Processing a stream of messages from a http server in clojure

Question

I am looking for an idiomatic way to do the following. I have a http server that on a particular GET request responds with a stream of messages. Now, since this message is non-terminating, when I use clj-http/get, the call just blocks forever (I am using LightTable). I would like to set up a callback or a core.async style channel to do some operation on the message as it comes in. Even writing the stream to a file would be a good first step for me. Any pointers? Here is the call:

    (require '[clj-http.client :as client])

    (def url "http://hopey.netfonds.no/tradedump.php?date=20150508&paper=AAPL.O&csv_format=txt")

    (client/get url)

The date has to be changed to today's date for the data to stream. Thanks!

Answer 1

To just write the stream to file, a simple approach is using clojure.java.io/copy (which takes an input-stream such as that returned by (:body (client/get some-url {:as :stream})) and an output stream and copies from one to the other). Something like

(ns http-stream
  (:require [clj-http.client :as client]
            [clojure.java.io :as io]))


(with-open [in-stream (:body (client/get "http://hopey.netfonds.no/tradedump.php?date=20150508&paper=AAPL.O&csv_format=txt" {:as :stream}))
            out-stream (->> "streamoutput.txt"
                        io/as-file
                        io/output-stream)]
  (io/copy in-stream out-stream))

That gave me several thousand lines of tab separated values over a couple seconds. Now, to process them with core.async at the level of lines we probably want to process the stream a bit more using a reader and a line-seq:

(ns http-stream
  (:require [clj-http.client :as client]
            [clojure.core.async :as async]
            [clojure.java.io :as io]
            [clojure.string :as str]))


(defn trades-chan
  "Open the URL as a stream of trades information. Return a channel of the trades, represented as strings."
  [dump-url]
  (let[lines (-> dump-url
                 (client/get {:as :stream})
                 :body
                 io/reader 
                 line-seq) ];;A lazy seq of each line in the stream.
    (async/to-chan lines))) ;;Return a channel which outputs the lines

;;Example: Print the first 250 lines.
(let [a (trades-chan "http://hopey.netfonds.no/tradedump.php?date=20150508&paper=AAPL.O&csv_format=txt")]
  (async/go-loop [takes 250]
                 (when (< 0 takes) 
                   (println (async/<! a))
                   (recur (dec takes)))))

Now, with this you are largely started up, but I notice that the stream always starts with a description of what the columns are

time    price   quantity    board   source  buyer   seller  initiator

and you can use that as a chance to improve a little bit. In particular, that's enough information to build a transducer for the trades-chan that can turn the trades into a more convenient format to work with, like a map. Also, we likely want a way to stop taking elements and close the connection sometime. I'm not that familiar with core.async myself but this seems to work:

(defn trades-chan
  "Open the URL as a tab-separated values stream of trades. 
  Returns a core.async channel of the trades, represented as maps.
  Closes the HTTP stream on channel close!"
  [dump-url]
  (let[stream (-> dump-url
                 (client/get {:as :stream})
                 :body)
       lines  (-> stream
                 io/reader 
                 line-seq) ;;A lazy seq of each line in the stream.
       fields (map keyword (str/split (first lines) #"\t")) ;; (:time :price :quantity ...
       transducer (map (comp #(zipmap fields %) #(str/split % #"\t")))  ;;A transducer that splits strings on tab and makes them into maps with keys from fields
       output-chan (async/chan 50 transducer)]
    (async/go-loop [my-lines (drop 1 lines)]
                   (if (async/>! output-chan (first my-lines))   ;;If we managed to put
                     (recur (rest my-lines))         ;;then the chan is not closed. Recur with the rest of the lines.
                     (.close stream)))               ;;else close the HTTP stream.
    output-chan))

Answer 2

I think user1571406's answer is reasonable and gives a good introduction to combining clj-http with core.async. However, if you do not stick to clj-http, I would like to strongly recommend the http-kit library, which is more designed for asynchronous response handling. Using http-kit, you can write your call back as follows.

user> (require '[clojure.java.io :as io]
               '[org.httpkit.client :as h])
nil

user> (def url "http://hopey.netfonds.no/tradedump.php?date=20150508&paper=AAPL.O&csv_format=txt")
#'user/url

user> (h/get url {:as :stream}
             (fn [{:keys [status body]}]
               (if (= status 200)
                 (with-open [out (io/output-stream "/tmp/output.txt")]
                   (io/copy body out)))))
#<core$promise$reify__6363@373b22df: :pending>

The last h/get function call returns immediately, and its callback fn writes the body of the response to the file /tmp/output.txt asynchronously.