class Async::HTTP::Protocol::HTTP1::Server
def closed(error = nil)
def closed(error = nil) super @ready.signal end
def each(task: Task.current)
def each(task: Task.current) task.annotate("Reading #{self.version} requests for #{self.class}.") while request = next_request if body = request.body finishable = Finishable.new(body) request.body = finishable end response = yield(request, self) version = request.version body = response&.body if hijacked? body&.close return end task.defer_stop do # If a response was generated, send it: if response trailer = response.headers.trailer! # Some operations in this method are long running, that is, it's expected that `body.call(stream)` could literally run indefinitely. In order to facilitate garbage collection, we want to nullify as many local variables before calling the streaming body. This ensures that the garbage collection can clean up as much state as possible during the long running operation, so we don't retain objects that are no longer needed. if body and protocol = response.protocol # We force a 101 response if the protocol is upgraded - HTTP/2 CONNECT will return 200 for success, but this won't be understood by HTTP/1 clients: write_response(@version, 101, response.headers) # At this point, the request body is hijacked, so we don't want to call #finish below. request = nil response = nil if body.stream? return body.call(write_upgrade_body(protocol)) else write_upgrade_body(protocol, body) end elsif response.status == 101 # This code path is to support legacy behavior where the response status is set to 101, but the protocol is not upgraded. This may not be a valid use case, but it is supported for compatibility. We expect the response headers to contain the `upgrade` header. write_response(@version, response.status, response.headers) # Same as above: request = nil response = nil if body.stream? return body.call(write_tunnel_body(version)) else write_tunnel_body(version, body) end else write_response(@version, response.status, response.headers) if request.connect? and response.success? # Same as above: request = nil response = nil if body.stream? return body.call(write_tunnel_body(version)) else write_tunnel_body(version, body) end else head = request.head? # Same as above: request = nil response = nil write_body(version, body, head, trailer) end end # We are done with the body: body = nil else # If the request failed to generate a response, it was an internal server error: write_response(@version, 500, {}) write_body(version, nil) request&.finish end if finishable finishable.wait(@persistent) else # Do not remove this line or you will unleash the gods of concurrency hell. task.yield end rescue => error raise ensure body&.close(error) end end end
def fail_request(status)
def fail_request(status) @persistent = false write_response(@version, status, {}) write_body(@version, nil) rescue => error # At this point, there is very little we can do to recover: Console::Event::Failure.for(error).emit(self, "Failed to write failure response!", severity: :debug) end
def initialize(...)
def initialize(...) super @ready = Async::Notification.new end
def next_request
def next_request if closed? return nil elsif !idle? @ready.wait end # Read an incoming request: return unless request = Request.read(self) unless persistent?(request.version, request.method, request.headers) @persistent = false end return request rescue ::Protocol::HTTP1::BadRequest => error fail_request(400) # Conceivably we could retry here, but we don't really know how bad the error is, so it's better to just fail: raise end