|This documentation is for the TIP (development tree) of NNG and may represent unreleased changes or functionality that is experimental, and is subject to change before release. The latest released version is v1.3.2. See the documentation for v1.3.2 for the most up-to-date information.|
nng - nanomsg next generation
cc ['flags'] 'files' -lnng ['libraries']
NNG provides a common messaging framework intended to solve common communication problems in distributed applications. It offers a number of protocols, and also a number of transports.
The protocols implement the semantics associated with particular communications scenarios, such as RPC style services, service discovery, publish/subscribe, and so forth.
The transports provide support for underlying transport methods, such as TCP, IPC, websockets, and so forth.
NNG is designed to permit easy creation of new transports and, to a lesser extent, new protocols.
NNG is wire compatible with the SP protocols described in the nanomsg project; projects using libnanomsg can inter-operate with nng as well as other conforming implementations. (One such implementation is mangos.) Applications using NNG which wish to communicate with other libraries must ensure that they only use protocols or transports offered by the other library.
NNG also offers a compatible API, permitting legacy code to be recompiled or relinked against NNG. When doing this, support for certain enhancements or features will likely be absent, requiring the application developer to use the new-style API.
NNG is implemented in pure C; if you need bindings for other languages please check the website.
Publisher side of publish/subscribe protocol
Pull side of pipeline protocol
Push side of pipeline protocol
Subscriber side of publish/subscribe protocol
Reply side of request/reply protocol
Request side of request/reply protocol
Respondent side of survey protocol
Surveyor side of survey protocol
NNG presents a socket view of networking. The sockets are constructed using protocol-specific functions, as a given socket implements precisely one protocol.
Each socket can be used to send and receive messages (if the protocol) supports it, and implements the appropriate protocol semantics. For example, sub sockets automatically filter incoming messages to discard those for topics that have not been subscribed.
NNG sockets are message oriented, so that messages are either delivered wholly, or not at all. Partial delivery is not possible. Furthermore, NNG does not provide any other delivery or ordering guarantees; messages may be dropped or reordered (Some protocols, such as req may offer stronger guarantees by performing their own retry and validation schemes.)
Each socket can have zero, one, or many endpoints, which are either listeners or dialers. (A given socket may freely choose whether it uses listeners, dialers, or both.) These endpoints provide access to underlying transports, such as TCP, etc.
Each endpoint is associated with a URL, which is a service address. For dialers, this will be the service address that will be contacted, whereas for listeners this is where the listener will accept new connections.
Endpoints do not themselves transport data. They are instead responsible for the creation of pipes, which can be thought of as message-oriented connected streams. Pipes frequently correspond to a single underlying byte stream. For example both IPC and TCP transports implement their pipes using a 1:1 relationship with a connected operating system socket.
Endpoints create pipes as needed. Listeners will create them when a new client connection request arrives, and dialers will generally create one, then wait for it to disconnect before reconnecting.
Most applications should not have to worry about endpoints or pipes at all; the socket abstraction should provide all the functionality needed other than in a few specific circumstances.
Most applications will use sockets in normal, or cooked, mode. This mode provides the full semantics of the protocol. For example, req sockets will automatically match a reply to a request, and resend requests periodically if no reply was received.
There are situations, such as with proxies, where it is desirable to bypass these semantics and simply pass messages to and from the socket with no extra semantic handling. This is possible using raw mode sockets.
Raw mode sockets are generally constructed with a different function,
Using these sockets, the application can simply send and receive messages,
and is responsible for supplying any additional socket semantics.
Typically this means that the application will need to inspect message
headers on incoming messages, and supply them on outgoing messages.
NNG uses universal resource locators (URLs) following the format specified in RFC 3986, including some schemes that are unique to SP. The URLs used in NNG are canonicalized as follows, mostly in accordance with RFC 3986 6.2.2:
The URL is parsed into scheme, userinfo, host, port, path, query and fragment components. (Not all of these members are necessarily present.)
The scheme, hostname, and port if present, are converted to lower case.
Percent-encoded values for unreserved characters converted to their unencoded forms.
Additionally URL percent-encoded values for characters in the path and with numeric values larger than 127 (i.e. not ASCII) are decoded.
The resulting path is checked for invalid UTF-8 sequences, consisting of surrogate pairs, illegal byte sequences, or overlong encodings. If this check fails, then the entire URL is considered invalid.
Path segments consisting of
..are resolved as per RFC 3986 184.108.40.206.
Further, empty path segments are removed, meaning that duplicate slash (
/) separators are removed from the path.
Note that steps 4, 5, and 7 are not specified by RFC 3986, but performing them is believed to improve both the usability and security of applications, without violating RFC 3986 itself.
|Port numbers may be service names in some instances, but it is recommended that numeric port numbers be used when known. If service names are used, it is recommended that they follow the naming conventions for C identifiers, and not be longer than 32 characters in length. This will maximize compatibility across systems and minimize opportunities for confusion when they are parsed on different systems.|
The library API is documented at libnng(3).