0
0
Fork 0
mirror of https://github.com/matrix-construct/construct synced 2024-10-31 10:58:54 +01:00
construct/include/ircd/server
2020-04-25 19:30:21 -07:00
..
link.h ircd::server: Add peer level aggregate for cleanup_canceled(). 2020-03-04 12:00:39 -08:00
peer.h ircd::server: Require service suffix for peer map keying; refactor host canonization. 2020-03-05 20:06:14 -08:00
README.md ircd::server: Add more blerb to README. 2019-02-11 10:13:38 -08:00
request.h ircd::server: Qualify noexcept interface functions. 2019-09-10 16:54:41 -07:00
server.h ircd::server: Additional convenience query for linked peers. 2020-04-25 19:30:21 -07:00
tag.h ircd::server::tag: Use explicit move assignment operations. 2019-09-14 16:55:57 -07:00

Interface To Remote Servers

This system manages connections and requests to remote servers when this server plays the role of client. This interface allows its user to make concurrent requests using the promise/future pattern. It is built on top of ircd::ctx::future, as well as many functions of ircd::net and also uses ircd::http.

API Overview

Developers will be concerned with the classes in request.h and the principal server::request.

A construction of the server::request class immediately launches a request to the remote target and processes the response in the background asynchronously. All connection establishment and management will be handled internally. The request object is somewhat "thin" and leaves buffer management for both sending and receiving to the user; buffers must stay valid for the duration of the request. The only internal buffer management is from a response-content dynamic-allocation feature which we'll detail later.

An instance is itself a ctx::future<http::code> and can be time-wait()'ed on and get()'ed; the latter returning a 20x HTTP code on success or throwing an exception on any failure. By default a failure is any network error and non-20x HTTP status codes. A request::opts option can be set to not throw exceptions for any HTTP code if that behavior is desired instead.

A request can be canceled at any time, and the preferred method is simply allowing the request to go out of scope and destruct. User buffers supplied to the request do not have to be maintained any further after cancellation.

Internal Overview

The ircd::server subsystem is fully asynchronous. It uses the callback pattern operating on the main stack without any blocking operations. It is not built on ircd::ctx to achieve maximum scale, servicing tens of thousands of requests concurrently. It is a relatively small subsystem with a high performance requirement providing a good motive to optimize this area specifically. Note that users of ircd::server are expected to be on an ircd::ctx which is required for the promise/future interface.

Structural

  • Peer: Represents a remote host (or srv-cluster, or matrix-origin) we want to communicate with. Instances of peers are held in a single large peers map keyed by host, and this map comprises the only runtime state of the ircd::server subsystem.

  • Link: A single TCP socket to a peer. There may be multiple links to a peer running concurrent operations.

  • Tag: A single HTTP request on a link to a peer. Tags will target a peer and be scheduled (and rescheduled) in the queue of the best link to a peer, or a new link may be opened if options allow.