tcbyrd

Rubber ducking discovery is a very good description of how I use it too. I don’t necessarily care if it’s 100% right, but it feels like it short circuits my “time to solution” by a significant percentage over hunting around through endless blog posts and SO answers. ChatGPT fills the gap between an idea and getting the LSP to tell me what will actually work.

Earlier today it totally hallucinated a built-in function, but when I started typing it out, the LSP kicked in and I could tell what the GPT was “thinking”. Before that I didn’t even know what the name of the function was that would do what I want, but it was close enough that it stopped me from having to parse the reference docs on my own.

Pages sites no longer work on github.com

https://github.blog/changelog/2021-01-29-github-pages-will-s...

I think there's a lot of it based in the fact that Rust was born at Mozilla and we want to see a modern language succeed that isn't tied to one of the big tech companies.

The integration mentions working with Microsoft Graph API, and it’s doing the generation on the client, so seems plausible this has no more access than any other OAuth app. The Graph API can be locked down with Access Policies. I’m sure there will be features to further cordon off data to the AI, similar to how OneDrive has an encrypted vault which is not accessible from the Graph API.

Hyperconverged is all about software defined storage and compute. You can create those service boundaries all on one cluster and pool like nodes together to create one big mesh of compute and storage. The precursor were things like EMC and NetApp storage clusters which typically had 2-4 "compute nodes" with a rack full of direct attached storage. This created a massive problem come upgrade time, and the term was called "forklift upgrade" [1] for a reason. With HCI you can add and replace single boxes as needed.

Also, latency is not an implementation detail for a lot of data-intensive workloads. We're talking differences of 1-10ms latency for Network Storage and 250-500µs for a local NVME SSD.

[1]: https://www.architecting.it/blog/what-is-a-forklift-upgrade/

Wait until you find out how many people in the 21st century are still iterating on the idea of the wheel itself. Imagine if nobody tried reinventing wheels since they were first introduced and you have your answer.

It’s an example of lenses designed for different workflows. Film style shoots it’s not uncommon to swap out a lens in between shots, and focus is sometimes a creative choice. Autofocus doesn’t work when you actually want to shift the focus from one actor to the next without them or the camera moving. You need manually focused lenses with precise control of the focus ring. With touch screens you can get close, because you can tell the software what subject to track, but a good focus puller that’s following the emotion of actors is tough to recreate in software. This is also why they’re generally not designed to be parfocal. Film shoots very rarely zoom the lens while recording.

This is a great example to me why production and broadcast equipment is and may always be so different. I'd never even consider something like an RX-10 camera when what I need is 4 cameras attached to a live switcher genlocked to the same timecode.

I went back and forth on this balance for years. I wanted that large sensor look and feel, with great full-frame lenses, but trying to use it live was a nightmare workflow. This was before Canon started making the C series that mixed the world of a camera with an EF mount that also had SDI I/O. I tested those, but still couldn't trust them in a live broadcast, mainly because of the lack of a parfocal lens I can focus remotely. I think with cameras like Blackmagic's Ursa and some more "prosumer" lenses that are parfocal we may be getting closer to having the best of both worlds, but I haven't researched that in a couple of years.

Broadcast optimized gear is like having Kubernetes. Everything built for broadcast talks SDI/NDI so it's generally possible to operate or at least monitor the device remotely. Then you have routers and mixers designed to modify those video feeds in real-time with as little latency as possible. One of my first gigs in broadcast involved learning how to remotely white balance cameras from the control room or patch the output of one video feed from a satellite truck into a monitor on set, while also recording that same feed in a completely separate room. I've also spent time doing "documentary style" and film production and was amazed at how different everything was, and how some workflows that are super straightforward in a newsroom are basically impossible with things like GoPros and DSLR setups.

Very cool! I've always wanted to see the inside of one. I'm speaking mostly from experience operating these cameras "in the field". My engineers never let me open one up.

I think I was just making sure it was clear (to others, since you probably are aware) that it's more than just lighting/apertures. Being good at a variety of lighting conditions, and being able to do that from 18mm-1700mm with a variable speed zoom without losing focus is where things get really big. There's a lot of features in the camera body to support this specific type of production beyond the lens, too. For live TV, an operator is making split-second decisions around composition (including adjusting for graphics on the screen), lighting, focus, and movement, while also monitoring on-air status and listening to both director and (sometimes) program audio. And then you have to add on the massive rigs that make sure everything's in perfect balance so you can whip the camera around quickly and have it stop on a dime. It's a whole system. A good high-end rig is like driving a Porsche. Sure the engine is a big part of the power and engineering, but if you don't also have a properly tuned suspension and grippy tires, you aren't going to have much fun driving it.

I’d recommend you watch the video. There’s a lot more to it than just the lens and it’s not just “to gather a lot of light”. You can gather a whole bunch of light on an f0.8 lens in way less weight. The video does a great job with the details of the physics.

That may explain the “expensive” part but it doesn’t explain the “huge” part. Could Fujinon make their huge lenses cheaper? Maybe. But as the video points out there’s practical reasons the equipment has to be the physical size that it is, and yes the market for productions that have the team and budget to support and operate the equipment is relatively small. And those cameras are usually a 5+ year investment.

I was working in a Dell call center in 2007 when this plague started. As soon as I pulled up the service tag I could tell how the call was going to go. Yellow lights on the back indicating power but no successful POST. Specific chassis manufactured in a specific time range. There was a massive recall but most people don’t return their working machine, so it took months before the calls slowed down.

Closest you’ll get is something like a Blackmagic ATEM switcher, using “Program” as the output. Then you can map a software key to swap it out in a single frame. As discussed in another thread, the computer has to be sending the display signal to some device that is seen as a display, and that switcher keeps the video from both PCs buffering at 60 or 30 fps. To the computer it can’t know the display switched and the display can’t handle two simultaneous streams, so you have a device in the middle that bridges that. Down side is depending on the quality of the switcher you might get 1-2 frames of lag because it’s optimizing for frame accurate sync to ensure the least possible “flicker” when the source is switched.

There’s actually a trick where you don’t have to use Godaddy. I don’t know why they don’t open this up to users who want it. All you have to do is find the DNS challenge in the query param that’s passed to Godaddy. After that there’s no real persistent connection. Have been using a custom domain on the family plan for over a year this way, with my domain on Namecheap.

Something I’ve been thinking about lately around all these external observability tools. Does streaming logs count towards your outbound bandwidth from providers like Vercel/Netlify/etc? One huge benefit to using everything the cloud provider gives you is all of it stays “on-net”, so to speak. I wonder if I use a tool like this or Honeycomb, am I not only monitoring ingest costs but also egress from the app provider. A lot of these solutions are hosted on top of AWS, so it would be awesome to find out they zero-rate traffic between them, but I feel like that’s probably not the case?

It's also a signed cookie, and we do the signature verification you mention if the format is correct, but if the format doesn't match, that cookie is discarded at the edge, before it even gets sent down to our app servers.

Because GitHub Pages isn't written in .NET. However, the general implementation is heavily inspired by this and other similar projects.

Hi from the Pages team! This was such a great find and can't thank you enough for helping ensure Private Pages is as secure as possible.

This report helped uncover:

- A bug in Openresty where `ngx.redirect` didn't handle unsafe characters [1]. While the fix is now in the latest version of Openresty, a quick patch was to build the URL safely before using it in the redirect.

- You should check for case sensitivity when reading `__Host` prefixed cookies, and verify the values against your expected format. It's possible for both `__HOST-Foo` and `__Host-Foo` cookies to exist, and only the `__Host` prefix requires the `Secure` and `HttpOnly` attributes [2]. In our case we strip all cookies at the edge using Varnish (VCL) to ensure no user-supplied cookies make it to our origin, and now we also ignore any "Secure" cookies that don't appear to have been set by our servers.

[1]: https://github.com/openresty/lua-nginx-module/pull/1654

[2]: https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Se...

I haven't personally done detailed cost analysis lately, but if you have systems that regularly operate at 80+% of capacity, I can't see how the operating costs of any cloud operator can be cheaper than operating it yourself. Their whole pricing model is based on being able to over-provision compute and move workloads around within a datacenter. As much as people talk about failing hard drives and other components at scale, failure rates are still low enough that you could operate dozens of systems at full capacity for 3+ years with maybe a handful of legit hardware failures. To rent that same compute from any cloud provider would cost significantly more. The cheapest "Dedicated Host" on AWS will cost you almost $12k over 3 years if you pay for it on-demand, and it's equivalent in specs to something you can buy for ~$2k.

> am I missing something?

I'd want more background behind what you mean by "at least for business". What kind of business? Obviously IaaS providers like Digital Ocean and Linode are are type of business that would not use other clouds. Dropbox and Backblaze as well would probably never use something like S3. And there are legitimate use cases outside of tech that have needs in specific teams for low latency compute, or its otherwise cost and time prohibitive to shuttle terabytes of data to the cloud and back (3D rendering, TV news rooms, etc). If you're talking about general business systems that can be represented by a website or app with a CRUD API, then most of that probably doesn't require on-prem. But that's not the only reason businesses buy servers.