Thursday, April 7, 2022
HomeSocial MediaWhy We Write Every little thing in Go

Why We Write Every little thing in Go

Okay, okay, we don’t write every thing in Go — we have now a complete front-end codebase in Typescript/Javascript, we have now some Python, we have now some straight Bash. Imagine it or not, we even had some Tcl up till early 2022 earlier than we ported it, after all, to Go.

At Bitly, we’ve been all in on Go as our most well-liked back-end language since about 2015. 


We like instruments which can be environment friendly and highly effective. We prefer to keep away from the unnecessarily complicated. We like repeatable patterns, open supply, clear requirements and excessive efficiency. We like easy-to-read code as a result of we like easy-to-maintain code. That is why we like Go.

As you would possibly guess, we energy the Bitly ecosystem with a group of internet apps, APIs, microservices and datastores.

The Bitly system processes tens of billions of redirects and a whole lot of hundreds of thousands of recent hyperlinks every month. We service prospects worldwide and run roughly 15 direct user-facing providers (internet apps and exterior APIs), 35 inside APIs, 130 queue shopper processes, 150 cron jobs, 20 datastores and who is aware of what number of advert hoc scripts. 

All advised, Go accounts for a little bit over half of our code base.

As soon as Upon a Time…

There as soon as was a time once we wrote most backend providers in Python, and once we needed severe efficiency, we dropped into C. We used Python as a result of it was straightforward to jot down, fairly performant, and there was a pleasant internet server framework known as Twister that scaled fairly properly with its efficient use of non-blocking community I/O. 

We used C as a result of, properly, it was C and you may’t beat its efficiency. You possibly can, nonetheless, beat its maintainability. (What number of C builders do you know on the market?)

However what if we might have a easy and easy language that gave us excessive efficiency as properly?

Enter Go

In 2014, we wrote a little bit open supply venture known as NSQ ( and put a promising new language known as Undergo its paces. We preferred what we noticed a lot that we began writing every thing new in Go, and shortly thereafter we started porting all legacy providers to Go as properly.

Some of the compelling findings from that early expertise was that we might run the identical workload in Go on fewer servers than the Python model and the response instances have been usually twice as quick. That, in and of itself, would have been sufficient, however add to the efficiency features the truth that builders unfamiliar with Go could be onboarded to it in a matter of weeks, and we have been offered.

Efficiency: Go vs. Python

Serving a easy request

Right here’s a easy instance of 1 path served up by an online app. That is merely the serving of a website affiliation file for iOS deep linking for a given customized area. (If you happen to don’t know what meaning, suffice it to say, iOS calls it and it’s a easy json response.)

Simply this:

% curl
  "applinks": {
    "apps": [],
    "particulars": [
        "appID": "",
        "paths":["NOT /","NOT /*+","NOT /m/*","NOT /.rss/*","NOT /index.php/*","NOT /index/*","NOT /static/*","NOT /etc/*","NOT /media/*","NOT /assets/*","NOT /sites/*","NOT /images/*","/*"]

Beneath you’ll discover Python and Go variations of the code to deal with this HTTP request.

The 2 items of code are mainly doing the identical factor:

  • Name our inside API to get the positioning affiliation information for a given customized area
  • Return some json

Right here’s the Python/Twister:

Route definition:

(r"^/apple-app-site-association$", app.site_associations.SiteAssociations, {"os": "ios"}),


from lib.deeplinks_api import get_site_associations

class SiteAssociations(app.primary.BaseHandler):

    BASE_TMPL = {
        "ios": {
            "applinks": {
                "apps": [],
                "particulars": [],

    def get(self):
        callback = functools.partial(self._finish_deeplinks_api_get_sas, os=self.os)
        get_site_associations(os=self.os,, callback=callback)

    def _finish_deeplinks_api_get_sas(self, response, os=None):
        if response.error:
            self.set_header("Content material-Sort", 'software/json')
            return self.end(json.dumps(deepcopy(self.BASE_TMPL[os])))

        resp = json.hundreds(response.physique)
        information = sorted(resp, key=lambda ok: int(ok.get('precedence', 1)))
        return self._generate_response(os=os, information=information)

    def _generate_response(self, os=None, information=[]):
        response = deepcopy(self.BASE_TMPL[os])
        for merchandise in information:
            particulars = {
                "appID": merchandise["apple_app_entitlement_id"],
                "paths": self.PATHS,

        self.set_header("Content material-Sort", 'software/json')

Right here’s the Go:

Route definition:

router.GET("/.well-known/apple-app-site-association", wrapper(siteAssociations.Apple)) // wrapper information metrics concerning the request


kind SiteAssociationHandler struct {
	deeplinksAPI *deeplinksapi.Shopper

kind AppleAppLinkDetail struct {
	AppID string   `json:"appID"`
	Paths []string `json:"paths"`

func (h *SiteAssociationHandler) Apple(w http.ResponseWriter, r *http.Request) {
	host, err := idna.ToASCII(r.Host)
	if err != nil {
		log.WithContext(r.Context()).Warn("invalid host")
	sa, err := h.deeplinksAPI.GetSiteAssociations(r.Context(), host, "ios")
	if err != nil {
		swap err.Error() {
		webresponse.UnknownError500(r.Context(), w, err)

	particulars := make([]AppleAppLinkDetail, 0, len(sa))
	for _, merchandise := vary sa {
		particulars = append(particulars, AppleAppLinkDetail{
			AppID: merchandise.AppleAppEntitlementID,
			Paths: siteAssociationPaths,

	w.Header().Set("Content material-Sort", "software/json")
	err = json.NewEncoder(w).Encode(AppleSiteAssociation{AppleAppLinks{
		Apps:    make([]interface{}, 0),
		Particulars: particulars,
	if err != nil {
		webresponse.InternalError500(r.Context(), w, err)

Now take a look at the distinction in efficiency.

Fig 1. Python – Response instances in milliseconds. Pattern measurement: 67,179 requests.

Fig 2. Go – Response instances in milliseconds. Pattern measurement: 50,192 requests.

That’s nearly half the typical response time!

A handful of milliseconds enchancment might look like no huge deal, however this code does nearly nothing. It makes a single API name and marshals some information to return. A 50% enchancment in response time at this stage will increase exponentially because the complexity of the code will increase. Extra complicated code is made up of easy constructing blocks like this.

Want a extra dramatic instance?

gsutil (Python) to gscopy (Go)

Perhaps you utilize gsutil, like we used to, to repeat recordsdata to Google Storage. Mundane activity, proper?

gsutil is a Python-based utility supplied by Google ( We rewrote the components of it that we wanted in Go and right here’s what we discovered.

The graph beneath reveals the length of a cron job that runs each half-hour. It copies 7,000 to 11,000 recordsdata per run. Common file measurement is 100Mb. (These of you crunching the numbers will see that the Python model of that cron job didn’t actually run each half-hour. That model took near an hour to run.)

Fig 3. Time to repeat 7,000 to 11,000 recordsdata to Google Storage, Python vs. Go

The Go model is greater than a full order of magnitude sooner than the Python model. And take a look at the CPU drop. 

Fig 4. Swap from gsutil to gscopy, impact on CPU utilization


I’m positive you need to see the code. In a nutshell, it makes use of the Google-provided Go package deal with a service account. Maybe we’ll get this into an open supply repo close to you.

The Affect

The influence of porting to Go must be apparent now.

Our backend providers usually take certainly one of solely a handful of varieties:

  • Net apps (, inside instruments)
  • Exterior-facing HTTP APIs (the Bitly API at
  • Inside HTTP APIs (shorten API, billing API, person administration API, and many others. — accessible solely to our inside techniques)
  • Queue readers (async message shoppers)

Out of roughly 270 backend providers right now, about 155 of these are Go. That makes for a efficiency acquire of…divide by 2…carry the 1…A LOT.

However Wait, There’s Extra!

Past the efficiency advantages of Go, there are a number of different components that we’ve come to understand concerning the language.


The Go group is a pleasant, welcoming, lively and supportive bunch. We love taking part in Gophercon and usually have Bitly builders in attendance yearly. As well as, on the time of this writing, there are 388 Go Meetups in 247 cities worldwide (

Documentation, tutorials & instruments

Go is just not solely properly documented however documentation is constructed into the language. Documentation is generated from the inline feedback within the supply code with godoc ( The supply code is accessible to all, therefore if you see the docs saying this:

Fig 5. Go docs.

… it maps on to the supply code that claims this:

Fig 6. Go supply code.;l=71

The “Tour of Go” tutorial ( offers a wonderful introduction to the language and its ideas, making onboarding builders who might have expertise in different languages very easy.

The Go Playground ( makes it straightforward to check code snippets with out having to arrange your total runtime surroundings.

Go Requirements & Bitly Go Requirements

Whereas there are many methods to jot down code in any language, Go offers a information “for writing clear, idiomatic Go code” at Efficient Go ( 

We even have a set of in-house requirements which can be laid on high of the Go requirements. Go lends itself to clear, constant patterns in our method to issues like software structure, database entry and API design. A number of the matters in our inside Go documentation seem like this:

Fig 7. Bitly Inside Go Requirements desk of contents

We extremely worth the potential of recent builders to shortly develop into productive on code contributions or code evaluate. Clear patterns from each the language itself and from our inside requirements assist make the onboarding expertise for brand new builders manageable. With a 12-year outdated code base, predictability is paramount. 

The March Continues…

Fig 8. Thus spake GitHub. Present language breakdown.

It’s a protracted march to make a longtime firm’s code base environment friendly, constant and clear. However it’s a worthy aim and Go has confirmed to be a wonderful option to get us there.

Bitly is hiring in Engineering and plenty of different departments. Take a look at our open listings in the event you’re :



Please enter your comment!
Please enter your name here

Most Popular

Recent Comments