AWS Lambda making video thumbnails

2019-01-30 16:37发布

问题:

I want make thumbnails from videos uploaded to S3, I know how to make it with Node.js and ffmpeg.

According to this forum post I can add libraries:

ImageMagick is the only external library that is currently provided by default, but you can include any additional dependencies in the zip file you provide when you create a Lambda function. Note that if this is a native library or executable, you will need to ensure that it runs on Amazon Linux.

But how can I put static ffmpeg binary on aws lambda?

And how can I call from Node.js this static binary (ffmpeg) with AWS Lambda?

I'm newbie with amazon AWS and Linux

Can anyone help me?

回答1:

The process as outlined by Naveen is correct, but it glosses over a detail that can be pretty painful - including the ffmpeg binary in the zip and accessing it within your lambda function.

I just went through this, it went like this:

  1. Include the ffmpeg static binary in your zipped lambda function package (I have a gulp task to copy this into the /dist every time it builds)
  2. When your function is called, move the binary to a /tmp/ dir and chmod it to give yourself access (Update Feb 2017: it's reported that this is no longer necessary, re: @loretoparisi and @allen's answers).
  3. update your PATH to include the ffmpeg executable (I used fluent-ffmpeg which lets you set two env vars to handle that more easily.

Let me know if more detail is necessary, I can update this answer.

The copy and chmod (step 2) is obviously not ideal.... would love to know if anyone's found a better way to handle this, or if this is typical for this architecture style.

(2nd Update, writing it before the first update b/c it's more relevant):

The copy + chmod step is no longer necessary, as @Allen pointed out – I'm executing ffmpeg in Lambda functions directly from /var/task/ with no trouble at this point. Be sure to chmod 755 whatever binaries before uploading them to Lambda (also as @Allen pointed out).

I'm no longer using fluent-ffmpeg to do the work. Rather, I'm updating the PATH to include the process.env['LAMBDA_TASK_ROOT'] and executing simple bash scripts.

At the top of your Lambda function:

process.env['PATH'] = process.env['PATH'] + "/" + process.env['LAMBDA_TASK_ROOT']

For an example that uses ffmpeg: lambda-pngs-to-mp4.

For a slew of useful lambda components: lambduh.

The below update left in for posterity, but no longer necessary:

UPDATE WITH MORE DETAIL:

I downloaded the static ffmpeg binary here. Amazon recommends booting up an EC2 and building a binary for your use on there, because that environment will be the same as the conditions Lambda runs on. Probably a good idea, but more work, and this static download worked for me.

I pulled only the ffmpeg binary into my project's to-be-archived /dist folder.

When you upload your zip to lambda, it lives at /var/task/. For whatever reason, I ran into access issues trying to use the binary at that location, and more issues trying to edit permissions on the file there. A quick work-around is to move the binary to /tmp/ and chmod permissions on it there.

In Node, you can run shell via a child_process. What I did looks like this:

require('child_process').exec(
  'cp /var/task/ffmpeg /tmp/.; chmod 755 /tmp/ffmpeg;',
  function (error, stdout, stderr) {
    if (error) {
      //handle error
    } else {
      console.log("stdout: " + stdout)
      console.log("stderr: " + stderr)
      //handle success
    }
  }
)

This much should give you an executable ffmpeg binary in your lambda function – but you still need to make sure it's on your $PATH.

I abandoned fluent-ffmpeg and using node to launch ffmpeg commands in favor of just launching a bash script out of node, so for me, I had to add /tmp/ to my path at the top of the lambda function:

process.env.PATH = process.env.PATH + ':/tmp/'

If you use fluent-ffmpeg, you can set the path to ffmpeg via:

process.env['FFMPEG_PATH'] = '/tmp/ffmpeg';

Somewhat related/shameless self-plug: I'm working on a set of modules to make building Lambda functions out of composable modules easier under the name Lambduh. Might save some time getting these things together. A quick example: handling this scenario with lambduh-execute would be as simple as:

promises.push(execute({
  shell: "cp /var/task/ffmpeg /tmp/.; chmod 755 /tmp/ffmpeg",
})

Where promises is an array of promises to be run.



回答2:

I created a GitHub repo that does exactly this (as well as resizes the video at the same time). Russ Matney's answer was extremely helpful to make the FFmpeg file executable.



回答3:

I am not sure what custom mode library you would use for the ffmpeg task; nevertheless the steps to accomplish that are the same.

  1. Create a separate directory for your lambda project
  2. Run npm install <package name> inside that directory ( this would automatically put in place the node_modules and appropriate files )
  3. Create index.js file in the lambda project directory then use the require(<package-name>) and perform your main task for video thumbnails creation
  4. Once you are done, you can zip the lambda project folder and upload it I'm AWS management console and configure the index file and handler.
  5. Rest of configurations follow the same process like IAM Execution Role, Trigger, Memory and Timeout specification etc.


回答4:

I got this working without moving it to /tmp. I ran chmod 755 on my executable and then it worked! I had problems when I previously set it to chmod 777.



回答5:

At the time I'm writing, as well described above there is no need anymore to copy binaries from current folder, that is the var/task or the process.env['LAMBDA_TASK_ROOT'] folder to the /tmp folder. So it is just necessary to do

chmod 755 dist/ff*

if you have your ffmpeg and ffprobe binaries there.

By the way, previously my 2 cents solution that wasted 2 days time was this

Configure : function(options, logger) {

        // default options
        this._options = {

          // Temporay files folder for caching and modified/downloaded binaries
          tempDir : '/tmp/',

          /**
           * Copy binaries to temp and fix permissions
           * default to false  - since this is not longer necessary
           * @see http://stackoverflow.com/questions/27708573/aws-lambda-making-video-thumbnails/29001078#29001078
          */
          copyBinaries : false

        };

        // override defaults
        for (var attrname in options) { this._options[attrname] = options[attrname]; }

        this.logger=logger;
        var self=this;

        // add temporary folder and task root folder to PATH
        process.env['PATH'] = process.env['PATH'] + ':/tmp/:' + process.env['LAMBDA_TASK_ROOT']

        if(self._options.copyBinaries)
        {
          var result = {}
          execute(result, {
            shell: "cp ./ffmpeg /tmp/.; chmod 755 /tmp/ffmpeg", // copies an ffmpeg binary to /tmp/ and chmods permissions to run it
            logOutput: true
          })
          .then(function(result) {
            return execute(result, {
               shell: "cp ./ffprobe /tmp/.; chmod 755 /tmp/ffprobe", // copies an ffmpeg binary to /tmp/ and chmods permissions to run it
               logOutput: true
             })
          })
          .then(function(result) {
             self.logger.info("LambdaAPIHelper.Configure done.");
          })
          .fail(function(err) {
            self.logger.error("LambdaAPIHelper.Configure: error %s",err);
          });
        } //copyBinaries

      }

helped by the good lambduh module:

// lambuh & dependencies
var Q = require('q');
var execute = require('lambduh-execute');

As described here and confirmed by module author now this can be considered not needed, by the way it's interesting to have a well understanding of the lambda runtime (the machine) environment that is well described in Exploring the Lambda Runtime environment.



回答6:

I just went through the same issues as described above and ended up moving with the same concept of moving my scripts requiring execution to the /tmp directory.

var childProcess = require("child_process");
var Q = require('q');

Code I used is below with promises:

.then(function(result) {
    console.log('Move shell ffmpeg shell script to executable state and location');
    var def = Q.defer();
    childProcess.exec("mkdir /tmp/bin; cp /var/task/bin/ffmpeg /tmp/bin/ffmpeg; chmod 755 /tmp/bin/ffmpeg",
      function (error, stdout, stderr) {
        if (error) {
          console.log("error: " + error)
        } else {
          def.resolve(result);
        }
      }
    )
    return def.promise;
  })


回答7:

In order for the binary to be directly executable on AWS Lambda (without first having to copy to /tmp and chmod), you need to ensure the binary has executable permission when it is added to the ZIP file.

This is problematic on Windows because Windows doesn't recognize Linux binaries. If you're using Windows 10, use the Ubuntu Bash shell to create the package.

I created a Node.js function template specifically for this purpose here. It allows you to deploy one or more binaries to Lambda, then execute an arbitrary shell command and capture the output.