robbles

This article mentions Snapchat's hundreds of employees and multiple offices. This is one of the most obvious examples of the "what are they all doing?" question for me. I know it must take quite a few people to run operations at that scale, and of course they have an advertising business too, which likely explains the need for multiple offices. But it seems like Snapchat is still an extremely minimal app with only a couple of extra features being added over the years. Instagram had only 13 employees when it was acquired, so what role are most of these people in?

Would you be willing to post a how-to or example of doing this? I'm curious about how this trick works!

here's a pretty cool example: https://blog.heroku.com/see_python_see_python_go_go_python_g...

> I can solve most problems correctly given enough time

Have you tried asking in advance of an interview for a few extra hours to do the task(s)? I think this is a totally reasonable request to make ahead of time. Plus, any company that thinks a hard time limit is necessary to evaluate your skills is not going to be worth working for.

Glide isn't ideal either. It gets confused about vendored dependencies by 3rd party libraries and tries to download them separately. Also, it's invented it's own format for the deps and lock files instead of trying to standardise on the ones already out there.

You're making the assumption that the additional features slowing down the service aren't adding value.

More common is a "death by 1000 cuts" scenario where the various causes of slowness are apparent to the developers, but quite difficult to remove because they've become necessary to the continued success of the business.

If the back-pressure is abstracted away by this interface, how do you monitor it?

Do you really want to type either of those prefixes thousands of times? Brevity and lack of special chars in the name would be nice to have.

Is it possible that medical advances in the last couple of decades have simply delayed the deaths of many sufferers? Heart disease and cancer survival rates are usually qualified with a number of years, given that they are difficult to fully cure. With a larger aging population just barely hanging on, perhaps an upswing is just around the corner but hard to see now. i.e. the rate of death has not actually dropped, the average survival time is longer.

How did you first discover the bug? Did your deploys just start failing in an obvious way, or did your instances disappear (from ECS's perspective)?

I'm wondering if Elastic Beanstalk is hiding this from us in some way. Or maybe just has a workaround built in?

Did you run ECS on a custom AMI, or use the stock one?

We've been running with vanilla EB + ECS for months and haven't seen this at all.

From an outside perspective, it sounds like the primary issue you referenced here (the agent disconnecting) could have been due to a mismatch in configuration between the agent and docker, or maybe just a permissions issue. IIRC, the ECS agent tries to clean up containers every few hours so perhaps not being configured correctly caused it to get stuck?

A much simpler solution might look like a whitelisted set of domains/domain patterns that are given priority for network and CPU usage. For example, a meta tag on every page that lists your domain and CDN domain as "primary", and every other domain is then aggressively rate-limited by the browser.

Quiver uses cells, similar to Jupyter/IPython Notebook, and supports inline LaTeX, which is a very useful feature for when you need to display something outside the scope of Markdown (like an equation).

Awesome - looking forward to trying it out again!

Definitely would use Sourcegraph with Python support.

Yes! That's the one - thanks!

A while back, there was a service on the front page that was a combination of search engine for code, static analysis, and git repository browser. The initial implementation was for Golang only.

Does anybody remember what that project was? Does it still exist?

Slightly OT, sorry. This project just reminded me of it.

Great tool - we used gor on our team recently to test a rewritten legacy API with captures of real production data. It was invaluable for exposing edge cases in the requests that we hadn't considered.

Machine Learning for _Rust_ Hackers. I had to read one page in to check, but it looks like it's not a cross-language framework - you need to be fluent in Rust to use this.

This "16.7M colours" stat that reporters wave around always gets me. It's just 24-bit color, doesn't mean it's "high tech". I can make you 16.7M colours with a $3 microcontroller and LEDs from the hardware store.

What they should discuss instead is the color gamut of the LEDs and how much of the visible light spectrum they can accurately reproduce.

Don't you need to ship all those builtin lua plugins around with your deployment of heka though? That was what turned me off - most of the other options can be used with a binary + config file.

What kind of physical sensor are you proposing that would give you data like that? I think most ambient light sensors in laptops are just LDRs, which are only capable of giving you a 1D number.

I looked into heka a while back and was turned off by its seeming reliance on lua scripts for almost every feature. It just seemed overcomplicated to deploy and maintain for a Go app as a result, and I didn't understand why all that functionality didn't come built in.

Is this unfair? Perhaps I misunderstood the docs.

> Hand-written HTTP request parser

Aren't they all hand-written?

Why doesn't Elixir/Erlang have this problem too?

It's using green threads as well, with 1 "process" per connection, right? Is it not the same kind of scheduling?

Thanks, it would be. We run on AWS so the ease of setting up on GCE doesn't really matter beyond the experimentation phase.

If it doesn't already exist, a multi-machine Vagrant setup would probably be helpful too.

Do you have recommendations for another tool to set up a Kubernetes cluster in the first place? At first glance, it seems somewhat complex and easy to misconfigure.

Or is that something you plan to support in redspread as well?

It seems like Amazon could fix this in a straightforward way without losing the "hourly updated" part of their bestseller list - just compute a longer time period for each hour, like a rolling average. That way the current bestseller still updates every hour, but they'd need to have the most sales over the last week/month instead.

There was some discussion leading up to the release about whether to merge the "SSA" branch, which seems to be a refactor that allows for easier compile time optimisations but also slows compile times for the time being.

Does anyone know if that was included in this release?

There's a lot of disagreement in this thread about the definition of MVP, and I think it stems from the interpretation of the "Viable" part.

Based on the definitions by writers who helped popularise the term, like Steve Blank and Eric Ries, I think this word should be clarified as "viable for the purpose of the entrepreneur". Then the definitions of MVP as a process or marketing experiment make more sense. A MVP does not need to be viable for the customer (e.g. a complete, but minimal product) , just as an experiment.

> fail to get the positive feedback I've always expected

I don't know what your idea is, but that statement makes me think "confirmation bias". Start looking for something people really want instead, and build from that. Otherwise you're effectively just building a product for yourself, and trying to convince yourself that others will want it.