Connection constructor: See the reference documentation for details on how to set these on connections for each language. The following table lists the connection options that can be used. option value semantics reconnect True, False Transparently reconnect if the connection is lost. reconnect_timeout N Total number of seconds to continue reconnection attempts before giving up and raising an exception. reconnect_limit N Maximum number of reconnection attempts before giving up and raising an exception. reconnect_interval_min N Minimum number of seconds between reconnection attempts. The first reconnection attempt is made immediately; if that fails, the first reconnection delay is set to the value of reconnect_interval_min; if that attempt fails, the reconnect interval increases exponentially until a reconnection attempt succeeds or reconnect_interval_max is reached. reconnect_interval_max N Maximum reconnect interval. reconnect_interval N Sets both reconnection_interval_min and reconnection_interval_max to the same value.

A receiver can only read from one source, but many programs need to be able to read messages from many sources, preserving the original sequence of the messages. In the Qpid Messaging API, a program can ask a session for the next receiver; that is, the receiver that is responsible for the next available message. The following example shows how this is done in C++ and Python. C++: Python:

Request / Response applications use the reply-to property, described in , to allow a server to respond to the client that sent a message. A server sets up a service queue, with a name known to clients. A client creates a private queue for the server's response, creates a message for a request, sets the request's reply-to property to the address of the client's response queue, and sends the request to the service queue. The server sends the response to the address specified in the request's reply-to property. This example shows the C++ code for a client and server that use the request / response pattern. The server creates a service queue and waits for a message to arrive. If it receives a message, it sends a message back to the sender. The client creates a sender for the service queue, and also creates a response queue that is deleted when the client closes the receiver for the response queue. In the C++ client, if the address starts with the character #, it is given a unique name. " << response.getContent() << std::endl; ]]> The client sends the string ping to the server. The server sends the response pong back to the same client, using the replyTo property.

Many messaging applications need to exchange data across languages and platforms, using the native datatypes of each programming language. AMQP provides a set of portable datatypes, but does not directly support a set of named type/value pairs. Java JMS provides the MapMessage interface, which allows sets of named type/value pairs, but does not provide a set of portable datatypes. The Qpid Messaging API supports maps in message content. Unlike JMS, any message can contain maps. These maps are supported in each language using the conventions of the language. In Java, we implement the MapMessage interface; in Python, we support dict and list in message content; in C++, we provide the Variant::Map and Variant::List classes to represent maps and lists. In all languages, messages are encoded using AMQP's portable datatypes.

In Python, Qpid supports the dict and list types directly in message content. The following code shows how to send these structures in a message: The following table shows the datatypes that can be sent in a Python map message, and the corresponding datatypes that will be received by clients in Java or C++. Python Datatype → C++ → Java boolboolbooleanintegerintegerlongshortintegershortlonglonglongfloatfloatfloatfloatdoubledoublestringunicodejava.lang.Stringuuidqpid::types::Uuidjava.util.UUIDdictVariant::Mapjava.util.MaplistVariant::Listjava.util.List

In C++, Qpid defines the the Variant::Map and Variant::List types, which can be encoded into message content. The following code shows how to send these structures in a message: The following table shows the datatypes that can be sent in a C++ map message, and the corresponding datatypes that will be received by clients in Java and Python. C++ Datatype → Python → Java boolboolbooleanunsigned charstringcharunsigned short intintegershortunsigned intintegerintegerunsigned longintegerlongcharstringcharshortintegershortintintegerintegerlonglonglongfloatfloatfloatdouble, long doublefloatdoublestringunicodejava.lang.Stringqpid::types::Uuiduuidjava.util.UUIDVariant::Mapdictjava.util.MapVariant::Listlistjava.util.List

The XML Exchange is an AMQP 0-10 custom exchange provided by the Apache Qpid C++ broker. It allows messages to be filtered using XQuery; queries can address either message properties or XML content in the body of the message. An instance of the XML Exchange must be added before it can be used: $ qpid-config add exchange xml xml When using the XML exchange, a sender's address string must provide a subject, e.g. xml/weather. If a receiver that is using the XML exchange also provides a subject, it receives messages if the subject exactly matches a message's subject. When using the XML Exchange, a receiver normally provides an XQuery as an x-binding argument. If the query contains a context item (a path starting with .), then it is applied to the content of the message, which must be well-formed XML. For instance, ./weather is a valid XQuery, which matches any message in which the root element is named weather. Here is an address string that contains this query: Note that each x-binding is created in addition to any other bindings that may exist, and each x-binding must include the exchange name, the key, and the xquery. If you specify the subject in the address string (e.g. xml/weather; link ...), it creates a binding that is used in addition to the x-bindings; the binding created for the subject matches any message whose subject is weather, the binding created for the x-binding matches any message that satisfies the query, i.e. any message with a root element named weather. The XML Exchange can also be used to query message properties by declaring an external variable with the same name as each property that is to be queried. The following XQuery queries the control property, as well as the content of the message: If the XQuery addresses message content, and the message is not well-formed XML, the message will not be received. If the XQuery does not query message content, the message need not contain XML. The following shows the arguments used with drain to retrieve messages whose root element is named weather: In C++, it is convenient to place an XQuery in a string, and use a stringstream to add the query to the template for an address string that specifies an x-binding. 50" " and $w/temperature_f - $w/dewpoint > 5" " and $w/wind_speed_mph > 7" " and $w/wind_speed_mph < 20"; stringstream address; address << "xml; {" " link: { " " x-bindings: [{ exchange: xml, key: weather, arguments: { xquery:\"" << query << "\"} }] " " } " "}"; Receiver receiver = session.createReceiver(address.str()); Message response = receiver.fetch(); session.acknowledge(); std::cout << response.getContent() << std::endl; ]]> In Python, it is often convenient to place the query in a separate string, and use the repr() value of the query string in an address template string. 50 and $w/temperature_f - $w/dewpoint > 5 and $w/wind_speed_mph > 7 and $w/wind_speed_mph < 20 """ address = """ xml; { link: { x-bindings: [{ exchange: xml, key: weather, arguments: { xquery: %r} }] } } """ % query receiver = session.receiver(address) # Retrieve matching message from the receiver and print it message = receiver.fetch(timeout=1) print message.content session.acknowledge() ]]>

Clients can often be made significantly faster by batching acknowledgements and setting the capacity of receivers to allow prefetch.

Many of the simple examples we have shown retrieve a message and immediately acknowledge it. Because each acknowledgement results in network traffic, you can dramatically increase performance by acknowledging messages in batches. For instance, an application can read a number of related messages, then acknowledge the entire batch, or an application can acknowledge after a certain number of messages have been received or a certain time period has elapsed. Messages are not removed from the broker until they are acknowledged, so guaranteed delivery is still available when batching acknowledgements.

By default, a receiver retrieves the next message from the server, one message at a time, which provides intuitive results when writing and debugging programs, but does not provide optimum performance. To create an input buffer, set the capacity of the receiver to the size of the desired input buffer; for many applications, a capacity of 100 performs well. C++ Receiver receiver = session.createReceiver(address); receiver.setCapacity(100); Message message = receiver.fetch();

If a queue is durable, the queue survives a messaging broker crash, as well as any durable messages that have been placed on the queue. These messages will be delivered when the messaging broker is restarted. Delivery is guaranteed if and only if both the message and the queue are durable. Guaranteed delivery requires a persistence module, such as the one available from QpidComponents.org. C++:

In AMQP, transactions cover the semantics of enqueues and dequeues. When sending messages, a transaction tracks enqueues without actually delivering the messages, a commit places messages on their queues, and a rollback discards the enqueues. When receiving messages, a transaction tracks dequeues without actually removing acknowledged messages, a commit removes all acknowledged messages, and a rollback discards acknowledgements. A rollback does not release the message, it must be explicitly released to return it to the queue. C++:

This section describes the AMQP 0-10 mapping for the Qpid Messaging API. The interaction with the broker triggered by creating a sender or receiver depends on what the specified address resolves to. Where the node type is not specified in the address, the client queries the broker to determine whether it refers to a queue or an exchange. When sending to a queue, the queue's name is set as the routing key and the message is transfered to the default (or nameless) exchange. When sending to an exchange, the message is transfered to that exchange and the routing key is set to the message subject if one is specified. A default subject may be specified in the target address. The subject may also be set on each message individually to override the default if required. In each case any specified subject is also added as a qpid.subject entry in the application-headers field of the message-properties. When receiving from a queue, any subject in the source address is currently ignored. The client sends a message-subscribe request for the queue in question. The accept-mode is determined by the reliability option in the link properties; for unreliable links the accept-mode is none, for reliable links it is explicit. The default for a queue is reliable. The acquire-mode is determined by the value of the mode option. If the mode is set to browse the acquire mode is not-acquired, otherwise it is set to pre-acquired. The exclusive and arguments fields in the message-subscribe command can be controlled using the x-subscribe map. When receiving from an exchange, the client creates a subscription queue and binds that to the exchange. The subscription queue's arguments can be specified using the x-declare map within the link properties. The reliability option determines most of the other parameters. If the reliability is set to unreliable then an auto-deleted, exclusive queue is used meaning that if the client or connection fails messages may be lost. For exactly-once the queue is not set to be auto-deleted. The durability of the subscription queue is determined by the durable option in the link properties. The binding process depends on the type of the exchange the source address resolves to. For a topic exchange, if no subject is specified and no x-bindings are defined for the link, the subscription queue is bound using a wildcard matching any routing key (thus satisfying the expectation that any message sent to that address will be received from it). If a subject is specified in the source address however, it is used for the binding key (this means that the subject in the source address may be a binding pattern including wildcards). For a fanout exchange the binding key is irrelevant to matching. A receiver created from a source address that resolves to a fanout exchange receives all messages sent to that exchange regardless of any subject the source address may contain. An x-bindings element in the link properties should be used if there is any need to set the arguments to the bind. For a direct exchange, the subject is used as the binding key. If no subject is specified an empty string is used as the binding key. For a headers exchange, if no subject is specified the binding arguments simply contain an x-match entry and no other entries, causing all messages to match. If a subject is specified then the binding arguments contain an x-match entry set to all and an entry for qpid.subject whose value is the subject in the source address (this means the subject in the source address must match the message subject exactly). For more control the x-bindings element in the link properties must be used. For the XML exchange,Note that the XML exchange is not a standard AMQP exchange type. It is a Qpid extension and is currently only supported by the C++ broker. if a subject is specified it is used as the binding key and an XQuery is defined that matches any message with that value for qpid.subject. Again this means that only messages whose subject exactly match that specified in the source address are received. If no subject is specified then the empty string is used as the binding key with an xquery that will match any message (this means that only messages with an empty string as the routing key will be received). For more control the x-bindings element in the link properties must be used. A source address that resolves to the XML exchange must contain either a subject or an x-bindings element in the link properties as there is no way at present to receive any message regardless of routing key. If an x-bindings list is present in the link options a binding is created for each element within that list. Each element is a nested map that may contain values named queue, exchange, key or arguments. If the queue value is absent the queue name the address resolves to is implied. If the exchange value is absent the exchange name the address resolves to is implied. The following table shows how Qpid Messaging API message properties are mapped to AMQP 0-10 message properties and delivery properties. In this table msg refers to the Message class defined in the Qpid Messaging API, mp refers to an AMQP 0-10 message-properties struct, and dp refers to an AMQP 0-10 delivery-properties struct. Python API C++ API AMQP 0-10 PropertyIn these entries, mp refers to an AMQP message property, and dp refers to an AMQP delivery property. msg.idmsg.{get,set}MessageId()mp.message_id msg.subjectmsg.{get,set}Subject()mp.application_headers["qpid.subject"] msg.user_idmsg.{get,set}UserId()mp.user_id msg.reply_tomsg.{get,set}ReplyTo()mp.reply_toThe reply_to is converted from the protocol representation into an address. msg.correlation_idmsg.{get,set}CorrelationId()mp.correlation_id msg.durablemsg.{get,set}Durable()dp.delivery_mode == delivery_mode.persistentNote that msg.durable is a boolean, not an enum. msg.prioritymsg.{get,set}Priority()dp.priority msg.ttlmsg.{get,set}Ttl()dp.ttl msg.redeliveredmsg.{get,set}Redelivered()dp.redelivered msg.propertiesmsg.{get,set}Properties()mp.application_headers msg.content_typemsg.{get,set}ContentType()mp.content_type

The following program shows how to use address strings and JNDI for Qpid programs that use Java JMS. The Qpid JMS client uses Qpid Messaging API to identify sources and targets. This program uses a JNDI file that defines a connection factory for the broker we are using, and the address of the topic exchange node that we bind the sender and receiver to. (The syntax of a ConnectionURL is given in .) In the Java JMS code, we use create a JNDI context, use the context to find a connection factory and create and start a connection, create a session, and create a destination that corresponds to the topic exchange. Then we create a sender and a receiver, send a message with the sender, and receive it with the receiver. This code should be straightforward for anyone familiar with Java JMS.

Apache Qpid defines JNDI properties that can be used to specify JMS Connections and Destinations. Here is a typical JNDI properties file: The following sections describe the JNDI properties that Qpid uses.

Apache Qpid supports the properties shown in the following table: Property Purpose connectionfactory.<jndiname> The Connection URL that the connection factory uses to perform connections. queue.<jndiname> A JMS queue, which is implemented as an amq.direct exchange in Apache Qpid. topic.<jndiname> A JMS topic, which is implemented as an amq.topic exchange in Apache Qpid. destination.<jndiname> Can be used for defining all amq destinations, queues, topics and header matching, using an address string. Binding URLs, which were used in earlier versions of the Qpid Java JMS client, can still be used instead of address strings.

In JNDI properties, a Connection URL specifies properties for a connection. The format for a Connection URL is: amqp://[<user>:<pass>@][<clientid>]<virtualhost>[?<option>='<value>'[&<option>='<value>']] For instance, the following Connection URL specifies a user name, a password, a client ID, a virtual host ("test"), a broker list with a single broker, and a TCP host with the host name localhost using port 5672: amqp://username:password@clientid/test?brokerlist='tcp://localhost:5672' Apache Qpid supports the following properties in Connection URLs: Option Type Description brokerlist see below The broker to use for this connection. In the current release, precisely one broker must be specified. maxprefetch -- The maximum number of pre-fetched messages per destination. sync_publish {'persistent' | 'all'} A sync command is sent after every persistent message to guarantee that it has been received; if the value is 'persistent', this is done only for persistent messages. sync_ack Boolean A sync command is sent after every acknowledgement to guarantee that it has been received. use_legacy_map_msg_format Boolean If you are using JMS Map messages and deploying a new client with any JMS client older than 0.7 release, you must set this to true to ensure the older clients can understand the map message encoding. failover {'roundrobin' | 'failover_exchange'} If roundrobin is selected it will try each broker given in the broker list. If failover_exchange is selected it connects to the initial broker given in the broker URL and will receive membership updates via the failover exchange. Broker lists are specified using a URL in this format: brokerlist=<transport>://<host>[:<port>](?<param>=<value>)?(&<param>=<value>)* For instance, this is a typical broker list: brokerlist='tcp://localhost:5672' A broker list can contain more than one broker address; if so, the connection is made to the first broker in the list that is available. In general, it is better to use the failover exchange when using multiple brokers, since it allows applications to fail over if a broker goes down. A broker list can specify properties to be used when connecting to the broker, such as security options. This broker list specifies options for a Kerberos connection using GSSAPI: This broker list specifies SSL options: The following broker list options are supported. Option Type Description heartbeat integer frequency of heartbeat messages (in seconds) sasl_mechs -- For secure applications, we suggest CRAM-MD5, DIGEST-MD5, or GSSAPI. The ANONYMOUS method is not secure. The PLAIN method is secure only when used together with SSL. For Kerberos, sasl_mechs must be set to GSSAPI, sasl_protocol must be set to the principal for the qpidd broker, e.g. qpidd/, and sasl_server must be set to the host for the SASL server, e.g. sasl.com. SASL External is supported using SSL certification, e.g. ssl='true'&sasl_mechs='EXTERNAL' sasl_encryption Boolean If sasl_encryption='true', the JMS client attempts to negotiate a security layer with the broker using GSSAPI to encrypt the connection. Note that for this to happen, GSSAPI must be selected as the sasl_mech. ssl Boolean If ssl='true', the JMS client will encrypt the connection using SSL. tcp_nodelay Boolean If tcp_nodelay='true', TCP packet batching is disabled. sasl_protocol -- Used only for Kerberos. sasl_protocol must be set to the principal for the qpidd broker, e.g. qpidd/ sasl_server -- For Kerberos, sasl_mechs must be set to GSSAPI, sasl_server must be set to the host for the SASL server, e.g. sasl.com. trust_store -- path to Keberos trust store trust_store_password Kerberos trust store password key_store path to Kerberos key store key_store_password -- Kerberos key store password ssl_verify_hostname Boolean When using SSL you can enable hostname verification by using "ssl_verify_hostname=true" in the broker URL. ssl_cert_alias If multiple certificates are present in the keystore, the alias will be used to extract the correct certificate.

The following table shows how Qpid Messaging API message properties are mapped to AMQP 0-10 message properties and delivery properties. In this table msg refers to the Message class defined in the Qpid Messaging API, mp refers to an AMQP 0-10 message-properties struct, and dp refers to an AMQP 0-10 delivery-properties struct. Java JMS Message Property AMQP 0-10 PropertyIn these entries, mp refers to an AMQP message property, and dp refers to an AMQP delivery property. JMSMessageIDmp.message_id qpid.subjectThis is a custom JMS property, set automatically by the Java JMS client implementation.mp.application_headers["qpid.subject"] JMSXUserIDmp.user_id JMSReplyTomp.reply_toThe reply_to is converted from the protocol representation into an address. JMSCorrelationIDmp.correlation_id JMSDeliveryModedp.delivery_mode JMSPrioritydp.priority JMSExpirationdp.ttlJMSExpiration = dp.ttl + currentTime JMSRedelivereddp.redelivered JMS Propertiesmp.application_headers JMSTypemp.content_type

Qpid supports the Java JMS MapMessage interface, which provides support for maps in messages. The following code shows how to send a MapMessage in Java JMS. colors = new ArrayList(); colors.add("red"); colors.add("green"); colors.add("white"); m.setObject("colours", colors); Map dimensions = new HashMap(); dimensions.put("length",10.2); dimensions.put("width",5.1); dimensions.put("depth",2.0); m.setObject("dimensions",dimensions); List> parts = new ArrayList>(); parts.add(Arrays.asList(new Integer[] {1,2,5})); parts.add(Arrays.asList(new Integer[] {8,2,5})); m.setObject("parts", parts); Map specs = new HashMap(); specs.put("colours", colors); specs.put("dimensions", dimensions); specs.put("parts", parts); m.setObject("specs",specs); producer.send(m); ]]> The following table shows the datatypes that can be sent in a MapMessage, and the corresponding datatypes that will be received by clients in Python or C++. Java Datatype → Python → C++ booleanboolboolcharstringThe Python string will contain one Unicode characterint??? Java char is 16-bitshortintegershortintegerintegerintlonglong integerlongfloatfloatfloatdoublefloatdouble, long doublejava.lang.Stringunicodestd::stringjava.util.UUIDuuidqpid::types::Uuidjava.util.MapdictVariant::Mapjava.util.ListlistVariant::List

The AMQP Java JMS Messaging Client supports the following syntax for JMS selectors. )? // eg: 5.5 or 5. or 5.5E10 or 5.E10 | "." (["0"-"9"])+ ()? // eg: .5 or .5E10 | (["0"-"9"])+ // eg: 5E10 ) EXPONENT: "E" (["+","-"])? (["0"-"9"])+ STRING_LITERAL: "'" ( ("''") | ~["'"] )* "'" ]]> andExpression )* ) andExpression := ( equalityExpression ( equalityExpression )* ) equalityExpression := ( comparisonExpression ( "=" comparisonExpression | "<>" comparisonExpression | | )* ) comparisonExpression := ( addExpression ( ">" addExpression | ">=" addExpression | "<" addExpression | "<=" addExpression | stringLitteral ( stringLitteral )? | ( )? | addExpression addExpression | addExpression addExpression | "(" ( "," )* ")" | "(" ( "," )* ")" )* ) addExpression := multExpr ( ( "+" multExpr | "-" multExpr ) )* multExpr := unaryExpr ( "*" unaryExpr | "/" unaryExpr | "%" unaryExpr )* unaryExpr := ( "+" unaryExpr | "-" unaryExpr | unaryExpr | primaryExpr ) primaryExpr := ( literal | variable | "(" orExpression ")" ) literal := ( | | | | | | | ) variable := ( | )]]>