This repository contains a set of examples for Envoy's External Processing Filter.
In particular, "go/cmd/processor" contains sample processor, written in Go and Rust, that exercise most of the functionality of ext_proc, including:
- Reading and validating header content
- Modifying request and response headers
- Reading the request body using several different streaming modes
- Modifying the request body using the different streaming modes
In addition, "go/cmd/httptarget" contains an HTTP server that can be used as an "upstream" server, along with a sample configuration that runs Envoy, the sample target, and the processor.
Follow these instructions to run all the parts together to exercise all the functionality.
Build the HTTP target and run it in the background on port 10001:
cd go
go build ./cmd/httptarget
./httptarget -p 10001 &
You can verify this using "curl":
curl 0:10001/hello
Hello, World!
Build the external processing server and run it in the background on port 10002:
cd go
go build ./cmd/processor
./processor -p 10002 &
The rust version works in a similar way:
cd rust
cargo build --profile release
./target/release/envoy-processor -p 10002 &ls targ
You'll need an Envoy proxy binary in order to test the processor. The Envoy documentation has many options listed. You can also build Envoy from source.
Run Envoy using the supplied configuration, which will:
- Listen on port 10000
- Proxy all requests to the HTTP target at port 10001
- Pass all requests through the processor for additional processing
Assuming that "envoy" points to an Envoy executable in your environment, then run the following from this directory:
envoy -c envoy.yaml
Ensure that a standard HTTP request gets through Envoy, the processor, to the HTTP target, and back again:
curl 0:10000/hello
Hello, World!
The combination of the processor and the HTTP target supports the following API:
Return help about the supported API.
Return "Hello, World!"
Return some valid JSON.
Return back the data that was sent in the body of the POST. The processor does not process the body, so an arbitrary amount of data may be streamed.
Same as "/hello", but the processing server adds the header "x-external-processor-status" to the response.
Echo back the request body, calculate the SHA-256 hash of each body and print out each to the output of the processor.
The processor does this with the STREAMING body mode. This means that the checksum will work regardless of the size of the request and response bodies. However, we can't manipulate the headers after the body chunks have been sent, so all that we can do is print them to standard output.
Try this with a big request, like a big file, to see this, and run the processor with the "-d" (debug) flag enabled to see the messages coming in. For example:
curl 0:10000/echohashstream -X POST -T <some big file> > /dev/null
Echo back the request body, but base64-encode the response. The processor does this using STREAMING mode (and the golang base64 module also operates in streaming mode) so it will work with very large bodies.
To verify:
curl 0:10000/echoencode -d 'Base64 encode this' | base64 -d -
Same as /echo, but if the Content-Type field on the request is set to "application/json", then return an error if the posted content is not a valid JSON message.
The golang JSON processor is not capable of streaming, so for this the processor uses BUFFERED body mode. So, an error will be returned if you try to send a request body that is too large.
For example:
$ curl 0:10000/checkJson -H "Content-Type: application/json" -d 'This is not JSON'
Invalid JSON
$ curl 0:10000/checkJson -H "Content-Type: application/json" -d '{"isThisJson": true}'
{"isThisJson": true}
Turn an HTTP GET from the client into a POST to the /echo API on the server. This will respond with a generic JSON message.
Just return a 404.