I'm not going to dive into the specifics of my thoughts on this question. I think a lot of comments here address this.

But does anyone else get embarrassed of their career choice when you read things like this?

I've loved software since I was a kid, but as I get older, and my friends' careers develop in private equity, medicine, law, {basically anything else}, I can tell a distinct difference between their field and mine. Like, there's no way a grown adult in another field evaluates another grown adult in the equivalent mechanism of what we see here. I know this as a fact.

I just saw a comment last week of a guy who proudly serves millions of webpages off a CSV-powered database, citing only reasons that were also covered by literally any other database.

It just doesn't feel like this is right.

[delayed]

I dunno, seems like a really confusing question. Communication is important but I can imagine that explaining this verbally on the spot to an interviewee would not be straightforward especially because the assumptions made around single threading get confusing. If it's just a Javascript question say that - because it seems it basically is. Writing this in go would be super easy so I think the question is just asking people how well they understand Javascript.

echoing the other comments about this being a rather terrible interview question...

> this interview can be given in JavaScript or any other language

it's a language-agnostic question...but it revolves around the assumption of making a callback on request completion. which is common in JS, but if you were solving this in some other language, that's usually not idiomatic at all.

followed by:

> For candidates without JavaScript experience or doing this interview in pseudo-code, you have to tell them that there's another function available to them now with the following signature:

> declare function setTimeout(callback: () => void, delayMs: number): number;

so you add in this curveball of delaying requests (it's unclear why?), and it's trivial to solve...using a function from the JS stdlib. and if the candidate is not using JS, you need to tell them "oh there's a function from JS that you can assume is available"

> After sendOnce is implemented, it's time to make the interview a lot more interesting. This is where you can start to distinguish less skilled software engineers from more skilled software engineers. You can do this by adding a bunch of new requirements to the problem

as you originally specified it, this code is a workaround for a buggy server. and for Contrived Interview Reasons we can't modify the server at all, only the client.

in that scenario, "extend it into a generic queue with a bunch of bells and whistles" is maybe the worst design decision you could pursue? this thing, if it existed in the real world, should be named something like SingleRequestQueueForWorkingAroundHopelesslyBuggyServer with comments explaining the backstory for why it needs to exist. working around the hopelessly buggy server should be roped off into one small corner of the codebase, and not allowed to infect other code that makes normal requests to non-buggy servers.

Maybe I came into this article knowing too much about the solution, but I dont agree with commenters saying this is a poorly designed interview question. Its a blog post as well, not the format that would be presented to a candidate.

I think it has clear requirements and opportunities for nudges from the interviewer without invalidating the assessment (when someone inevitably gets tunnel vision on one particular requirement). It has plenty of ways for an interviewee to demonstrate their knowledge and solve the problem in different ways.

Ive run debounce interview questions that attempt to exercise similar competency from candidates, with layering on of requirements time allowing (leading/trailing edge, cancel, etc) and this queue form honestly feels closer to what Id expect devs to actually have built in their day to day.

for the record (and disregarding how appropriate this is as an interview question): in JS you can (ab)use the event loop and promise chains to do this for you without managing any queues or lists manually. You have a single `let job = Promise.success();` as a global var, and scheduling a new job becomes `job = job.then(f, errHandler).then(callback, errHandler)`. It's a nightmare to debug (because you can't "see" the in-process queue) but it means you don't have to muck around with manual lists, queues, loops, shift/unshift, "isProcessing" flags etc, all of which is basically you reimplementing that native functionality in user space. It completely sidesteps the bug of TFAs naive implementation.

Not advocating for this in prod but in the context of a programming puzzle it can be neat.

late edit: ironically this is also a comment on the LLM talk in TFA: messing with the event loop like this can give you a strong mental model of JS semantics. Using LLMs I would just have accepted a loop and never learned about promise chains. This is the risk in using LLMs: you plateau. If you will allow a tortured metaphor: my naive understanding of SR is that you always move at light speed, but in 4 dimensions, so the faster you move in the 3D world, the slower you move through time, and vice versa. Skill is similar: your skill vector is always a fixed size (= "talent"?). If you use LLMs, it's basically flat: complete tasks fast but learn nothing. Without them, you move diagonally upwards: always improving, but slower in the "task completion" plane. Are you ready to plateau?

      let isProcessing = false;

      async function checkFlagAndRun(task) {
          if (isProcessing) {
              return setTimeout(() => checkFlagAndRun(task), 0);
          }

          isProcessing = true;
          await task();
          isProcessing = false;
      }

      function delayedLog(message, delay) {
          return new Promise(resolve => {
              setTimeout(() => {
                  console.log(message);
                  resolve();
              }, delay);
          });
      }

      function test(name,num) {
          for (let i = 1; i <= num; i++) {
              const delay = Math.floor(Math.random() * 1000 + 1);
              checkFlagAndRun(() => delayedLog(`${name}-${i} waited ${delay} ms`, delay));
          }
      }

      test('t1',20); test('t2',20); test('t3',20); 

Here’s an idea for fair interviews:

Interviewer and candidate meet at time X for 1h session of “live coding”. A saas throws at them both one problem at random. Let the game begin. The company can decide if they want interviewer and candidate to collaborate together to solve the problem (the saas is the judge) or perhaps they both need to play against each other and see who gets the optimal solution.

You can add a twist (faangs most likely): if the candidate submits a “better” answer than the interviewer’s , candidate takes over their job.

An LLM could be very well behind the saas.

Oh boy, I wouldn’t feel that nervous anymore in any interview. Fairness is the trick. One feels so underpowered when you know that the interviewer knows every detail about the proposed problem. But when both have no idea about the problem? That’s levelling the field!

As a self promoting post I think the author did a good job. As an interview format, I would rather work somewhere less ego driven development and more real problem oriented workplace. But that is just me. Someone could prefer these kind of interviews. I also did a set of questions for java engineers in the past and I always felt there is something really icky. I also noticed the engineers with huge ego revel in these kind of candidate assessments as it makes the feel good, but the candidate performance is poorly tested. Thats what the probation period is for. Just ask the candidate whats his experience. Asking these "cleverly" designed problems is nice for the interviever importance of keeping his job, but is not really usefull. You could even miss a good engineer. Perhaps i see this too narrow and you just really want to observe what the candidate is thinking, but you could make a couple of not really complicated questions and you could see where he is at. I dont bite this head-game at all.

I've implemented multiple production versions of this problem (but not in JavaScript)[1], so maybe my view of this problem is miscalibrated...

This feels both too easy and too hard for an interview? I would expect almost any new grad to be able to implement this in the language of their choice. Adding delays makes it less trivial, except that the answer is... Just use the function provided by the language. That's the right answer for real code, but what are you really assessing by asking it?

[1] https://github.com/google/guava/blob/master/guava/src/com/go...

Oh I see what you're doing here...this is just an interview to massage the interviewer's ego.

It must be so boring working you

> "This is a good way to test how "AI-native" the candidate is."

yuck

You can also schedule code to run each "tick" of the event loop, which is a non-blocking version of a while loop.

Or you could promisify the send function and use normal async/await.

   let q = Promise.resolve(),
     sendAsync = (p) => new Promise(r => send(p, r)),
     sendOnce = (p, c, ms) => setTimeout(_ => q.then(_ => sendAsync(p)).then(c), ms)

Or you could actually spin up a new worker thread and get multithreading :P

Do I have to use JavaScript? I'd write it in Java in a way that it would be trivial to ramp up the number of connections in the pool once they fix their stupid server.

> Here's a naive, faulty implementation

For this first implementation, I don't see anything ever added to the queue. Am I missing something? New task is added to the queue if the queue is not empty only, but when the queue is empty the task is executed and the queue remains empty so in the end the queue is always empty?

Promises in JS make this stuff much easier (at least to my mind):

const lockify = f => {

  let lock = Promise.resolve()

  return (...args) => {

    const result = lock.then(() => f(...args))

    lock = result.catch(() => {})

    return result.then(v => v)

  }

}

Interviews are a two way street. If you strongly imply that working around servers that only do one thing is part of your day to day work, a lot of people will want to work somewhere they can learn about more modern software.

Interestingly I think I would over-think this. The interviewer is assuming a single server, running in a VPS type environment. There's also no notion of state persistence/timeout/recovery etc. I think I'd immediately have started factoring those things in.

ALSO while JavaScript is a single threaded environment, the while solution would still basically work due to the scheduler (at least if you yield, await sleep, etc.)

Definitely one of those where the interviewer wants to show how smart they are.

If I were asked this question, the first thing I said would be this is a poorly designed architecture. Client is the poor place to do throttling by itself. It has no information on the aggregated load of the system. It makes assumption that leads to complicate code in the sample code. There're more robust and better ways to do flow control and throttling.

The use-case described is ill suited to be addressed by the client.

Which make the whole coding exercise moot.

What if there are 1 million users opening the browser at the same time?

The queue question is fun but doing it in the client is not right.

In another thread I asked if leetcode style questions are still common, and the answer I got was yes, so where would I encounter this kind of question? I only ever got leetcode style questions or something like "what is the difference between var and let"

Is the send function considered non-blocking?

I don’t know anything about the author, so just speculating here: assuming that the interview lasts 1h, it’s not realistic (nor fair) to judge the candidate’s answer if the interviewer has spent more than 1h to think about the problem and potential solution(s).

Interviewers have thought about the problem they propose countless of times (at least once per interview they have hold) each time refines their understanding of the problem, and so they become god of their tiny realm. Candidates have less than one hour, add to that stress and a single shot to get it more or less right. You’re not assessing candidate’s ability to code nor their ability to handle new requirements as they come.

thanks for sharing and by reading the blogpost and the comments I think I get the whole point: it's all about how engineers understand the requests and the reasoning about how does it approach that more than the code itself. If this raw code really works or not it's almost secondary here - IMHO nobody I'll start coding a real queue out of blue like this.

I’m really confused why this is an “async queue.” Seems pretty synchronous to me since requests are completed in order, one at a time. I literally wrote something to do this in the past few months. This was to work around JavaScript’s asynchronous IO making a shit show of updates from library being used in the client. I had to queue requests in JavaScript, have them execute in order (FIFO), and explicitly described it “synchronous.”

Is it only “async” because it’s doing it in JavaScript and the underlying network request API is asynchronous? Seems like, IMHO, a really bad way to describe the desired result since all IO in JavaScript is going to be async by default.

The minDelay extension feels contrived and also the solution sucks. None of the pending requests are actually added to the queue until the timeout is finished, meaning you have no knowledge of all these delayed requests until the timeout passes and they all enqueue themselves

The explanation is way too long, in an area that is pretty big and can be done in many ways. Couple this with candidates who will variously be fearful of asking to many or too few questions, and you just have confusion about who is good and who is not.

This is one of the most confusing and badly worded interview problems I've ever seen. If I had been given this problem, I'd view it as a signal that I'd be wasting my time working with the folks that thought it was good.

Hmm… this code doesn’t work in the real world unless you only run it on a single machine. Perhaps a more interesting question is how to make a multi-node queue with max N concurrent requests.