As Jabber/XMPP technologies have matured, the need for a generic publish-subscribe ("pubsub") mechanism has arisen in a number of domains. These include (but are not limited to): news feeds and content syndication, extended presence, avatar management, shared bookmarks, auction and trading systems, online catalogs, workflow systems, network management systems, NNTP gateways, profile management, and event notification.

In all of these domains, it is desirable for data communication to follow the classic "publish-subscribe" or "observer" design pattern: a person or application publishes information, and an event notification (without or without payload) is broadcasted to all authorized subscribers. In general, the relationship between the publisher and subscriber is mediated by a service that receives publication requests, broadcasts event notifications to subscribers, and enables privileged entities to manage lists of people or applications that are authorized to publish or subscribe. In most pubsub services, the focal point for publication and subscription is a "topic" or "node" to which publishers send data and from which subscribers receive notifications. Additionally, some nodes may also maintain a history of events and provide other services that supplement the pure pubsub model.

This document defines a single, cohesive, generic protocol that all forms of pubsub can use. While compliant implementations are not required to implement all of the features defined herein, this document addresses most use cases that may be requested of a pubsub service. (For information about which features are required and which are recommended or optional, consult the Feature Summary.) Other specifications may define "subsets" or "profiles" of publish-subscribe for use in specialized contexts (e.g., &xep0163;), but such profiles are out of scope for this document.

This specification is large. However, the basic idea behind pubsub is rather simple (see Publish an Item to a Node):

1. An entity publishes information to a node at a publish-subscribe service.
2. The pubsub service pushes a notification to all entities that are authorized to learn about the published information.

Perhaps the most popular application of pubsub-like functionality is content syndication, which has become familiar from the RSS and Atom (&rfc4287;) feeds associated with weblogs, news sites, and other frequently-updated information available on the Internet. Consider the example of a weblog published by <hamlet@denmark.lit>. When Hamlet writes a new weblog entry, his blogging software publishes the entry to a pubsub node hosted at <pubsub.shakespeare.lit>:

So that is the "pub" part of publish-subscribe.

Now the pubsub service notifies all the subscribers about the new blog entry:

Here is an even simpler example: a transient node that sends only notifications without a payload:

Naturally, the entities involved may need to complete other use cases in order to enable full pubsub functionality -- for example, the publisher may need to create the node (see Create a Node) and subscribers may need to sign up for notifications (see Subscribe to a Node). These use cases are fully described in the remainder of this document. (For information about which features are required and which are recommended or optional, consult the Feature Summary.)

To manage permissions, the protocol defined herein uses a hierarchy of affiliations, similiar to those introduced in &xep0045;.

Support for the "owner" and "none" affiliations is REQUIRED. Support for all other affiliations is RECOMMENDED. Particular kinds of pubsub services MAY enforce additional requirements.

* Note: A service MAY allow any publisher to delete any item once it has been published to that node instead of allowing only the original publisher to remove it (this is discoverable via the "pubsub#delete-any" feature).

The ways in which an entity changes its affiliation with a node are well-defined. Typically, action by an owner is required to make an affiliation state transition. Affiliation changes and their triggering actions are specified in the following table.

Subscriptions to a node may exist in several states.

The requirements for the publish-subscribe protocol imply that there are two major dimensions along which we can measure events: persistent vs. transient, and pure notification vs. inclusion of payload. An implementation SHOULD enable an owner to configure a node along both of these dimensions.

No matter whether a node is configured for persistent or transient events, a service MAY cache the last item published to the node, in which case it SHOULD send that item to subscribers based on configuration of the "send_last_published_item" option (see the Item Caching section of this document); if the service supports the "http://jabber.org/protocol/pubsub#last-published" feature then the value of this option MUST default to "on_sub_and_presence" (though the service SHOULD allow the node owner to override the default).

A pubsub service MUST validate publish requests against the configuration of the node along both of these dimensions (see the Publish An Item to a Node section of this document for the relevant error conditions).

Whether an item must be provided by the publisher, and whether an item ID is provided by the publisher or generated by the pubsub service, depends on the type of event being published. We can summarize the relevant rules as follows:

There are two types of nodes:

In order to make node creation simpler for clients, we define the following node access models (in order of openness):

A generic publish-subscribe implementation SHOULD support all of the defined access models, although specialized publish-subscribe implementations MAY support only a subset of the access models. Which access models are provided in a particular deployment is a matter of service provisioning (e.g., some restricted deployments may wish to lock down permissions so that only the "authorize" and "whitelist" access models are provided, or even only the "whitelist" access model).

A node creator or owner can override the default access model by specifying an appropriate value for the 'pubsub#access_model' configuration field (see the Create a Node With Default Configuration and Configure a Node sections of this document).

If a pubsub node is addressable, it MUST be addressable either (1) as a JID or (2) as the combination of a JID and a node. These nodes are equivalent to those used in XEP-0030: Service Discovery.

A service MUST respond to service discovery information requests qualified by the 'http://jabber.org/protocol/disco#info' namespace. The "disco#info" result returned by a pubsub service MUST indicate the identity of the service and indicate which pubsub features are supported.

The possible pubsub features are noted throughout this document and have been registered as described in the XMPP Registrar Considerations section of this document. For information regarding which features are required, recommended, and optional, see the Feature Summary section of this document.

If a service implements a hierarchy of nodes (by means of Collection Nodes), it MUST also enable entities to discover the nodes in that hierarchy by means of the Service Discovery protocol, subject to the recommendations in XEP-0030 regarding large result sets (for which &xep0055; or some other protocol SHOULD be used). The following examples show the use of service discovery in discovering the nodes available at a hierarchical pubsub service.

Note: Node hierarchies and collection nodes are OPTIONAL. For details, refer to the NodeID Semantics and Collection Nodes sections of this document.

In the first example, an entity sends a service discovery items ("disco#items") request to the root node (i.e., the service itself), which is a Collection Node:

In the second example, an entity sends a disco#items request to one of the first-level nodes, which is also a collection node:

If a node is a leaf node rather than a collection node and items have been published to the node, the service MAY return one &ITEM; element for each published item as described in the Discover Items for a Node section of this document, however such items MUST NOT include a 'node' attribute (since they are published items, not nodes).

A pubsub service MUST allow entities to query individual nodes for the information associated with that node. The Service Discovery protocol MUST be used to discover this information. The "disco#info" result MUST include an identity with a category of "pubsub" and a type of either "leaf" or "collection".

Note: If a node has an identity type of "leaf", then it MUST NOT contain other nodes or collections (only items); if a node has an identity type of "collection", then it MUST NOT contain items (only other nodes or collections).

The "disco#info" result MAY include detailed meta-data about the node, encapsulated in the &xep0004; format as described in &xep0128;, where the data form context is specified by including a FORM_TYPE of "http://jabber.org/protocol/pubsub#meta-data" in accordance with &xep0068;. If meta-data is provided, it SHOULD include values for all configured options as well as "automatic" information such as the node creation date, a list of publishers, and the like.

Note: Node meta-data can be set in many ways. Some of it is based on node configuration (e.g., the owner's JID) whereas some of it may be dynamic (e.g., the number of subscribers). Any static information to be provided in the node meta-data SHOULD be provided as fields in the node configuration form.

Much of the meta-data provided about a node maps directly to selected &DUBLINCORE; meta-data attributes; specifically:

To discover the published items which exist on the service for a specific node, an entity MAY send a "disco#items" request to the node itself, and the service MAY return each item as a Service Discovery &ITEM; element. The 'name' attribute of each Service Discovery item MUST contain its ItemID and the item MUST NOT possess a 'node' attribute. This ItemID MAY then be used to retrieve the item using the protocol defined in the Retrieve Items from a Node section of this document.

An entity way want to query the service to retrieve its subscriptions for all nodes at the service. Support for this feature ("retrieve-subscriptions") is RECOMMENDED.

In order to make the request, the requesting entity MUST send an IQ-get whose &PUBSUB; child contains an empty <subscriptions/> element with no attributes.

If the service returns a list of subscriptions, it MUST return all subscriptions for all JIDs that match the bare JID (&BAREJID;) portion of the 'from' attribute on the request.

For each subscription, a <subscription/> element is returned specifying the NodeID, the JID that is affiliated (which MAY include a resource, depending on how the entity subscribed), and the current subscription state. If subscription identifiers are supported by the service, the 'subid' attribute MUST be present as well.

If the requesting entity has no subscriptions, the pubsub service MUST return an empty <subscriptions/> element.

If the service does not support subscriptions retrieval, the service MUST respond with a &feature; error, specifying a pubsub-specific error condition of <unsupported/> and a feature of "retrieve-subscriptions".

An entity may want to query the service to retrieve its affiliations for all nodes at the service. Support for this feature ("retrieve-affiliations") is RECOMMENDED.

In order to make the request, the requesting entity includes an empty <affiliations/> element with no attributes.

If the service returns a list of affiliations, it MUST return all affiliations for all JIDs that match the bare JID (&BAREJID;) portion of the 'from' attribute on the request.

For each affiliation, an <affiliation/> element is returned containing the NodeID and the affiliation state (owner, publisher, or outcast).

If the requesting entity has no affiliations, the pubsub service MUST return an empty <affiliations/> element.

If the service does not support affiliations retrieval, the service MUST respond with a &feature; error, specifying a pubsub-specific error condition of <unsupported/> and a feature of "retrieve-affiliations".

An implementation MAY allow subscribers to configure subscription options. Implementations SHOULD use the Data Forms protocol to accomplish this configuration (however, an out-of-band mechanism such as a web interface could be offered as well).

Implementations of pubsub that choose to persist items MAY allow entities to request existing items from a node (e.g., an entity may wish to do this after successfully subscribing in order to receive all the items in the publishing history for the node).

A publisher may want to delete an item once it has been published to a node that supports persistent items. Support for this feature ("delete-any") is RECOMMENDED.

After creating a new node, the node owner may want to modify the node configuration. Support for this feature is RECOMMENDED.

An entity way want to request information about the default node configuration, e.g. in order to determine whether to perform create-and-configure as previously described. Support for this feature is OPTIONAL.

If a service supports node creation, it MUST support node deletion. If an implementation persists items, it MUST remove all items from persistent storage before the node itself is deleted.

If a service persists published items, a node owner may want to purge the node of all published items (thus removing all items from the persistent store, with the exception of the last published item, which MAY be cached). It is OPTIONAL for a service to implement this feature.

A service MAY send subscription approval requests to the node owner(s) at any time. An approval request consists of a message stanza containing a Data Form scoped by the "http://jabber.org/protocol/pubsub#subscribe_authorization" FORM_TYPE. The form MUST contain a boolean field that has a 'var' attribute of "pubsub#allow", which is the field that designates whether or not to allow the subscription request. The form SHOULD include fields that specify the node identifier, the subscription id (if applicable), and the JID of the pending subscriber. The message MAY include a &BODY; element that contains natural-language text explaining that the message contains a pending subscription form.

In order to approve the request, the owner shall submit the form and set the "pubsub#allow" field to a value of "1" or "true"; for tracking purposes the message MUST reflect the 'id' attribute originally provided.

The service then SHOULD send notification to the approved subscriber (see the Notification of Subscription State Changes section of this document).

In order to deny the request, the owner shall submit the form and set the "pubsub#allow" field to a value of "0" or "false"; as above, the message MUST reflect the 'id' attribute originally provided.

The service then SHOULD send notification to the denied subscriber (see the Notification of Subscription State Changes section of this document).

In order to cancel the form submission, the owner shall reply with the form's 'type' attribute set to "cancel".

The service MUST check the "pubsub#allow" field to see if the subscription should be allowed or denied. If the owner cancels the Data Form, then the subscription request MUST remain in the pending state.

A node owner may want to request all of the pending subscription requests for all of their nodes at a service. It is OPTIONAL for a service to implement this feature.

This feature MUST be implemented using the &xep0050; protocol, where the command name ('node' attribute of the command element) MUST have a value of "http://jabber.org/protocol/pubsub#get-pending".

A node owner may want to edit the list of subscriptions associated with a given node. Support for this feature ("pubsub#manage-subscriptions") is OPTIONAL.

Note: This feature enables a node owner to set subscriptions for new entities; for nodes of type "whitelist", this enables the node owner to add subscribers.

A node owner may want to edit the affiliations of entities associated with a given node and to set affiliations for new entities. Support for this feature ("pubsub#modify-affiliations") is OPTIONAL.

A service that implements collection nodes SHOULD allow entities to subscribe to collection nodes (subject to access models and local security policies).

In addition to the subscription configuration options already defined, there are two subscription configuration options specific to collection nodes:

• pubsub#subscription_type

  This subscription option enables the subscriber to subscribe either to items or to nodes.

  If the subscription type is "items", the subscriber shall be notified whenever any node contained in the collection generates a notification (e.g., when an item is published or deleted), as modified by the value of the "pubsub#subscription_depth" option.

  If the subscription type is "nodes", the subscriber shall be notified whenever a new node is added to the collection, as modified by the value of the "pubsub#subscription_depth" option.

  The default value of this subscription option MUST be "nodes".

• pubsub#subscription_depth

  This subscription option enables the subscriber to specify whether it wants to receive notifications only from first-level children of the collection (a value of "1") or from all descendents (a value of "all").

  For subscriptions of type "items", this enables the subscriber to be informed only when an item is published to a leaf node that is a direct child of the collection node to which it has subscribed, or to be informed whenever an item is published to any leaf node in the "tree" that begins at the level of the collection to which it has subscribed.

  For subscriptions of type "nodes", this enables the subscriber to be informed only when a new node is added in the specific collection to which it has subscribed, or to be informed whenever a node is added anywhere in the "tree" that begins at the level of the collection to which it has subscribed.

  The default value of this subscription option MUST be "1".

In order to subscribe to a collection node, an entity MUST send a subscription request to the node; the subscription request MAY include subscription options, but this is not strictly necessary (especially if the entity does not wish to override the default settings for the "pubsub#subscription_type" and "pubsub#subscription_depth" options).

The subscriber will now receive notification of new first-level nodes created within the "blogs" collection.

The subscriber will now receive item notifications from nodes at any depth within the "blogs" collection.

Depending on the nature of the node "tree", a subscription type of "items" and depth of "all" may result in an extremely large number of notifications. Therefore, a service MAY disallow such a combination of subscription options, in which case it MUST return a ¬allowed; error to the requesting entity.

A service MAY allow an entity to subscribe to a collection node in two ways, once with a subscription of type "nodes" (to receive notification of any new nodes added to the collection or the entire tree) and once with a subscription of type "items" (to receive all items published within the tree). However, a service SHOULD NOT allow an entity to subscribe twice to a collection node (once with a subscription depth of "1" and once with a subscription depth of "all") for the same subscription type, since two such subscriptions are unnecessary (a depth of "all" includes by definition a depth of "1"); in this case the service SHOULD return a &conflict; error to the requesting entity.

A service that implements collections SHOULD support a root collection. The root collection shall be identified by the lack of a node identifier (i.e., the address of the pubsub service itself, such as "pubsub.shakespeare.lit").

Subscribing to this node with a subscription of type "nodes" and a depth of "1" enables an entity to be notified whenever a new first-level node is created at the pubsub service. Subscribing to this node with a subscription of type "nodes" and a depth of "all" enables an entity to be notified whenever a new node is created anywhere at the pubsub service.

In order to send notification of new nodes, the service shall send an event that contains a <collection/> element that in turns contains a <node/> element whose 'id' attribute specifies the NodeID of the new node.

The notification event MAY also include the node meta-data, formatted using the Data Forms protocol.

To create a new collection node, the requesting entity MUST include a Data Form containing a 'pubsub#node_type' field whose <value/> is "collection".

In addition to the errors already defined for leaf node creation, there are several reasons why the collection node creation request might fail:

1. The service does not support collection nodes.
2. The service does not support creation of collection nodes.
3. The requesting entity does not have sufficient privileges to create collection nodes.

These error cases are described more fully in the following sections.

If the service does not support collection nodes, it MUST respond with a &feature; error, specifying a pubsub-specific error condition of <unsupported/> and a feature of "collections".

If the service supports collection nodes but does not allow new collection nodes to be created, it MUST respond with a ¬allowed; error.

If the requesting entity has insufficient privileges to create new collections, the service MUST respond with a &forbidden; error.

A service MAY offer some node configuration options that are specific to collection nodes and not provided in configuration forms related to leaf nodes. The following are RECOMMENDED:

• pubsub#children_association_policy -- the policy regarding who may associate child nodes with the collection (values: all, owner, whitelist).
• pubsub#children_association_whitelist -- the whitelist of entities that may associate child nodes with the collection.
• pubsub#children_max -- the maximum number of child nodes that may be associated with a collection.

Although a node can be associated with a collection when it is created (as described above), it can also be associated with a collection after it has been created. This can be done in two ways:

• By modifying the node's "pubsub#collection" configuration field.
• By modifying the collection node's "pubsub#children" configuration field.

These methods are described below.

In order to modify the (child) node's "pubsub#collection" configuration field, the owner of the node shall submit a request to edit the node's configuration, receive a configuration form from the service, and then submit a modified configuration form:

In order to modify the (parent) node's "pubsub#children" configuration field, the owner of the node shall submit a request to edit the node's configuration, receive a configuration form from the service, and then submit a modified configuration form:

A node can be disassociated from a collection after it has been associated (whether at creation time or afterward). This can be done in two ways:

• By modifying the node's "pubsub#collection" configuration field.
• By modifying the collection node's "pubsub#children" configuration field.

These methods are described below.

In order to modify the (child) node's "pubsub#collection" configuration field, the owner of the node shall submit a request to edit the node's configuration, receive a configuration form from the service, and then submit a modified configuration form:

Note: To disassociate the node from all collection nodes, the node owner MUST submit an empty <value/> element within the 'pubsub#collection' field as shown in the foregoing example.

In order to modify the (parent) node's "pubsub#children" configuration field, the owner of the node shall submit a request to edit the node's configuration, receive a configuration form from the service, and then submit a modified configuration form:

If a node is disassociated from a collection node and a new association is not formed, the implementation MAY associate the node with the root collection node or associate it with no collection node.

Note: The combination of associating and disassociating a node with a collection can be used to move a node from one collection to another.

If an item is published to a node which is also included by a collection, and an entity is subscribed to that collection with a subscription type of "items", then the notifications generated by the service MUST contain additional information. The &ITEMS; element contained in the notification message MUST specify the node identifier of the node that generated the notification (not the collection) and the &ITEM; element MUST contain a SHIM header that specifies the node identifier of the collection.

When a contact is affiliated with the account owner through sharing of XMPP presence, the "auto-subscribe" feature greatly simplifies the subscription process. In particular, support for the "auto-subscribe" has the following implications:

A contact may not want to receive notifications for all payload types. A contact SHOULD signal its preferences to the account owner's server by including XEP-0115 information that specifies the namespaces for which the contact wishes to receive notifications (if any). This information is used by a pubsub service that supports the "filtered-notifications" feature to send only those payload types that the subscriber wishes to receive.

In order to make this possible, all possible payload namespaces can be appended with the string "+notify" to indicate that the contact wishes to receive notifications for the payload format. Thus if Romeo wants to receive notifications for location data and tune data but not activity data, his client would advertise support for the following namespaces in the disco#info results it sends: Including, say, the 'http://jabber.org/protocol/geoloc' namespace indicates that the client understands the geolocation namespace, whereas including the 'http://jabber.org/protocol/geoloc+notify' namespace indicates that the client wishes to receive notifications related to geolocation.

• http://jabber.org/protocol/geoloc+notify
• http://jabber.org/protocol/tune+notify

This set of namespaces would then be advertised by including the namespace in the identity+features hash encapsulated via the 'ver' attribute as described in XEP-0115.

It is the responsibility of the account owner's server to cache XEP-0115 information. When the server receives presence from a contact, it MUST check that presence information for entity capabilities data and correlate that data with the desired namespaces for the contact's client. The server MUST NOT send notifications related to any data formats that the contact's client has not asked for via the relevant "namespace+notify" disco#info feature. This enables a client to turn off all notifications (e.g., because of bandwidth restrictions) and to easily receive all desired data formats simply by adding support for the appropriate "namespace+notify" combination in its disco#info results and client capabililies. However, it also implies that a client can request notifications only on a global basis and cannot request, say, mood information only from certain contacts in the user's roster. Community consensus is that this is an acceptable tradeoff. Also, note that this works only if the account owner has a presence subscription to the contact and the contact has a presence subscription to the account owner.

Some examples may help to illustrate the concept of notification filtering. Here we show presence generated by two of the contacts listed above (benvolio@montague.lit does not have any presence subscriptions to or from juliet@capulet.lit and therefore is not involved in these protocol flows).

We assume that Juliet's server doesn't know anything about these capabilities, so it sends service discovery information requests to each of the clients on Juliet's behalf (realistically, the capulet.lit server will quickly build up a cache of client capabilities, with the result that it will not need to send these service discovery requests):

The server shall also query the identity+features for <romeo@montague.lit>:

(As noted in XEP-0115, the server MUST check the hash provided in the 'ver' attribute against the generation method to ensure that no poisoning has occurred.)

Now we revisit account owner publication and server generation of notifications, with filtering enabled because the server has caps information:

• If Juliet publishes a tune item to the presence-access "http://jabber.org/protocol/tune" node, her server will send notifications to <nurse@capulet.lit/chamber> and <romeo@montague.lit/orchard> (full JIDs).

• If Juliet publishes an activity item to the presence-access "http://jabber.org/protocol/activity" node, her server will send notifications only to <nurse@capulet.lit/chamber>.

• If Juliet publishes a geolocation item to the roster-access "http://jabber.org/protocol/geoloc" node, her server will send notifications only to <romeo@montague.lit/orchard>.

When a pubsub service generates notifications, it MUST adhere to the delivery rules implicit in the subscription option configuration for each subscriber. In particular, the 'to' address SHOULD be that of the subscribed JID only. The service SHOULD NOT attempt to guess at the most available resource associated with the subscribed JID (e.g., in the context of instant messaging systems).

As noted above, a pubsub service SHOULD ensure that the &MESSAGE; stanza for each event notification it generates possesses an 'id' attribute with a unique value. (This notification ID is not to be confused with either the node ID or the item ID.) This ID MUST be unique within the context of the pubsub service in order to ensure proper tracking of any delivery-related errors.

Exactly how a service shall handle delivery-related errors is a matter of implementation. In general, such handling is effectively similar to the bounce processing performed by other message delivery systems, such as mail transfer agents and mailing list software. The following are some suggested guidelines regarding the handling of XMPP-specific error conditions in relation to pubsub event notifications (see RFC 3920 and XEP-0086 regarding XMPP error condition semantics):

• If the XMPP error is of type "cancel" (e.g., ¬found;), or the error condition is &gone;, the pubsub service SHOULD terminate the subscription of the entity to that node and MAY terminate the subscription of that entity to all nodes hosted at the service.
• If the XMPP error is of type "auth" (e.g., ®istration;) or "wait" (e.g., &timeout;), or the error condition is &badrequest;, &redirect;, or ¬acceptable;, the pubsub service SHOULD increment a bounce counter for that entity and MAY attempt to resend the event notification after some configurable amount of time. The service MAY terminate the subscription of the entity to that node if the bounce counter has reached some configurable limit.

Implementations of pubsub MAY deliver event notifications only when the subscriber is online. In these cases, the option may be a node configuration option as shown in the examples above. To facilitate this, the pubsub service needs to subscribe to the subscriber's presence and check the subscriber's current presence information before sending any event notifications (as described in RFC 3921). Presence subscriptions MUST be based on the subscribed JID.

Sending events to users of existing Jabber servers may force event notifications to be routed to offline storage for later delivery (as described in &xep0160;). This may not always be desirable. The possible ways of preventing this behavior include:

• Use presence-based subscription options as described above.
• Use delivery semantics as defined by &xep0079;.
• Specify a message type of "headline", which in most existing server implementations will prevent offline storage.

If a service understands the semantics for a particular payload type and an entity's subscription is so configured (via the "pubsub#include_body" option), the service SHOULD include an appropriate XMPP &BODY; child element along with the payloads it sends in event notifications for a given node, where the body's XML character data summarizes or represents the information contained in the payload (this enables clients that do not understand the payload format to present the appropriate information to an end user). For example, the Atom <summary/> element (see RFC 4287) could be mapped to the XMPP &BODY; element. A service MUST NOT provide the "pubsub#include_body" subscription option for a node if it does not have a defined way to transform part or all of the payload format into a sensible message body. A node owner MAY define an XSLT for transforming the payload format into a message body, via the "pubsub#body_xslt" node configuration option.

If the service does not understand the semantics for a particular payload type, it MAY include a <body/> child whose XML character data informs the subscriber that the message contains an event notification (e.g., text such as "This message contains an event notification" would be appropriate).

NodeIDs MUST be treated as unique identifiers within the context of a particular pubsub service.

If item identifiers are used, they MUST be treated as unique within the scope of the node. The combination of the NodeID + ItemID MUST be unique within a given service, and MUST specify a single published item at a single node.

If a publisher publishes an item and the ItemID matches that of an existing item, the pubsub service MUST overwrite the existing item and generate a new event notification.

Because it is possible for a node's configuration to change such that ItemIDs are required (e.g., a change from transient to persistent), a service SHOULD use ItemIDs for internal tracking purposes even if it does not include them with the notifications it generates prior to the configuration change.

A service MAY cache the last item published to a node, even if the node is configured for transient publication (i.e., configured to not persist items). The last published item SHOULD be sent to new subscribers upon successful processing of a subscription request or approval by a node owner.

Note: If a publisher requests Batch Processing of item publications, the concept of "last published item" is undefined; therefore, if information coherence is needed, a publisher SHOULD publish items in separate requests rather than in batch mode.

Note: Particular profiles of the generic publish-subscribe protocol MAY define more stringent requirements regarding the "cache-last-item" feature.

A publisher MAY include multiple &ITEM; elements in a publish request and MAY include multiple &ITEM; elements in a retract request. This results in "batch processing" of publications or retractions. If the service cannot process any one of the items to be published or retracted, the entire batch MUST fail. Also note that batch publication renders the concept of "last published item" problematic; therefore, if information coherence is needed, a publisher SHOULD publish items in separate requests rather than in batch mode.

A service MUST allow owners and publishers to subscribe to a node, and to retrieve items from a node even if they are not subscribed. A service MAY auto-subscribe owners and publishers if they are not already subscribed, in which case it SHOULD generate a subscription ID if necessary for the subscription and SHOULD send a notification of successful subscription as described in the Notification of Subscription State Changes section of this document.

How subscription requests are sent to node owners is a matter of implementation. Possibilities include:

• Send requests to all owners (these may be placed in offline storage as described in XEP-0160) and first approval wins.
• The service could subscribe to owner presence, and send only to the owners that are online.
• All owners vote on the new subscriber.
• Any owner is allowed to veto the subscriber.

An implementation MAY use any of these methods, or some other method not defined herein.

Various actions and events may result in changes to a subscription state:

• Approval or denial of a subscription request as described in the Manage Subscription Requests use case

• Cancellation of an existing subscription, for which many "triggers" are possible:

  • The entity simply unsubscribes from the node
  • The node is of type "presence" and the underlying presence subscription is cancelled
  • The node is of type "roster" and the entity is moved to an unauthorized roster group

When a subscription state change occurs, a service SHOULD send a message to the (new, former, or denied) subscriber informing it of the change, where the message contains an <event/> element with a single <subscription/> child that specifies the node, JID, and subscription state. The notification MAY contain a &BODY; element specifying natural-language text regarding the subscription change. Examples are shown below.

If the service has knowledge of the (former or denied) subscriber's presence, it SHOULD send the message to all of the subscriber's resources; if not, it MUST send the message to the subscriber's affiliated JID.

If a service or node supports this feature, it MUST return a feature of "subscription-notifications" in its response to service discovery information requests.

NodeIDs MAY have semantic meaning in particular profiles, implementations, or deployments of pubsub. However, it is STRONGLY RECOMMENDED that such semantic meaning not be used to encapsulate the hierarchical structure of nodes; instead, node hierarchy SHOULD be encapsulated using collections and their associated child nodes.

A user may publish information that adheres to a certain profile at multiple pubsub nodes, and a potential subscriber may want to discover all of these pubsub nodes. The user may make a list of pubsub nodes publicly available for querying even when the user is offline by using the Service Discovery mechanism for "publishing" items (see Section 4 of XEP-0030). The potential subscriber may then send a "disco#items" request to the user's bare JID (<user@host>), including the 'node' attribute set to the value of the namespace to which the desired information adheres. The user's server then returns a list of pubsub nodes that meet that criterion (with each pubsub node being a separate Service Discovery item). Here is an example.

Alternatively, a user may be registered with a server that offers personal eventing services, in which case the user will have one node per namespace as described in XEP-0163.

When SubIDs are used, Stanza Headers and Internet Metadata (SHIM) headers are to be included in order to differentiate notifications sent regarding a particular subscription. The relevant use cases and scenarios are:

• Sending notifications regarding newly-published items as described in the Publish an Item to a Node use case.
• Sending notifications regarding deleted items as described in the Delete an Item from a Node use case.

The SHIM headers are generated by the node to which the subscriber has a subscription, which may be either a leaf node or a collection node.

SHIM headers are not to be included when the content does not differ based on subscription ID, e.g., when a node sends notification of a configuration change to the node itself, notification that the node has been purged, or notification that the node has been deleted.

An implementation MAY enable the node configuration to specify an association between the event notification and the entity to which the published information pertains, but such a feature is OPTIONAL. Here are some possible examples:

• In the context of a geolocation notification service using &xep0080;, the user may generate the geolocation information or the information may be generated by an automated service (e.g., a service offered by a mobile telephony provider), but in either case the information is about the user's geolocation and therefore all replies should go to the user (who is probably the node owner).
• In the context of a group weblog, different users might publish to the weblog and replies might go to the publisher of an entry rather than to the weblog owner.
• In the context of an integrated pubsub and multi-user chat system, the node owner might be the room owner but all replies need to be sent to the room rather than to the owner.

Therefore we define three node configuration options:

• pubsub#itemreply -- the per-item policy for replies (the value is "owner" or "publisher").
• pubsub#replyto -- the specific user JID(s) to which replies should be sent.
• pubsub#replyroom -- the multi-user chat room to which replies should be sent.

A node owner MUST NOT define more than one of these options.

An example is shown below.