vasync: observable asynchronous control flow

This module provides several functions for asynchronous control flow. There are many modules that do this already (notably async.js). This one's claim to fame is improved debuggability.

Observability is important

Working with Node's asynchronous, callback-based model is much easier with a handful of simple control-flow abstractions, like:

• waterfalls and pipelines (which invoke a list of asynchronous callbacks sequentially)
• parallel pipelines (which invoke a list of asynchronous callbacks in parallel and invoke a top-level callback when the last one completes).
• queues
• barriers

But these structures also introduce new types of programming errors: failing to invoke the callback can cause the program to hang, and inadvertently invoking it twice can cause all kinds of mayhem that's very difficult to debug.

The functions in this module keep track of what's going on so that you can figure out what happened when your program goes wrong. They generally return an object describing details of the current state. If your program goes wrong, you have several ways of getting at this state:

• On illumos-based systems, use MDB to find the status object and then print it out.
• Provide an HTTP API (or AMQP, or whatever) that returns these pending status objects as JSON (see kang).
• Incorporate a REPL into your program and print out the status object.
• Use the Node debugger to print out the status object.

Functions

• parallel: invoke N functions in parallel (and merge the results)
• forEachParallel: invoke the same function on N inputs in parallel
• pipeline: invoke N functions in series (and stop on failure)
• tryEach: invoke N functions in series (and stop on success)
• forEachPipeline: invoke the same function on N inputs in series (and stop on failure)
• filter/filterSeries/filterLimit: filter N inputs serially or concurrently
• whilst: invoke a function repeatedly until a stopping condition is met
• waterfall: like pipeline, but propagating results between stages
• barrier: coordinate multiple concurrent operations
• queue/queuev: fixed-size worker queue

parallel: invoke N functions in parallel

Synopsis: parallel(args, callback)

This function takes a list of input functions (specified by the "funcs" property of "args") and runs them all. These input functions are expected to be asynchronous: they get a "callback" argument and should invoke it as callback(err, result). The error and result will be saved and made available to the original caller when all of these functions complete.

This function returns the same "result" object it passes to the callback, and you can use the fields in this object to debug or observe progress:

• operations: array corresponding to the input functions, with
  • func: input function,
  • status: "pending", "ok", or "fail",
  • err: returned "err" value, if any, and
  • result: returned "result" value, if any
• successes: "result" field for each of "operations" where "status" == "ok" (in no particular order)
• ndone: number of input operations that have completed
• nerrors: number of input operations that have failed

This status object lets you see in a debugger exactly which functions have completed, what they returned, and which ones are outstanding.

All errors are combined into a single "err" parameter to the final callback (see below).

Example usage:

console.log(mod_vasync.parallel({
    'funcs': [
        function f1 (callback) { mod_dns.resolve('joyent.com', callback); },
        function f2 (callback) { mod_dns.resolve('github.com', callback); },
        function f3 (callback) { mod_dns.resolve('asdfaqsdfj.com', callback); }
    ]
}, function (err, results) {
        console.log('error: %s', err.message);
        console.log('results: %s', mod_util.inspect(results, null, 3));
}));

In the first tick, this outputs:

status: { operations:
   [ { func: [Function: f1], status: 'pending' },
     { func: [Function: f2], status: 'pending' },
     { func: [Function: f3], status: 'pending' } ],
  successes: [],
  ndone: 0,
  nerrors: 0 }

showing that there are three operations pending and none has yet been started. When the program finishes, it outputs this error:

error: first of 1 error: queryA ENOTFOUND

which encapsulates all of the intermediate failures. This model allows you to write the final callback like you normally would:

if (err)
  return (callback(err));

and still propagate useful information to callers that don't deal with multiple errors (i.e. most callers).

The example also prints out the detailed final status, including all of the errors and return values:

results: { operations:
   [ { func: [Function: f1],
       funcname: 'f1',
       status: 'ok',
       err: null,
       result: [ '165.225.132.33' ] },
     { func: [Function: f2],
       funcname: 'f2',
       status: 'ok',
       err: null,
       result: [ '207.97.227.239' ] },
     { func: [Function: f3],
       funcname: 'f3',
       status: 'fail',
       err: { [Error: queryA ENOTFOUND] code: 'ENOTFOUND',
          errno: 'ENOTFOUND', syscall: 'queryA' },
       result: undefined } ],
  successes: [ [ '165.225.132.33' ], [ '207.97.227.239' ] ],
  ndone: 3,
  nerrors: 1 }

You can use this if you want to handle all of the errors individually or to get at all of the individual return values.

Note that "successes" is provided as a convenience and the order of items in that array may not correspond to the order of the inputs. To consume output in an ordered manner, you should iterate over "operations" and pick out the result from each item.

forEachParallel: invoke the same function on N inputs in parallel

Synopsis: forEachParallel(args, callback)

This function is exactly like parallel, except that the input is specified as a single function ("func") and a list of inputs ("inputs"). The function is invoked on each input in parallel.

This example is exactly equivalent to the one above:

console.log(mod_vasync.forEachParallel({
    'func': mod_dns.resolve,
    'inputs': [ 'joyent.com', 'github.com', 'asdfaqsdfj.com' ]
}, function (err, results) {
    console.log('error: %s', err.message);
    console.log('results: %s', mod_util.inspect(results, null, 3));
}));

pipeline: invoke N functions in series (and stop on failure)

Synopsis: pipeline(args, callback)

The named arguments (that go inside args) are:

• funcs: input functions, to be invoked in series
• arg: arbitrary argument that will be passed to each function

The functions are invoked in order as func(arg, callback), where "arg" is the user-supplied argument from "args" and "callback" should be invoked in the usual way. If any function emits an error, the whole pipeline stops.

The return value and the arguments to the final callback are exactly the same as for parallel. The error object for the final callback is just the error returned by whatever pipeline function failed (if any).

This example is similar to the one above, except that it runs the steps in sequence and stops early because pipeline stops on the first error:

console.log(mod_vasync.pipeline({
    'funcs': [
        function f1 (_, callback) { mod_fs.stat('/tmp', callback); },
        function f2 (_, callback) { mod_fs.stat('/noexist', callback); },
        function f3 (_, callback) { mod_fs.stat('/var', callback); }
    ]
}, function (err, results) {
        console.log('error: %s', err.message);
        console.log('results: %s', mod_util.inspect(results, null, 3));
}));

As a result, the status after the first tick looks like this:

{ operations:
   [ { func: [Function: f1], status: 'pending' },
     { func: [Function: f2], status: 'waiting' },
     { func: [Function: f3], status: 'waiting' } ],
  successes: [],
  ndone: 0,
  nerrors: 0 }

Note that the second and third stages are now "waiting", rather than "pending" in the parallel case. The error and complete result look just like the parallel case.

tryEach: invoke N functions in series (and stop on success)

Synopsis: tryEach(funcs, callback)

The tryEach function invokes each of the asynchronous functions in funcs serially. Each function takes a single argument: an interstitial-callback. tryEach will keep calling the functions until one of them succeeds (or they all fail). At the end, the terminating-callback is invoked with the error and/or results provided by the last function that was called (either the last one that failed or the first one that succeeded).

This example is similar to the one above, except that it runs the steps in sequence and stops early because tryEach stops on the first success:

console.log(mod_vasync.tryEach([
        function f1 (callback) { mod_fs.stat('/notreal', callback); },
        function f2 (callback) { mod_fs.stat('/noexist', callback); },
        function f3 (callback) { mod_fs.stat('/var', callback); },
        function f4 (callback) { mod_fs.stat('/noexist', callback); }
    ],
    function (err, results) {
            console.log('error: %s', err);
            console.log('results: %s', mod_util.inspect(results));
}));

The above code will stop when it finishes f3, and we will only print a single result and no errors:

error: null
results: { dev: 65760,
  mode: 16877,
  nlink: 41,
  uid: 0,
  gid: 3,
  rdev: -1,
  blksize: 2560,
  ino: 11,
  size: 41,
  blocks: 7,
  atime: Thu May 28 2015 16:21:25 GMT+0000 (UTC),
  mtime: Thu Jan 21 2016 22:08:50 GMT+0000 (UTC),
  ctime: Thu Jan 21 2016 22:08:50 GMT+0000 (UTC) }

If we comment out f3, we get the following output:

error: Error: ENOENT, stat '/noexist'
results: undefined

Note that: there is a mismatch (inherited from async) between the semantics of the interstitial callback and the sematics of the terminating callback. See the following example:

console.log(mod_vasync.tryEach([
        function f1 (callback) { callback(new Error()); },
        function f2 (callback) { callback(new Error()); },
        function f3 (callback) { callback(null, 1, 2, 3); },
        function f4 (callback) { callback(null, 1); }
    ],
    function (err, results) {
            console.log('error: %s', err);
            console.log('results: %s', mod_util.inspect(results));
}));

We pass one or more results to the terminating-callback via the interstitial-callback's arglist -- (err, res1, res2, ...). From the callback-implementor's perspective, the results get wrapped up in an array (err, [res1, res2, ...]) -- unless there is only one result, which simply gets passed through as the terminating callback's second argument. This means that when we call the callback in f3 above, the terminating callback receives the list [1, 2, 3] as its second argument. If, we comment out f3, we will end up calling the callback in f4 which will end up invoking the terminating callback with a single result: 1, instead of [1].

In short, be mindful that there is not always a 1:1 correspondence between the terminating callback that you define, and the interstitial callback that gets called from the function.

forEachPipeline: invoke the same function on N inputs in series (and stop on failure)

Synopsis: forEachPipeline(args, callback)

This function is exactly like pipeline, except that the input is specified as a single function ("func") and a list of inputs ("inputs"). The function is invoked on each input in series.

This example is exactly equivalent to the one above:

console.log(mod_vasync.forEachPipeline({
    'func': mod_dns.resolve,
    'inputs': [ 'joyent.com', 'github.com', 'asdfaqsdfj.com' ]
}, function (err, results) {
    console.log('error: %s', err.message);
    console.log('results: %s', mod_util.inspect(results, null, 3));
}));

waterfall: invoke N functions in series, stop on failure, and propagate results

Synopsis: waterfall(funcs, callback)

This function works like pipeline except for argument passing.

Each function is passed any values emitted by the previous function (none for the first function), followed by the callback to invoke upon completion. This callback must be invoked exactly once, regardless of success or failure. As conventional in Node, the first argument to the callback indicates an error (if non-null). Subsequent arguments are passed to the next function in the "funcs" chain.

If any function fails (i.e., calls its callback with an Error), then the remaining functions are not invoked and "callback" is invoked with the error.

The only difference between waterfall() and pipeline() are the arguments passed to each function in the chain. pipeline() always passes the same argument followed by the callback, while waterfall() passes whatever values were emitted by the previous function followed by the callback.

Here's an example:

mod_vasync.waterfall([
    function func1(callback) {
 	setImmediate(function () {
		callback(null, 37);
	});
    },
    function func2(extra, callback) {
	console.log('func2 got "%s" from func1', extra);
	callback();
    }
], function () {
	console.log('done');
});

This prints:

func2 got "37" from func1
better stop early

filter/filterLimit/filterSeries: filter N inputs serially or concurrently

Synopsis: filter(inputs, filterFunc, callback)

Synopsis: filterSeries(inputs, filterFunc, callback)

Synopsis: filterLimit(inputs, limit, filterFunc, callback)

These functions take an array (of anything) and a function to call on each element of the array. The function must callback with a true or false value as the second argument or an error object as the first argument. False values will result in the element being filtered out of the results array. An error object passed as the first argument will cause the filter function to stop processing new elements and callback to the caller with the error immediately. Original input array order is maintained.

filter and filterSeries are analogous to calling filterLimit with a limit of Infinity and 1 respectively.

var inputs = [
    'joyent.com',
    'github.com',
    'asdfaqsdfj.com'
];
function filterFunc(input, cb) {
    mod_dns.resolve(input, function (err, results) {
        if (err) {
            cb(null, false);
        } else {
            cb(null, true);
        }
    }
}
mod_vasync.filter(inputs, filterFunc, function (err, results) {
    // err => undefined
    // results => ['joyent.com', 'github.com']
});

whilst: invoke a function repeatedly until a stopping condition is met

Synopsis: whilst(testFunc, iterateFunc, callback)

Repeatedly invoke iterateFunc while testFunc returns a true value. iterateFunc is an asychronous function that must call its callback (the first and only argument given to it) when it is finished with an optional error object as the first argument, and any other arbitrary arguments. If an error object is given as the first argument, whilst will finish and call callback with the error object. testFunc is a synchronous function that must return a value - if the value resolves to true whilst will invoke iterateFunc, if it resolves to false whilst will finish and invoke callback with the last set of arguments iterateFunc called back with.

whilst also returns an object suitable for introspecting the current state of the specific whilst invocation which contains the following properties:

• finished: boolean if this invocation has finished or is in progress
• iterations: number of iterations performed (calls to iterateFunc)

Compatible with async.whilst

var n = 0;

var w = mod_vasync.whilst(
    function testFunc() {
        return (n < 5);
    },
    function iterateFunc(cb) {
        n++;
        cb(null, {n: n});
    },
    function whilstDone(err, arg) {
        // err => undefined
        // arg => {n: 5}
        // w => {finished: true, iterations: 5}
    }
);

// w => {finished: false, iterations: 0}

barrier: coordinate multiple concurrent operations

Synopsis: barrier([args])

Returns a new barrier object. Like parallel, barriers are useful for coordinating several concurrent operations, but instead of specifying a list of functions to invoke, you just say how many (and optionally which ones) are outstanding, and this object emits 'drain' when they've all completed. This is syntactically lighter-weight, and more flexible.

• Methods:

  • start(name): Indicates that the named operation began. The name must not match an operation which is already ongoing.
  • done(name): Indicates that the named operation ended.

• Read-only public properties (for debugging):

  • pending: Set of pending operations. Keys are names passed to "start", and values are timestamps when the operation began.
  • recent: Array of recent completed operations. Each element is an object with a "name", "start", and "done" field. By default, 10 operations are remembered.

• Options:

  • nrecent: number of recent operations to remember (for debugging)

Example: printing sizes of files in a directory

var mod_fs = require('fs');
var mod_path = require('path');
var mod_vasync = require('../lib/vasync');

var barrier = mod_vasync.barrier();

barrier.on('drain', function () {
  console.log('all files checked');
});

barrier.start('readdir');

mod_fs.readdir(__dirname, function (err, files) {
  barrier.done('readdir');

  if (err)
    throw (err);

  files.forEach(function (file) {
    barrier.start('stat ' + file);

    var path = mod_path.join(__dirname, file);

    mod_fs.stat(path, function (err2, stat) {
      barrier.done('stat ' + file);
      console.log('%s: %d bytes', file, stat['size']);
    });
  });
});

This emits:

barrier-readdir.js: 602 bytes
foreach-parallel.js: 358 bytes
barrier-basic.js: 552 bytes
nofail.js: 384 bytes
pipeline.js: 490 bytes
parallel.js: 481 bytes
queue-serializer.js: 441 bytes
queue-stat.js: 529 bytes
all files checked

queue/queuev: fixed-size worker queue

Synopsis: queue(worker, concurrency)

Synopsis: queuev(args)

This function returns an object that allows up to a fixed number of tasks to be dispatched at any given time. The interface is compatible with that provided by the "async" Node library, except that the returned object's fields represent a public interface you can use to introspect what's going on.

• Arguments

  • worker: a function invoked as worker(task, callback), where task is a task dispatched to this queue and callback should be invoked when the task completes.
  • concurrency: a positive integer indicating the maximum number of tasks that may be dispatched at any time. With concurrency = 1, the queue serializes all operations.

• Methods

  • push(task, [callback]): add a task (or array of tasks) to the queue, with an optional callback to be invoked when each task completes. If a list of tasks are added, the callback is invoked for each one.
  • length(): for compatibility with node-async.
  • close(): signal that no more tasks will be enqueued. Further attempts to enqueue tasks to this queue will throw. Once all pending and queued tasks are completed the object will emit the "end" event. The "end" event is the last event the queue will emit, and it will be emitted even if no tasks were ever enqueued.
  • kill(): clear enqueued tasks and implicitly close the queue. Several caveats apply when kill() is called:
    • The completion callback will not be called for items purged from the queue.
    • The drain handler is cleared (for node-async compatibility)
    • Subsequent calls to kill() or close() are no-ops.
    • As with close(), it is not legal to call push() after kill().

• Read-only public properties (for debugging):

  • concurrency: for compatibility with node-async
  • worker: worker function, as passed into "queue"/"queuev"
  • worker_name: worker function's "name" field
  • npending: the number of tasks currently being processed
  • pending: an object (not an array) describing the tasks currently being processed
  • queued: array of tasks currently queued for processing
  • closed: true when close() has been called on the queue
  • ended: true when all tasks have completed processing, and no more processing will occur
  • killed: true when kill() has been called on the queue

• Hooks (for compatibility with node-async):

  • saturated
  • empty
  • drain

• Events

  • 'end': see close()

If the tasks are themselves simple objects, then the entire queue may be serialized (as via JSON.stringify) for debugging and monitoring tools. Using the above fields, you can see what this queue is doing (worker_name), which tasks are queued, which tasks are being processed, and so on.

Example 1: Stat several files

Here's an example demonstrating the queue:

var mod_fs = require('fs');
var mod_vasync = require('../lib/vasync');

var queue;

function doneOne()
{
  console.log('task completed; queue state:\n%s\n',
      JSON.stringify(queue, null, 4));
}

queue = mod_vasync.queue(mod_fs.stat, 2);

console.log('initial queue state:\n%s\n', JSON.stringify(queue, null, 4));

queue.push('/tmp/file1', doneOne);
queue.push('/tmp/file2', doneOne);
queue.push('/tmp/file3', doneOne);
queue.push('/tmp/file4', doneOne);

console.log('all tasks dispatched:\n%s\n', JSON.stringify(queue, null, 4));

The initial queue state looks like this:

initial queue state:
{
    "nextid": 0,
    "worker_name": "anon",
    "npending": 0,
    "pending": {},
    "queued": [],
    "concurrency": 2
}

After four tasks have been pushed, we see that two of them have been dispatched and the remaining two are queued up:

all tasks pushed:
{
    "nextid": 4,
    "worker_name": "anon",
    "npending": 2,
    "pending": {
        "1": {
            "id": 1,
            "task": "/tmp/file1"
        },
        "2": {
            "id": 2,
            "task": "/tmp/file2"
        }
    },
    "queued": [
        {
            "id": 3,
            "task": "/tmp/file3"
        },
        {
            "id": 4,
            "task": "/tmp/file4"
        }
    ],
    "concurrency": 2
}

As they complete, we see tasks moving from "queued" to "pending", and completed tasks disappear:

task completed; queue state:
{
    "nextid": 4,
    "worker_name": "anon",
    "npending": 1,
    "pending": {
        "3": {
            "id": 3,
            "task": "/tmp/file3"
        }
    },
    "queued": [
        {
            "id": 4,
            "task": "/tmp/file4"
        }
    ],
    "concurrency": 2
}

When all tasks have completed, the queue state looks like it started:

task completed; queue state:
{
    "nextid": 4,
    "worker_name": "anon",
    "npending": 0,
    "pending": {},
    "queued": [],
    "concurrency": 2
}

Example 2: A simple serializer

You can use a queue with concurrency 1 and where the tasks are themselves functions to ensure that an arbitrary asynchronous function never runs concurrently with another one, no matter what each one does. Since the tasks are the actual functions to be invoked, the worker function just invokes each one:

var mod_vasync = require('../lib/vasync');

var queue = mod_vasync.queue(
    function (task, callback) { task(callback); }, 1);

queue.push(function (callback) {
  console.log('first task begins');
  setTimeout(function () {
    console.log('first task ends');
    callback();
  }, 500);
});

queue.push(function (callback) {
  console.log('second task begins');
  process.nextTick(function () {
    console.log('second task ends');
    callback();
  });
});

This example outputs:

$ node examples/queue-serializer.js
first task begins
first task ends
second task begins
second task ends