Ben, you are correct the dial thing, not sure how that ended up commented. Anyway, I've removed the custom dial (and tried with the timeout enabled as well), it did increase r/s a lil bit (5%-10% or so) but it increased the amounts of timeouts on the remote urls as well.
El jueves, 2 de febrero de 2017, 14:11:29 (UTC-3), James Bardin escribió: > > First things I notice are that you're overriding the default dialer with > one that doesn't timeout, and you've commented out ReadTimeout in the > client. Both of those could indefinitely hold up client connections > regardless of the DoTimeout call, which just ensures that the Do function > returns before the deadline. > > Otherwise, I think you're going to have to instrument the code a little > better to see what is holding you up. > > > On Thursday, February 2, 2017 at 11:27:38 AM UTC-5, emarti...@gmail.com > wrote: >> >> Thanks for the answer. >> >> Yes, it seems to be blocking, I just fixed it with: >> http://blog.sgmansfield.com/2016/01/the-hidden-dangers-of-default-rand/ >> >> After that change my code is working a lil bit better but still I see a >> ton of timeouts + high latency on responses. Maybe the code is locking on >> another part of the code? >> >> El jueves, 2 de febrero de 2017, 4:09:07 (UTC-3), land...@gmail.com >> escribió: >>> >>> func randInt(min int, max int) int { >>> rand.Seed(int64(time.Now().Nanosecond())) >>> return min + rand.Intn(max-min) >>> } >>> >>> Is the culprit. the default rand locks globally for concurrent access. >>> You need to create a new rand in each goroutine you want to use it in, for >>> maximum speed. >>> >>> On Wednesday, February 1, 2017 at 8:26:07 PM UTC-6, emarti...@gmail.com >>> wrote: >>>> >>>> Hello, >>>> >>>> I'm writing a POC for a future RTB platform. Basically I'm doing stress >>>> tests by receiving HTTP requests and performing HTTP GET requests in the >>>> background. >>>> >>>> Issue I face is that when i try to scale up, URLs starts to timeout. I >>>> am trying to find what our bottleneck is but so far no luck (we aren't >>>> running out of ephemeral ports). >>>> >>>> Remote URLs takes about 500ms-1000ms to respond. We are stress testing >>>> it with wrk with 5000 concurrent requests. (5000 incoming requests which >>>> translated into 30 remote URL requests = 150k requests) >>>> >>>> Here is our code: >>>> >>>> package main >>>> >>>> import ( >>>> "net/http" >>>> "time" >>>> "runtime" >>>> "io" >>>> "net" >>>> "github.com/valyala/fasthttp" >>>> "math/rand" >>>> ) >>>> >>>> type HttpResponse struct { >>>> url string >>>> response *http.Response >>>> err error >>>> } >>>> >>>> var urls = []string{ >>>> //lots of urls >>>> } >>>> >>>> >>>> //func asyncHttpGets(urls []string) []*HttpResponse { >>>> >>>> var( >>>> clients []fasthttp.Client >>>> total_clients int >>>> max_urls int >>>> ) >>>> >>>> func init(){ >>>> max_urls=30 >>>> clients =append(clients, create_client() ) >>>> } >>>> >>>> func create_client() fasthttp.Client{ >>>> >>>> return fasthttp.Client{ >>>> Dial: func(addr string) (net.Conn, error) { >>>> var dialer = net.Dialer{} >>>> return dialer.Dial("tcp", addr) >>>> }, >>>> >>>> MaxIdleConnDuration:30*time.Second, >>>> MaxConnsPerHost:2024, >>>> //ReadTimeout: 1*time.Second, >>>> } >>>> } >>>> >>>> func randInt(min int, max int) int { >>>> rand.Seed(int64(time.Now().Nanosecond())) >>>> return min + rand.Intn(max-min) >>>> } >>>> >>>> >>>> func asyncHttpGets(urls []string) []string{ >>>> ch:=make(chan string,max_urls) >>>> var responses []string >>>> >>>> cl:=0 >>>> for i := 0; i <= max_urls; i++ { >>>> url:=urls[i] >>>> go func(url string) { >>>> >>>> req := fasthttp.AcquireRequest() >>>> req.SetRequestURI(url) >>>> req.Header.Add("Connection", "keep-alive") >>>> resp := fasthttp.AcquireResponse() >>>> clients[cl].DoTimeout(req, resp,1*time.Second) >>>> >>>> bodyBytes := resp.Body() >>>> ch <- string(bodyBytes) >>>> >>>> fasthttp.ReleaseRequest(req) >>>> fasthttp.ReleaseResponse(resp) >>>> }(url) >>>> } >>>> >>>> for { >>>> select { >>>> case r := <-ch: >>>> responses = append(responses, r) >>>> if len(responses) == max_urls { >>>> return responses >>>> } >>>> } >>>> } >>>> return responses >>>> } >>>> >>>> func hello(w http.ResponseWriter, r *http.Request) { >>>> results := asyncHttpGets(urls) >>>> for _, result := range results { >>>> io.WriteString(w,"%s status: %s" + " " + result + "\n") >>>> } >>>> } >>>> >>>> >>>> func main() { >>>> runtime.GOMAXPROCS(0) >>>> >>>> server8000 := http.NewServeMux() >>>> server8000.HandleFunc("/", hello) >>>> http.ListenAndServe(":8001", server8000) >>>> } >>>> >>>> >>>> Any help is really appreciated. >>>> >>> -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
