grpc-transcoder

gRPC-JSON Transcoder Config Generator

A simple utility to generate an Istio EnvoyFilter CRD configuring Envoy's gRPC-JSON transcoder filter. This filter allows a gRPC server to serve JSON HTTP requests without any additional work on the server's part, regardless of the language the gRPC server is written in. Envoy transcodes the client JSON requests into gRPC before handing the request to the server, and transcodes the server's response from gRPC back into JSON before sending it to the client. Even better, it can do this while still serving gRPC clients on the same port!

Usage

1. Build by binary via make build which creates a binary named gen-transcoder.

   Alternatively, use go run github.com/tetratelabs/istio-tools/grpc-transcoder and pass the same CLI arguments as we use in the other examples.

2. Build your protobuf API definitions with protoc, instructing the compiler to produce descriptors, a binary file that describes the runtime format of protobufs alongside their metadata. These descriptors are used to perform transocoding at runtime. Note that your gRPC service's need to use google.api.http options to describe their mapping to a REST API.

   Google's Cloud Endpoints' documentation provides an overview of using these proto/gRPC features, as well as how generate descriptors.

   protoc \
     -I path/to/google/protobufs \
     -I path/to/your/protos \
     --descriptor_set_out=path/to/output/dir/YOUR_SERVICE_NAME.proto-descriptor \
     --include_imports \
     --go_out=plugins=grpc:. \
     path/to/your/protos/service.proto

3. Note the fully qualified name of your gRPC service's protobuf, i.e. proto.package.name.Service. You may choose to provide prefix of matching package names e.g. --packages proto.package. If none provided, all packages will be chosen, and listed.

4. You may choose to match services using comma-separated regular expressions e.g. --services http.*,echo.*. If none provided, all services will be chosen and listed.

5. Find the app label of the Kubernetes Service you want to enable transcoding for. For our example, we'll assume our Kubernetes uses the label app: foo.

6. Note the port your gRPC server is running on; for our example we'll assume the gRPC server listens on port 9080.

7. Use gen-transcoder to generate your configuration for Istio:

   gen-transcoder \
     --port 9080 \
     [--service foo] \
     [--packages proto.package.name] \
     [--services 'Service.*'] \
     --descriptor path/to/output/dir/YOUR_SERVICE_NAME.proto-descriptor

   Which will spit out config looking like:

   # Created by github.com/tetratelabs/istio-tools/grpc-transcoder
   apiVersion: networking.istio.io/v1alpha3
   kind: EnvoyFilter
   metadata:
     name: foo
   spec:
     workloadLabels:
       app: foo
     filters:
     - listenerMatch:
         portNumber: 9080 
         listenerType: SIDECAR_INBOUND
       insertPosition:
         index: BEFORE
         relativeTo: envoy.router
       filterName: envoy.grpc_json_transcoder
       filterType: HTTP
       filterConfig:
         protoDescriptorBin: <Base 64 Encoded String, the binary data inside of path/to/output/dir/YOUR_SERVICE_NAME.proto-descriptor>
         services:
         - proto.package.name.Service1
         - proto.package.name.Service2
         printOptions:
           alwaysPrintPrimitiveFields: True

---

We have included a few sample proto services, compiled into a single proto descriptor that you can use in the following way:

gen-transcoder \
  --port 9080 \
  --service echo \
  --packages proto \
  --services 'Echo.*' \
  --descriptor proto/onebig.proto-descriptor

Which spits out the config below:

# Created by github.com/tetratelabs/istio-tools/grpc-transcoder
apiVersion: networking.istio.io/v1alpha3
kind: EnvoyFilter
metadata:
  name: echo
spec:
  workloadLabels:
    app: echo
  filters:
  - listenerMatch:
      portNumber: 9080
      listenerType: SIDECAR_INBOUND
    insertPosition:
      index: BEFORE
      relativeTo: envoy.router
    filterName: envoy.grpc_json_transcoder
    filterType: HTTP
    filterConfig:
      services:
      - proto.EchoService
      protoDescriptorBin: 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
      printOptions:
        alwaysPrintPrimitiveFields: True

TODO: full example including protos, k8s service + deployment definition showing full e2e setup