go_serialization_benchmarks VS grpc-web

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

The bebop definition specifies fixed-width types inside a struct. The format of structs cannot be changed, but there are efficiency gains by omitting all of the indices and header data. It's useless as the root message, but it's small and fast for a benchmark.

Hey everyone! Let me introduce you to mus-go - the fastest Golang serializer today. If you look at benchmarks (https://github.com/alecthomas/go_serialization_benchmarks), you can see that it could be almost twice as fast as its closest "competitor":

This might be of interest: https://github.com/alecthomas/go_serialization_benchmarks

The protobuf vs msgpack benchmarks are not too bad. Msgpack performs very decently.

https://github.com/alecthomas/go_serialization_benchmarks

Maybe submit a PR against https://github.com/alecthomas/go_serialization_benchmarks? That covers a ton of serialization formats already, so adding your library would be cool and avoid wheel reinvention.

* Additionally, client can stream data to the backend server (if bidirectional GRPC streams are used). I.e. client sends WebSocket messages, those will be transformed to GRPC messages by WebSocket server and delivered to the application backend.

As a result we have a system which allows to quickly create individual streams by using strict GRPC contract but terminating connections over WebSocket transport. So it works well in web browsers. After that no need to write WebSocket protocol, client implementation, handle WebSocket connection. This all will be solved by a suggested WebSocket server and its client SDKs.

The mechanics is similar to Websocketd (https://github.com/joewalnes/websocketd), but instead of creating OS processes we create GRPC streams. The difference from grpc-web (https://github.com/grpc/grpc-web) is that we provide streaming capabilities but not exposing GRPC contract to the client - just allowing to stream any data as payload (both binary and text) with some wrappers from our client SDKs side for managing subscriptions. I.e. it's not native GRPC streams on the client side - we expose just Connection/Subscription object to stream in both directions. GRPC streams used only for communication between WebSocket server and backend. To mention - grpc-web does not support all kinds of streaming now (https://github.com/grpc/grpc-web#streaming-support) while proposed solution can. This all should provide a cross-platform way to quickly write streaming apps due to client SDKs and language-agnostic nature of GRPC.

I personally see both pros and cons in this scheme (without concentrating on both too much here to keep the question short). I spent some time thinking about this myself, already have some working prototypes – but turned out need more opinions before moving forward with the idea and releasing this, kinda lost in doubts.

My main question - whether this seems interesting for someone here? Do you find this useful and see practical value?

By default, web browsers do not support gRPC, but we will use gRPC-web to make it possible.

You just have to use a library implementation for JavaScript https://github.com/grpc/grpc-web

TypeScript support remains an experimental feature of gRPC.

The gRPC-Web protocol supports HTTP/1 and can be used from a browser.

-- grpc-web

gRPC is mainly used in server-to-server communication, but it can also be used in client-to-server communication. gRPC-web is a gRPC implementation for web browsers. It is a JavaScript library that allows you to call gRPC services from a web browser. It supports Unary and Streaming Server API calls.

Since we're using Envoy, there's one more neat trick that we can employ. It turns out that Envoy also support gRPC-Web out of the box, a JavaScript client designed to support gRPC communication from the browser! That means that we can send gRPC messages over HTTP/1.1 as base64 encoded strings or as binary protobufs. Messages will be sent through our proxy and on to our backend service. The advantage of this is smaller and more efficient wire communication which should lead to better performance.

protoc-gen-grpc-web — a plugin that allows our front end to communicate with the backend using gRPC calls. A separate blog post on this coming up in the future.