Serverless-chrome: Support for Google Cloud Functions

[adieuadieu]

Add support for Google Cloud Functions with package @serverless-chrome/cloud-functions and appropriate headless chrome binaries, and include Dockerfile for building Chrome binary. Not super sure about the package name:

@serverless-chrome/lambda
@serverless-chrome/cloud-functions
@serverless-chrome/az-functions

.. a bit inconsistent.

Related: #44 — Support for Azure Functions

[alex88]

Someone tried to run chrome on GCF? I've tried to use the binary included with the repo and chrome silently exits with exit code 127, no other stdout/err

Update: after some digging there are some .so files missing, I've tried to run chrome binary on ubuntu docker image and from there I've copied these libraries:

|-- libnspr4.so
|-- libnss3.so
|-- libnssutil3.so
|-- libosmesa.so
|-- libplc4.so
|-- libplds4.so
|-- libsmime3.so
|-- libssl3.so
`-- nss
    |-- libfreebl3.chk
    |-- libfreebl3.so
    |-- libnssckbi.so
    |-- libnssdbm3.chk
    |-- libnssdbm3.so
    |-- libnsssysinit.so
    |-- libsoftokn3.chk
    `-- libsoftokn3.so

This makes chrome at least start.
The problem is, due the fact the only writable folder on GCF is /tmp I had to move chrome there along with the above libraries and set LD_LIBRARY_PATH accordingly, this makes libnss3 search for the nssdb in /tmp/.pki/nssdb.
I've tried to copy the contents of /var/lib/nssdb there but chrome still crashes with:

[0729/072216.817840:WARNING:resource_bundle.cc(338)] locale_file_path.empty() for locale
[0729/072216.822776:WARNING:resource_bundle.cc(338)] locale_file_path.empty() for locale
[0729/072216.825442:VERBOSE1:zygote_main_linux.cc(626)] ZygoteMain: initializing 0 fork delegates
[0729/072216.825727:INFO:cpu_info.cc(50)] Available number of cores: 1
[0729/072216.835971:ERROR:devtools_http_handler.cc(759)] Error writing DevTools active port to file
[0729/072216.836504:VERBOSE1:webrtc_internals.cc(106)] Could not get the download directory.
[0729/072216.841256:VERBOSE1:proxy_service.cc(958)] PAC support disabled because there is no system implementation
[0729/072216.844030:ERROR:nss_util.cc(706)] Error initializing NSS with a persistent database (sql:/tmp/.pki/nssdb): NSS error code: -8023
[0729/072216.844228:VERBOSE1:nss_util.cc(712)] Initializing NSS without a persistent database.
[0729/072216.844354:ERROR:nss_util.cc(201)] Error initializing NSS without a persistent database: NSS error code: -8023
[0729/072216.856111:WARNING:histograms.cc(40)] Started multiple compositor clients (Browser, Renderer) in one process. Some metrics will be disabled.
[0729/072216.844795:FATAL:nss_util.cc(203)] nss_error=-8023, os_error=0
...very long stacktrace...

If someone is able to make some progress please ping me since I'll have to make this work somehow :D

[adieuadieu]

Hi @alex88 the Lambda binary isn't compiled with the GCF execution environment in mind. However, I have been working on making a GCF-specific build. If nothing unexpected comes up, I should have it working sometime this coming week. :smile:

[alex88]

Thank you @adieuadieu, you know the OS/environment GCF runs on?

[samueller]

@adieuadieu, eager to know if you were able to get the GCF-specific build working.

[rosskevin]

@adieuadieu I'm starting on this and ready to contribute. Do you have a branch published?

[PaddyMann]

Hej, we'd love to use a GCF implementation. Any progress here? What are the current blockers?

[steren]

Note that the Node.js 8 runtime for Google Cloud Functions now includes all the necessary OS packages to run Headless Chrome.

Here is a code sample of an HTTP function that returns screenshots:

Main index.js file:

const puppeteer = require('puppeteer');

exports.screenshot = async (req, res) => {
  const url = req.query.url;

  if (!url) {
    return res.send('Please provide URL as GET parameter, for example: <a href="?url=https://example.com">?url=https://example.com</a>');
  }

  const browser = await puppeteer.launch({
    args: ['--no-sandbox']
  });
  const page = await browser.newPage();
  await page.goto(url);
  const imageBuffer = await page.screenshot();
  browser.close();

  res.set('Content-Type', 'image/png');
  res.send(imageBuffer);
}

and package.json

{
  "name": "screenshot",
  "version": "0.0.1",
  "dependencies": {
    "puppeteer": "^1.2.0"
  }
}

[adieuadieu]

@steren nice! Then there's no need for us to do anything! 🎉

[gotson]

what about Java ? I'm using Selenium Java and would like to use GCP instead of AWS.

[sejka]

what about Java ? I'm using Selenium Java and would like to use GCP instead of AWS.

afair gcp functions support only js and python

[ncruces]

@steren It seems some of those libraries (libssl3) where just dropped from the Node 8 runtime. Any idea why?

[steren]

The base image has changed between the Node6 and Node8 runtimes.
As I commented, the Node8 runtime has all libraries needed to run headless chrome.

[ncruces]

Sorry @steren, this is probably not the most appropriate place to discuss this.

I'm not exactly using this package (serverless-chrome), nor puppeteer mentioned in this GCP blog.

But I am using something else (aws-lambda-libreoffice) that depended on much the same libs.

Something changed between 12 and 2pm GMT today. libssl3.so is now missing from every function I've redeployed since 3pm. This a beta and I really can't complain, but I'd still like to know what happened, can I reach the team somewhere more appropriate?

I'm in the process of testing puppeteer directly to see if it's similarly affected.

[steren]

Let's move this discussion to the Cloud Function issue tracker. Please open an issue there and ping it to me.

[ncruces]

@steren, thanks: https://issuetracker.google.com/issues/118423830

And sorry adieuadieu for hijacking this issue.

[adieuadieu]

Closing this as GCP has built-in package support for Chrome. Nothing for us to do.