How should I run NodeJS from a Java application?

问题:

I'm writing a Java library, actually, a Clojure library, but for this question, what matters is that it runs on the JVM. This library needs to execute some JavaScript. I tried with Nashorn but I encounter some limitations that might be too hard to overcome. As an alternative, I want to try NodeJS.

I want my library to be self contained, to not depend on the system running NodeJS independently and thus requiring a particular deployment mechanism to place the Java and NodeJS artifacts in the right places to be picked up by the two different network servers. This approach, though, brings some issues.

I will be talking to NodeJS over HTTP but I don't want NodeJS to open a specific port. I want to find a random unused one so there's no collisions. I also want to control where the logs from NodeJS go, as to keep them with the rest of my application. Lastly, my app should be able to detect when NodeJS crashed and re-run it or report an error with information.

What's the best way to approach this? Are there any Java libraries to help manage child process in this fashion? Anything in particular I should do from the NodeJS side (I'm very new to NodeJS, I never used it before).

回答1:

My solution in the end was to use ProcessBuilder like this:

(defn create-process-builder [js-engine]
  (doto (ProcessBuilder. ["node" (:path js-engine)
                          "--port-file" (:port-file js-engine)
                          "--default-ajax-host" (:default-ajax-host js-engine)
                          "--default-ajax-port" (str (:default-ajax-port js-engine))])
    .inheritIO))

and then call start in it. inheritIO causes the output of it to go to the output of the current process which effectively merges stdout and stderr.

On top of that NodeJS opens a random port by specifying 0 as the port number and writes it to a file:

(let [app (-> (express)
              (.use (cookie-parser))
              (.get "/" (fn [_req res] (.send res "Universal JavaScript engine for server side pre-rendering single page applications.")))
              (.get "/render" render))
      server (.createServer http app)]
  (.listen server 0 (fn [] (.writeFile fs (:port-file options) (.-port (.address server)))))))))

which then is opened by the Java side (waiting for it to appear):

(defn get-port-number [js-engine]
  (or (with-timeout
        (:start-timeout js-engine)
        (loop [port-number (read-port-file js-engine)]
          (if (string/blank? port-number)
            (if (is-running? js-engine)
              (do (Thread/sleep 100)
                  (recur (read-port-file js-engine)))
              (throw (Exception. (str "While waiting for port number, process died: " (:path js-engine)))))
            port-number)))
      (throw (Exception. (str "Waited for " (:start-timeout js-engine) " for " (:path js-engine) " to start and report its port number but it timed out.")))))

回答2:

There is a pretty good answer here on how to run javascript in java. Would something like that be doable for your case? If not here are some resources:

• Random port in nodejs You could hit yet another service to find an open port during the build, or have your node app fire an http request to your java server based on the port it grabs.
• Winston is the best logging library I've found, you shouldn't have any issues logging to the same path.
• Forever and PM2 are the common node process managers which keep node running. I currently prefer forever (not sure why)

It sounds like you will be using a lot of cpu within node. If that is the case you will probably want to use the cluster module (so nodejs can utilize multiple cores). If you block the event loop (which cpu based processing will, then you will only be able to perform 1 concurrent request per forked process).

回答3:

Nashorn does have some issues that I've run into as well, such as finding information about some of their APIs (documentation leaves a lot to be desired), and the slow boot up.

What I would recommend instead: treat your server-side rendering as a service and not a child process.

In other words, you could run a Node.js instance on your internal net on say port 10015 and only allow local connections to it (you could also send logs wherever you want). Then you can make a request to the service with the pages that you want to render, such as localhost:10015/posts/ and have that Node.js app render the page inside of a headless browser (using something like Phantom or Slimer).

To keep your Node server up, you can use Forever or supervisord, and to help you gain traction faster, you could look at what the Prerender team has made: https://github.com/prerender/prerender-node

回答4:

I have been in a similar position where I had to run fortran from a python script, so here is my two cents. Run your node.js script with a terminal command in Java like this:

Runtime rt = Runtime.getRuntime();
String[] commands = {"node example.js", "args"};
Process proc = rt.exec(commands);

BufferedReader stdInput = new BufferedReader(new 
     InputStreamReader(proc.getInputStream()));

BufferedReader stdError = new BufferedReader(new 
     InputStreamReader(proc.getErrorStream()));

// read the output from the command
System.out.println("Here is the standard output of the command:\n");
String s = null;
while ((s = stdInput.readLine()) != null) {
    System.out.println(s);
}

// read any errors from the attempted command
System.out.println("Here is the standard error of the command (if any):\n");
while ((s = stdError.readLine()) != null) {
    System.out.println(s);
}

With setup you can pass in parameters to your node program and get responses. Although I am not sure what you are using your node.js program for, so not sure if this is helpful.