Official Go implementation of the Ethereum protocol
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 
go-ethereum/beacon/light/request/server_test.go

158 lines
4.9 KiB

package request
import (
"testing"
"github.com/ethereum/go-ethereum/common/mclock"
)
const (
testRequest = "Life, the Universe, and Everything"
testResponse = 42
)
var testEventType = &EventType{Name: "testEvent"}
func TestServerEvents(t *testing.T) {
rs := &testRequestServer{}
clock := &mclock.Simulated{}
srv := NewServer(rs, clock)
var lastEventType *EventType
srv.subscribe(func(event Event) { lastEventType = event.Type })
evTypeName := func(evType *EventType) string {
if evType == nil {
return "none"
}
return evType.Name
}
expEvent := func(expType *EventType) {
if lastEventType != expType {
t.Errorf("Wrong event type (expected %s, got %s)", evTypeName(expType), evTypeName(lastEventType))
}
lastEventType = nil
}
// user events should simply be passed through
rs.eventCb(Event{Type: testEventType})
expEvent(testEventType)
// send request, soft timeout, then valid response
srv.sendRequest(testRequest)
clock.WaitForTimers(1)
clock.Run(softRequestTimeout)
expEvent(EvTimeout)
rs.eventCb(Event{Type: EvResponse, Data: RequestResponse{ID: 1, Request: testRequest, Response: testResponse}})
expEvent(EvResponse)
// send request, hard timeout (response after hard timeout should be ignored)
srv.sendRequest(testRequest)
clock.WaitForTimers(1)
clock.Run(softRequestTimeout)
expEvent(EvTimeout)
clock.WaitForTimers(1)
clock.Run(hardRequestTimeout)
expEvent(EvFail)
rs.eventCb(Event{Type: EvResponse, Data: RequestResponse{ID: 1, Request: testRequest, Response: testResponse}})
expEvent(nil)
}
func TestServerParallel(t *testing.T) {
rs := &testRequestServer{}
srv := NewServer(rs, &mclock.Simulated{})
srv.subscribe(func(event Event) {})
expSend := func(expSent int) {
var sent int
for sent <= expSent {
if !srv.canRequestNow() {
break
}
sent++
srv.sendRequest(testRequest)
}
if sent != expSent {
t.Errorf("Wrong number of parallel requests accepted (expected %d, got %d)", expSent, sent)
}
}
// max out parallel allowance
expSend(defaultParallelLimit)
// 1 answered, should accept 1 more
rs.eventCb(Event{Type: EvResponse, Data: RequestResponse{ID: 1, Request: testRequest, Response: testResponse}})
expSend(1)
// 2 answered, should accept 2 more
rs.eventCb(Event{Type: EvResponse, Data: RequestResponse{ID: 2, Request: testRequest, Response: testResponse}})
rs.eventCb(Event{Type: EvResponse, Data: RequestResponse{ID: 3, Request: testRequest, Response: testResponse}})
expSend(2)
// failed request, should decrease allowance and not accept more
rs.eventCb(Event{Type: EvFail, Data: RequestResponse{ID: 4, Request: testRequest}})
expSend(0)
srv.unsubscribe()
}
func TestServerFail(t *testing.T) {
rs := &testRequestServer{}
clock := &mclock.Simulated{}
srv := NewServer(rs, clock)
srv.subscribe(func(event Event) {})
expCanRequest := func(expCanRequest bool) {
if canRequest := srv.canRequestNow(); canRequest != expCanRequest {
t.Errorf("Wrong result for canRequestNow (expected %v, got %v)", expCanRequest, canRequest)
}
}
// timed out request
expCanRequest(true)
srv.sendRequest(testRequest)
clock.WaitForTimers(1)
expCanRequest(true)
clock.Run(softRequestTimeout)
expCanRequest(false) // cannot request when there is a timed out request
rs.eventCb(Event{Type: EvResponse, Data: RequestResponse{ID: 1, Request: testRequest, Response: testResponse}})
expCanRequest(true)
// explicit server.Fail
srv.fail("")
clock.WaitForTimers(1)
expCanRequest(false) // cannot request for a while after a failure
clock.Run(minFailureDelay)
expCanRequest(true)
// request returned with EvFail
srv.sendRequest(testRequest)
rs.eventCb(Event{Type: EvFail, Data: RequestResponse{ID: 2, Request: testRequest}})
clock.WaitForTimers(1)
expCanRequest(false) // EvFail should also start failure delay
clock.Run(minFailureDelay)
expCanRequest(false) // second failure delay is longer, should still be disabled
clock.Run(minFailureDelay)
expCanRequest(true)
srv.unsubscribe()
}
func TestServerEventRateLimit(t *testing.T) {
rs := &testRequestServer{}
clock := &mclock.Simulated{}
srv := NewServer(rs, clock)
var eventCount int
srv.subscribe(func(event Event) {
if !event.IsRequestEvent() {
eventCount++
}
})
expEvents := func(send, expAllowed int) {
eventCount = 0
for sent := 0; sent < send; sent++ {
rs.eventCb(Event{Type: testEventType})
}
if eventCount != expAllowed {
t.Errorf("Wrong number of server events passing rate limitation (sent %d, expected %d, got %d)", send, expAllowed, eventCount)
}
}
expEvents(maxServerEventBuffer+5, maxServerEventBuffer)
clock.Run(maxServerEventRate)
expEvents(5, 1)
clock.Run(maxServerEventRate * maxServerEventBuffer * 2)
expEvents(maxServerEventBuffer+5, maxServerEventBuffer)
}
type testRequestServer struct {
eventCb func(Event)
}
func (rs *testRequestServer) Subscribe(eventCb func(Event)) { rs.eventCb = eventCb }
func (rs *testRequestServer) SendRequest(ID, Request) {}
func (rs *testRequestServer) Unsubscribe() {}