Facilitates wrapping Julia functions into a remote callable API via message queues (e.g. ZMQ, RabbitMQ) and HTTP.
It can plug in to a different messaging infrastructure through an implementation of transport (AbstractTransport
) and message format (AbstractMsgFormat
).
Multiple instances of the front (HTTP API) and back (Julia methods) end can help scale an application.
Bundled with the package are implementations for:
- ZMQTransport: use ZMQ for transport
- InProcTransport: use Julia
Channel
for transport within the same process - JSONMsgFormat: JSON as message format
- SerializedMsgFormat: Julia serialization as message format
- DictMsgFormat: Julia
Dict
as message format, for use within the same process
Combined with a HTTP/Messaging frontend (like JuliaBox), it helps deploy Julia packages and code snippets as hosted, auto-scaling HTTP APIs.
Some amount of basic request filtering and pre-processing is possible by registering a pre-processor with the HTTP frontend. The pre-processor is run at the HTTP server side, where it has access to the complete request. It can examine headers and data and take decision whether to allow calling the service or respond directly and immediately. It can also rewrite the request before passing it on to the service.
A pre-processor can be used to implement features like authentication, request rewriting and such. See example below.
Create a file srvr.jl
with the following code
# Load required packages
using JuliaWebAPI
using Compat
# Define functions testfn1 and testfn2 that we shall expose
function testfn1(arg1, arg2; narg1="1", narg2="2")
return (parse(Int, arg1) * parse(Int, narg1)) + (parse(Int, arg2) * parse(Int, narg2))
end
testfn2(arg1, arg2; narg1="1", narg2="2") = testfn1(arg1, arg2; narg1=narg1, narg2=narg2)
# Expose testfn1 and testfn2 via a ZMQ listener
process([(testfn1, true), (testfn2, false)], "tcp://127.0.0.1:9999"; bind=true)
Start the server process in the background. This process will run the ZMQ listener.
julia srvr.jl &
Then, on a Julia REPL, run the following code
using JuliaWebAPI #Load package
using Logging
Logging.configure(level=INFO);
#Create the ZMQ client that talks to the ZMQ listener above
const apiclnt = APIInvoker("tcp://127.0.0.1:9999");
#Starts the HTTP server in current process
run_http(apiclnt, 8888)
Then, on your browser, navigate to http://localhost:8888/testfn1/4/5?narg1=6&narg2=4
This will return the following JSON response to your browser, which is the result of running the testfn1
function defined above:
{"data"=>44,"code"=>0}
Example of an authentication filter implemented using a pre-processor:
function auth_preproc(req::Request, res::Response)
if !validate(req)
res.status = 401
return false
end
return true
end
run_http(apiclnt, 8888, auth_preproc)
Deploying on JuliaBox takes care of most of the boilerplate code. To expose a simple fibonacci generator on JuliaBox, paste the following code as the API command:
fib(n::AbstractString) = fib(parse(Int, n));
fib(n::Int) = (n < 2) ? n : (fib(n-1) + fib(n-2));
process([(fib, false)]);
Notice that we need to specify a lot less detail on JuliaBox. JuliaBox connects the API servers to a queue, instead of the server having to listen for requests. The obvious packages are aleady imported.
The JuliaBox API command must however be concisely expressed within 512 bytes without new lines. To run larger applications, simply package up the code as a Julia package and install the package as part of the command. For an example, see the Juliaset API package.
Note: A JuliaBox deployment may or may not have this enabled, depending on how it has been configured.