fasthttp VS fresh

Someone made a benchmark of serialization libraries in go [1], and I was surprised to see gobs is one of the slowest ones, specially for decoding. I suspect part of the reason is that the API doesn't not allow reusing decoders [2]. From my explorations it seems like both JSON [3], message-pack [4] and CBOR [5] are better alternatives.

By the way, in Go there are a like a million JSON encoders because a lot of things in the std library are not really coded for maximum performance but more for easy of usage, it seems. Perhaps this is the right balance for certain things (ex: the http library, see [6]).

There are also a bunch of libraries that allow you to modify a JSON file "in place", without having to fully deserialize into structs (ex: GJSON/SJSON [7] [8]). This sounds very convenient and more efficient that fully de/serializing if we just need to change the data a little.

--

1: https://github.com/alecthomas/go_serialization_benchmarks

2: https://github.com/golang/go/issues/29766#issuecomment-45492...

--

3: https://github.com/goccy/go-json

4: https://github.com/vmihailenco/msgpack

5: https://github.com/fxamacker/cbor

--

6: https://github.com/valyala/fasthttp#faq

--

7: https://github.com/tidwall/gjson

8: https://github.com/tidwall/sjson

Fasthttp is one of the most powerful webservers written in Go, I'm working on a project that makes it possible to use it as a webserver for Python.

IDE: use whatever make you productive. I personally use vscode. VCS: git, as golang communities use github heavily as base for many libraries. AFAIK Linter: use staticcheck for linting as it looks like mostly used linting tool in go, supported by many also. In Vscode it will be recommended once you install go plugin. Libraries/Framework: actually the standard libraries already included many things you need, decent enough for your day-to-day development cycles(e.g. `net/http`). But here are things for extra: - Struct fields validator: validator - Http server lib: chi router , httprouter , fasthttp (for non standard http implementations, but fast) - Web Framework: echo , gin , fiber , beego , etc - Http client lib: most already covered by stdlib(net/http), so you rarely need extra lib for this, but if you really need some are: resty - CLI: cobra - Config: godotenv , viper - DB Drivers: sqlx , postgre , sqlite , mysql - nosql: redis , mongodb , elasticsearch - ORM: gorm , entgo , sqlc(codegen) - JS Transpiler: gopherjs - GUI: fyne - grpc: grpc - logging: zerolog - test: testify , gomock , dockertest - and many others you can find here

The best example I can give you is https://github.com/nutsdb/nutsdb it’s great project that got me started, one thing one should know is Go is different “yep” so there’re some coding habits that may bite you in Go and the Go compiler won’t correct you, you wanna learn about optimizations, unsafe usage check out https://github.com/valyala/fasthttp (note this is deep the rabbit hole), wanna learn concurrency check out ants https://github.com/panjf2000/ants with a little aid from “Go by example” you’re good to go

As I said in another comment, I think net/http is a good cautionary tale here. It was designed to be easy to use, and then grew organically, but performance never seems to have been a goal. fasthttp solves this, but bifurcates the ecosystem and passes on those costs to everyone who uses it. If net/http had been designed with performance in mind, this could have been avoided. net/http can't be removed or optimized, so this is a situation the Go ecosystem is effectively stuck with forever. At best, a faster version may end up in the std lib, just like netip is more modern and faster than net but the ecosystem is still bifurcated and adoption of the new package has been slow.

Fiber, while has a lot of batteries included and decent for many use cases, is known for having corner cases (because of internals like fasthttp) like https://github.com/valyala/fasthttp/issues/622

Deno: Deno with one of it's frameworks (like Fresh

The road to Fresh 2.0 (https://github.com/denoland/fresh/issues/2363) by Marvin Hagemeister Can't wait for seeing the end of the road! All in all great changes ahead.

Because I have secret magical powers that you probably don't, it's basically free for me. Here's the breakdown though:

The application server uses Deno and Fresh (https://fresh.deno.dev) and requires a shared-1x CPU at 512 MB of ram. That's $3.19 per month as-is. It also uses 2GB of disk volume, which would cost $0.30 per month.

As far as post generation goes: when I first set it up it used GPT-3.5 Turbo to generate prose. That cost me rounding error per month (maybe like $0.05?). At some point I upgraded it to GPT-4 Turbo for free-because-I-got-OpenAI-credits-on-the-drama-day reasons. The prose level increase wasn't significant.

With the GPU it has now, a cold load of the model and prose generation run takes about 1.5 minutes. If I didn't have reasons to keep that machine pinned to a GPU (involving other ridiculous ventures), it would probably cost about 5 minutes per day (increased the time to make the math easier) of GPU time with a 40 GB volume (I now use Nous Hermes Mixtral at Q5_K_M precision, so about 32 GB of weights), so something like $6 per month for the volume and 2.5 hours of GPU time, or about $6.25 per month on an L40s.

In total it's probably something like $15.75 per month. That's a fair bit on paper, but I have certain arrangements that make it significantly less cheap for me. I could re-architect Arsène to not have to be online 24/7, but it's frankly not worth it when the big cost is the GPU time and weights volume. I don't know of a way to make that better without sacrificing model quality more than I have to.

For a shitpost though, I think it'd totally worth it to pay that much. It's kinda hilarious and I feel like it makes for a decent display of how bad things could get if we go full "AI replaces writers" like some people seem to want for some reason I can't even begin to understand.

I still think it's funny that I have to explicitly tell people to not take financial advice from it, because if I didn't then they will.

Deno has also created a Next.js competitor, Fresh. I found it a few weeks ago and am starting to go through the docs, looks like a good overall concept. https://fresh.deno.dev/

https://fresh.deno.dev/ - Fresh embraces the tried and true design of server side rendering and progressive enhancement on the client side.

Along the way, I not only got the oppurtunity to revise old concepts that had blurred in my memory, but also learnt about new technologies like Fresh.js, a framework from Deno (a js runtime engine) that uses Preact, a React Routing library and used Chakra UI for the first time.

A whole project might be released as a server or framework. Frameworks like fresh, and astro) both have had things deep within them that I've wanted to reuse, within fresh it's the esbuild configuration, and islands functionality, and within astro it's the rendering of astro files themselves.

The Fresh framework by Deno cited an improved developer experience due to tighter feedback loops.