http2-client-grpc
http2-client-grpc copied to clipboard
A native gRPC client in Haskell
THIS REPO IS MOVING
NOTE: we'll be moving the code in this repository to https://github.com/haskell-grpc-native .
http2-client-grpc
A native HTTP2 gRPC client library using proto-lens and http2-client.
Summary
This project provides a library that leverages the code generated using proto-lens (in particular, proto-lens-protoc) to implement the client-side of gRPC services.
Usage
Prerequisites
In addition to a working Haskell dev environment, you need to:
- build the
proto-lens-protocexecutable (proto-lens) - install the
protocexecutable
Adding .proto files to a Haskell package
In order to run gRPC:
- generate the
Protostubs in somegendirectory
A single protoc invocation may be enough for both Proto and GRPC outputs:
protoc "--plugin=protoc-gen-haskell-protolens=${protolens}" \
--haskell-protolens_out=./gen \
-I "${protodir1} \
-I "${protodir2} \
${first.proto} \
${second.proto}
- add the
gensourcedir for the generated to your .cabal/package.yaml file (cf. 'hs-source-dirs'). - add the generated Proto modules to the 'exposed-modules' (or 'other-modules') keys
A reliable way to list the module names is the following bash invocation:
find gen -name "*.hs" | sed -e 's/gen\///' | sed -e 's/\.hs$//' | tr '/' '.'
Unlike proto-lens, this project does not yet provide a modified Setup.hs.
As a result, we cannot automate these steps from within Cabal/Stack. Hence,
you'll have to automate these steps outside your Haskell toolchain.
Calling a GRPC service
In short, use http2-client with the Network.GRPC.call function and let the types guide you.
Example
You'll find an example leveraging the awesome grpcb.in service at https://github.com/lucasdicioccio/http2-client-grpc-example .
gRPC service mapping
The Protobuf format specifies the notion of Service which can have one or more
RPCs. The gRPC protocal maps these services onto HTTP2 headers and HTTP2
frames. In general, gRPC implementation generate one inlined function per RPC.
This implementation differs by decoupling the HTTP2 transport and leveraging
generics to provide generic functions. This design allows the
Network.GRPC.call to be type safe and multi-usage. For instance, you can wrap
this function with whatever metrics/benchmark code and have all your RPC calls
be monitored in a uniform way.
Status
This library is currently an early-stage library. Expect breaking changes.
Next steps
- provide function to map raw results into commonly-understood GRPC errors
- (nice-to-have) support effectful compression (Zlib pure compression is supported)