字幕表 動画を再生する
STEREN GIANNINI: Hello, hello.
Welcome.
MYLES BORINS: Hi, everyone.
Thanks for coming.
[APPLAUSE]
MYLES BORINS: It's been a long day.
It's been a long three days.
And I know Steren and myself really
appreciate all of you sticking around and coming
to our session.
So just one more time, thank you everyone.
STEREN GIANNINI: It's good to have you with us.
This is Myles, Myles Borins.
Myles is a developer advocate on Google Cloud
and we've been working together for the last year.
He also happens to be the director of the Node.js
Technical Steering Committee.
[APPLAUSE]
And of course he's a long time Node.js lover.
I think the first time you used Node.js
was for an Earth installation?
Something like that.
Nice.
MYLES BORINS: So this is Steren.
Steren is a product manager for Google Cloud Platform
on the serverless team.
And we've been working really closely on new Node.js
technologies.
Previously was involved in the Stackdriver Error Reporting
product.
And you used Node in production in 2011?
Was that like 04 or 06?
STEREN GIANNINI: 06 I think.
MYLES BORINS: 06.
Did anyone start with Node 06 in the audience?
Right on.
That was buggy.
I think that's like right when they introduced [? Util. ?]
But fun fact about Steren.
His wedding invitation was actually written in JavaScript
and it's still online.
And that means your wedding invitation will never break.
STEREN GIANNINI: Never.
The web.
All right.
So who can name these French pastry?
So it's called a millefeuille, literally
translating to thousand layers.
So it's basically a cake made of many, many, many layers.
So one of my teachers used to say
in French [SPEAKING FRENCH] Sort of translating, every problem
is millefeuille of simple things.
So you should understand every problem can be broken down
into many simple things.
And I think this is also true for software, for code,
but also for software architectures.
If you break down your big app into simple things,
it will be simpler.
Until you break them down into Microservices.
So that brings us to Microservices.
Why Microservices?
Well, maybe your app has grown too big.
Maybe it's actually hard to understand, deadlocked,
maintain, manage.
This is why Microservices can be a solution.
You break up the monolith.
You decouple those Microservices.
You make them simpler to understand and maintain.
And potentially each service will
be written in a different language,
owned by a different team, and deployed independently.
As a consequence, they will scale independently
if you've decoupled them.
The scaling is independent, too.
And by making them more fine-grained,
that will result in a better utilization of your resources.
That will result in a lower bill for you.
And lastly, Microservices pair great with Node.js
due to the asychronous nature of Node.js.
So Microservices break up the monolith,
scale independently, and pair great with Node.js.
But what is Node.js?
MYLES BORINS: So I was given this slide.
Node.js.
JavaScript, how many people like JavaScript?
Sweet.
So I don't if any of you have ever
seen this little emoji thing before, the rocket turtle.
That was the Node.js logo.
It was kind of like a joke.
Substack drew this turtle on a rocket ship
because JavaScript was not supposed to be a fast language.
It was pretty amazing how fast, both as a
runtime, but also for iteration.
Node is one of those things that just you have an idea
and you can get it done in like 10 or 15
minutes because someone's already written the module.
You just need to find it.
At its base, Node is a JavaScript runtime.
We take the V8 virtual machine that
comes in the Chrome browser, we wrap it
with libuv, which is an underlying architecture
wrapping all the system calls.
And that kind of abstracts all the different platforms
and allows developers to just write JavaScript and be
able to do compute, write servers,
and all those different things.
It allows you to leverage your front end skills.
It's event driven.
What that means is that compute is driven by events.
Something computes when something happens.
That can be an HTTP request.
That can be some sort of message coming over a socket.
It can be user input with a keyboard.
It can be some sensor if you're working with IoT.
Any form of I/O. So anytime there's
input, some sort of compute runs.
The [? adventure ?] of an architecture
plays really great with async non-blocking I/O. Multiple
things can run in parallel without blocking each other.
When I make some sort of request over a socket,
I'm able to suspend, let other things execute,
and resume when I'm done without requiring the developer
to program any of this.
It just happens.
Many events execute without blocking each other.
And this ends up being really great for web servers
because it means multiple people can
connect to the same process at the same time
and it just works.
You don't have to worry about mutex.
You don't have to worry about timing.
You just worry about handling the events.
So to sum that up, Node.js is JavaScript runtime
that's event driven with async non-blocking I/O.
STEREN GIANNINI: And the last term
that we need to define from the title of this talk
is serverless.
So who here is using a serverless product?
So we should agree that there are a lot of definitions
for serverless.
I'm going to highlight here the key characteristics
that we use along this talk.
The first one that, to me, is the main characteristic
is that you pay for what you use.
So it's a billing model.
If you don't use it, you don't pay,
and then the bill is proportional to your usage.
Of course you don't need to provision in advance
and your resource, all of this, is done on demand.
So why are we able to achieve that billing model?
Because of an operational model.
So a serverless product will scale with usage
without having to worry about it.
It grows with you from a single developer to a company
with thousands of developers.
It follows your usage.
And why can it scale?
Because of a programming model.
The cloud providers takes care of the rest.
You focus on your code.
That means that no infrastructure management,
no VM management.
Everything is abstracted from you.
So you focus on your code.
That also means that you focus on delivering business value
and value to your users.
So to recap, serverless, pay for what you use,
it scales and allows you to focus on code.
These are the principles we have to use along that talk
and you will see it's not actually not only for compute.
MYLES BORINS: So Google Cloud has
a number of different serverless offerings.
We have Cloud Functions.
How many people here have been using Node in Cloud Functions
so far?
Cool.
We've got App Engine Standard, we've
got Data Store, Cloud storage, BigQuery Pub/Sub, Dataflow,
ML Engine, the Firebase suite, as well as AppScripts.
And all of these different products,
what defines them as serverless and makes them all serverless
is the fact that you don't have to manage them.
We will scale them for you.
We will manage them for you.
You can utilize them as much as you need it.
STEREN GIANNINI: And today, we are
very proud to share with you the preview of Node.js on the App
Engine standard environment.
[APPLAUSE]
STEREN GIANNINI: Thank you.
So yes, we are adding a new language to App Engine's
standard, Node.js.
We will release it in a few weeks.
So let's see a small demo of Node.js on App Engine standard.
Myles is going to tell you more about it.
MYLES BORINS: So this package json file
is outlining a hello world application.
We have a single start script that calls node on server.js.
Which fun little fact, if you don't define the start script
and you just run npm start, that's actually
the default command it will just do,
which can be really confusing if you're not
expecting it to do that.
And it has a single dependency, which is Express.
Here's our Express app.
We're making an app using Express.
We have a single route.
And we celebrate Node.js on App Engine standard.
We listen on an environment variable for the port
or fall back to 80-84 local development.
And there's one more thing that we'll
need to get this onto App Engine standard.
It's just a tiny bit of YAML.
This one line of YAML runtime Node.js 8
is the only extra thing that you need
to add to the majority of Node.js apps.
And it will work as soon as you run a single command.
And that command is gcloud app deploy.
So Steren has something loaded up on his computer.
So we're going to switch over to it right now.
And your going to see Steren's in the Cloud Shell.
He's got that app that we were just showing you.
And he's going to type in that command
that I just mentioned, gcloud app deploy.
So this is how you know it's a real demo.
But we haven't actually deployed yet, we're ready to deploy now.
STEREN GIANNINI: Ready to go?
MYLES BORINS: You're ready?
STEREN GIANNINI: Let's do it.
MYLES BORINS: So we're starting to deploy.
And in the time it takes to deploy,
I'm going to try to list all the things I like about it.
So it's an idiomatic node runtime environment.
So that means we're using the exact version of Node that
gets published by Node.js.
There's no special runtimes, there are no special APIs,
it's the same app you can run in other places.
It scales to zero.
So when you're not using it, you're not paying for it.
It's a managed runtime.
So that means everything but your code
and the node modules that you're using will get updated by us.
If there is a security fix that needs to land into Node
or open SSL-- and trust me, there
are-- we're going to patch that for you
without you or your users really ever noticing.
Oh, wow.
I couldn't even get to it.
It's already deployed.
You also get npm install in the Cloud.
So that also did npm install.
You get hybrid monitoring for free and free and free HTTPS.
Wow, that was fast.
If we can go, we can see the service.
Right now it's already running.
STEREN GIANNINI: Sorry.
MYLES BORINS: The bar was hiding it.
So there you go.
This service has been deployed.
We're already running it.
And this is, in my opinion, it's a really great user experience.
Being able to deploy, have it up in the Cloud so quickly,
and be able to share that with people to take a look.
So now that we're done with the demo,
we're going to talk a little bit more.
STEREN GIANNINI: So this was Node.js on App Engine
standard environment.
It is idiomatic, which means that you
can use any module from the npm registry that you want.
There is no API or language restriction.
As you've seen, it deploys fast and scales instantly.
And as Myles said, it is secure.
So you get one-click HTTPS for your custom domains,
and [INAUDIBLE] updates.
So I will now do a deep dive into three updates
pieces of our infrastructure.
The first one that is what we call internally
faster than light builds.
So what's that?
When you type gcloud app deploy, we
will first only upload the files that have actually
changed since last time.
This is an optimization for the upload process.
The faster than light build arrives after.
So what we do is we look at your package.json
and package-lock.json to install your dependencies in the Cloud.
So we do npm install for you in the cloud.
Good.
So this is what we defined as the speed of light.
But actually, if we notice that your package.json
and your dependencies haven't changed, well,
what we are going to do is just give you
what we just gave you before if you do a new deployment.
So we use our caching system to give user cache to modules.
And that's how we can go faster than light.
This was the first piece of data infrastructure
for this new runtime.
The second one is a new runtime environment.
So let's hear a bit of the technical stack.
So first, you gave us your code when you did gcloud app deploy.
So this is your application code.
As I said, we installed the Node modules for you.
So including the native modules, they can be any modules.
Good.
Under that, of course, there is Node.js.
And here, this is an off-the-shelf Node.js.
There's no special customization here.
Node is relying on the OS packages.
And for example, you might have heard about Headless Chrome.
And yes, all of the packages needed to run Headless Chrome
are present in the App Engine image.
And that lives on top of a regular OS, which is Ubuntu.
The last three pieces--
Node, OS packages, and Ubuntu--
are automatically updated without you
having to worry about it.
So you'll automatically update your Node.js version,
the minor and patch version of the Node.js.
The OS package and the OS is also automatically updated.
All of that lives in a secured sandbox.
And this is a new sandbox technology
that we are introducing with this new runtime.
Under the hood, the sandbox is based on G Vysor.
So G Vysor is a content runtime sandbox
that we open sourced last week.
Check it out.
This is our new runtime environment.
And the last piece of infrastructure
I would like to highlight is automatic scaling.
So what is automatic scaling?
Let's take an example.
So App Engine constantly monitors the instance capacity
and traffic, your application capacity and traffic.
And it ensures that the appropriate number of instances
are available to run your application.
So what does that mean?
Let's take an example.
So as I said, if there is no request arriving
to your application, there is no instance scheduled.
And when I say instance, it's not a VM instance,
it's an abstract that we named instance.
OK, now some traffic arrives to your application.
Well, we will scale up as fast as possible one instance.
Here I'm going to pause for a second.
I want to highlight one concept that is
very important for developers.
The very first of these requests really
hits what we call a cold start.
That means that this request will actually hit the instance
while it is starting up.
And this usually results in a higher latency than usual.
Of course we try to not retry, but we do boot up
the instance very, very fast.
But you also have a responsibility here.
You have to ensure that your start command is fast too.
So in the start script of your package [INAUDIBLE],,
make sure to not do things like build
or a lot of initialization.
Do the minimum required to start your application,
because this will have an impact on the cold start
of your application.
OK, so here as I said, the very first request
will hit this cold start.
But the next ones, everything is fine.
But then your traffic will start to increase a little bit.
And here, before this first instance arrives at capacity,
we will anticipate that you need a second one.
And we will actually schedule a second one.
So that when you actually need it, it's there.
Doesn't need to scale more.
And now let's imagine your app is featured on a popular blog
post and you have a huge spike of traffic.
Well, don't worry.
App Engine scales very fast.
So it would be to scale up instances as much as needed.
You control, you can tell us a maximum of number of instances
if that matters to you.
And of course, when things go down,
well we will ramp down the number of instances to zero.
So that's what we call automatic scaling.
You did not have to do anything to get this behavior.
But this is not new at all.
App Engine has been doing that for 10 years, 10 years now.
What is new is that actually we updated our automatic scaling
algorithm.
So the updates include it is more predictable.
As a developer, you will better understand how it behaves.
When you look at your instance chart,
you will better understand it.
Second, it actually improves the instance utilization
and performance.
So we measured a reduction of 30% of those cold starts.
So that means that the detail, the latency
of all of our apps across all of App Engine has went down.
And also better utilization means lower bills
for everybody using App Engine.
And this was released in the last months.
The third point are new settings that we
are introducing to better control this automatic scaling
algorithms.
Those new settings help you balance
between cost and performance.
Also, as I said, help you cap the number of instances
so that you can cap your bill but also the number of database
connections, for example.
So this was three pieces of infrastructure we updated.
Up to you, Myles.
MYLES BORINS: So we built a product
with Node.js comprised of Microservices
to share with you today.
And Steren's going to demo it on the machine for us right now.
STEREN GIANNINI: All right, so go for it.
MYLES BORINS: So we created a web page watcher.
And we're able to put an address into the submission bar
and it will start watching it.
And at five-minute increments, it's
going to use Headless Chrome to take a screen
capture of the website.
And it's going to compare it to other screen captures.
And it's going to only take screen captures
when there's actually been a difference in the page.
This website allows you to set any number
of different websites to watch.
And this has been a lot of fun to use on the Google website
to see the changes to the Google header.
You could use it to watch your own site to make sure
that things aren't going down if you have a personal page that
isn't changing a lot.
But it was a way for us to experiment with breaking things
into different services, and put together
a really great example.
So Steren, did you have anything you wanted to add to that?
STEREN GIANNINI: I want to tell you how it's done.
MYLES BORINS: That sounds great.
STEREN GIANNINI: Let's get back to the slides.
So this is our product.
This is our application the website watcher.
How did we build it?
Well, every five minutes there is
a cron job that's we trigger of a first Microservices.
This Microservice is called Task Scheduler.
The job of this Microservice is to ask our database for a list
of web pages to screenshot.
When it receives this list, it will actually [? enqueue ?]
into a Pub/Sub topic a message for each page to screenshot.
This is the first Microservice.
Then Pub/Sub will dispatch, will push to a second
Microservice, the web pages.
And the job of this Microservice,
is it has one job, is to take screenshots.
So as I said previously, this is using
Headless Chrome under the hood, [INAUDIBLE]
to take the screenshots.
So it opens a web page, takes a screenshot,
saves it into Google Cloud Storage.
That's its only job.
Now a new file arrives into Cloud Storage.
This will send an event that will trigger a Google Cloud
function that will compare this new screenshot
with the last known reference screenshot for that web page.
We call it image diff because the job of that function
is to compare two images.
And if there is a difference between the images,
it will store the new image into a new folder in Google Cloud
Storage.
So this is how we continuously get
the updates of all of the web pages that we want to track.
OK now, a user arrives onto a website.
So an HTP request arrives onto our web Frontend,
another Microcservice, that will query our database asking
for a list of web pages.
You've seen that list.
And when I see the detail of one of those web pages,
we actually display the screenshots.
So it also queries Cloud Storage.
So that's how we build our products.
Four custom Microservices all communicating with each other.
And now, Myles is going to tell you more about the code.
MYLES BORINS: So the first Microservice that we're showing
is our screenshot server.
You're going to notice we're using async functions
because App Engine standard comes
with Node 8 out of the box.
We're going to start a browser session by awaiting Puppeteer
to launch.
We're going to then await a new page from the browser object.
We're going to await surfing to that URL.
We'll await setting the viewport to the size of the image
that we want to take.
And then we'll await a screenshot in the image buffer.
We then await closing the browser.
And then we return the image buffer.
We're doing a lot of asynchronous work in here
that is going to allow us to use the same service to pull
lots of different images at the exact same time
without blocking.
We're able to do it in a single linear fashion
without having to manage a whole bunch of callbacks
or any sorts of promises.
Although under the hood, there's lots
of different promises going on.
Now, this is deployed using App Engine standard.
And there were some other talks going
on this week about Headless Chrome.
And you can find those if you go and check
the I/O webpage later.
So this is the part of the service that
actually saves to a bucket.
Similarly we're using Async/Await
to be able to do this in a linear fashion.
But we go to the bucket, we rename
the images to be based on a time stamp,
and we save it, and await that information, and then log it.
And because all of the Node.js client libraries support
promises, you're going to be able to use our various client
libraries to talk to all of our various servers
using Async/Await right out of the box.
And then the last thing that we wanted
to show you was the image diff from our Cloud Function
that's doing the image diff.
Every time a new file gets added to a Cloud Storage bucket,
this function's automatically called.
And what we wanted to just show is how quickly
from the event that the function is executed with
we can grab the file and the context at which
of that file inside of Cloud Storage.
So we don't even need to write any logic
to tie the two together.
We just use one service to save it,
and this new service just reacts.
And one thing that's really great about all
these different services is that out of the box,
if you start logging, Stackdriver's
going to grab all of those logs and you're
going to be able to use Stackdriver logging to look
at all the logs of what's going on.
And I don't know about you, but I never write a service right
at the first sign.
I write it locally, and then it works differently in the cloud.
Being able to just log and have it working is really awesome.
For the demo that we put together,
you'll see online later we used Bunyan
and have a pretty robust logging system
that you'll be able to take a look at.
STEREN GIANNINI: So let's look at the Cloud Console
to see those Microservices in the console.
So as I said, we do have three App Engine services.
So you will see them on the App Engine dashboard.
So opening my services-- or we have four because I just
deployed one earlier with you.
So defaults, screenshot, task editor.
We have one cloud function.
So if I open it,
it will display here.
Image diff, as you see, the trigger
is a Cloud Storage bucket trigger.
Now, as Myles said, out of the box all of your logs
are going to be collected into Stackdriver logging.
So here, let me filter out some of those logs
and only show the application logs that are here.
So these are the logs for our app across all of our services.
So as you see, we see we loaded the website.
We save screenshots so all of this
gets logged, and with a different severity.
Here we had an error, apparently.
And the last thing I wanted to show
is that when your logs go to Stackdriver,
they are going to be collected and grouped together
by Stackdriver Error Reporting.
So what you see here is a list of error groups.
So all of your errors, if they are the same,
are going to be grouped together in this place as one group.
So this is out of the box when you use App Engine, especially
Node.js App Engine.
That is all for the demo of the resources.
Let's go back to the slides.
And now I'm going to talk a little bit more
about how to connect Microservices.
So of course, you can directly call a service.
So that's the case for example, when
a client calls our front end.
So an HTP request arrives, that's a direct call.
It's also the case when you receive a webhook.
When you receive a webhook, that would be a direct call.
And lastly, two Microservices can call each other directly,
of course.
Another kind of communication between Microservices,
which actually can be quite convenient and useful,
is what we call asynchronous execution and events.
So if you remember, every five minutes [INAUDIBLE] services
was scheduled.
So this is one kind of scheduled services.
Now you remember any time we put a file into Cloud Storage,
something happened.
The function was triggered.
So this is another kind of communication.
We call that reactive services.
So when something happens, this is
triggered [INAUDIBLE] events.
Another way to do an asychronous execution
is to actually use a task queue.
So one service will make you a task into a queue.
And then this queue will make sure
that this task is executed by retrying, if needed,
and by dispatching at the rate that you want.
So task queues can be quite convenient for
asynchronous task execution.
And lastly, as we made in our demo,
you can use a publisher and subscriber
pattern when one service will publish
a message into a Pub/Sub topic, and one or many other services
that are subscribed to that topic,
will receive the message.
And lastly, maybe this is obvious,
but sharing storage is also a way
to communicate between services.
You can have two services sharing the same database.
That is a kind of connection.
In our case, we used Cloud Datastore and Google Cloud
Storage.
So connecting Microservices, you can directly call them,
you can use asynchronous execution and events,
and shared storage.
MYLES BORINS: So after kind of talking
about all these different types of services and products,
you may be wondering when should I use functions?
Or when should I use App Engine?
Or maybe when should I use a Kubernetes?
These decisions are things that teams need to make.
So I want to get a little bit more into functions first.
So we talked about events and async
and that's a lot of the programming model
that's used for that.
So you have different kinds of triggers.
You have Pub/Sub, you have Google Cloud Storage.
So whenever something ends up in a bucket, it triggers it.
You can have database transactions.
One that's really interesting is logs.
So you can have a function run every time
you see things in logs.
You can use that to filter certain kinds of error messages
or automate pager duty or something fun like that.
Maybe not pager duty.
But you can also use webhooks.
Any sort of HTTP can go and call in.
One of the things that's nice with Cloud Functions
is they can be really single-purpose.
It has a well-defined programming model
with a really quick way to get to the data that's
coming in from the event.
One that I really like is an emailer, for example.
So you may have a whole bunch of different services
that need to send email.
So if you have one Cloud Function that
just does all the auth and all the message
handling to whatever email service you're running,
you can make a email topic on Pub/Sub and just publish a to,
a from, a body, and a title, and an email will just go out.
And let's say you decide to change your email provider
service.
You just have to change that one Cloud Function.
And no one else needs to refactor their code.
As long as you have a defined API surface between all
your services that isn't broken, you
could change it to any language or any service you want to.
So on the other hand, we have App Engine.
App Engine, for example, HTTP services
are generally what are exposed.
Now, it's multi-purpose, though.
You could use it for front end applications.
You can use it to serve static resources.
It does not have a defined function wrapper.
So you can just take any Node application for the most part
that you've already been running and just launch it.
If you have multiple different things that
are going, multiple routes that you're handling,
database queries templating, you may
find that running something on App Engine
makes more sense for the use case than a Cloud Function.
But you can define your own compute relationship.
Card functions and App Engine work really great together.
If your team prefers thinking about things as functions,
run with that.
I know entire websites that have thousands of page views
and hours and they've written their entire infrastructure
on functions.
And there's tons of big note apps
that could just run on an App Engine and scale as necessary.
You could really break apart the problem
as it makes sense to your team as you need to go.
And they both share a lot of similarities.
They're both going to scale to zero
when you're not using them.
You're going to pay for what you use.
And under the hood, they're using the same sandbox.
They're using the same auto scaling.
There's no right or wrong.
One of the cool things about it, too,
is you can start thinking--
it almost becomes more of a difference
of how you want to think about your compute
and how you want to factor your code.
But from an OPS perspective, they're very similar.
And we also wanted to announce today
that [INAUDIBLE] is going to be coming to GCF really soon.
So you're going to be able to use all of these features
both on Cloud Functions as well as App Engine.
STEREN GIANNINI: All right, to recap,
deploying Node.js Microservices, serverless Microservices.
What have we seen today?
First, breaking up the monolith.
So as you've seen, we've built an entire product made out
of very, very simple pieces.
Each of these pieces is a few files, or even one file
for some of them.
Break up the monolith.
Microservice architectures can help you build rich products
from a de-coupled, small, scalable, maintainable,
understandable components.
Second, we showed you a preview of Node.js
on App Engine standard environment.
This is coming in a few weeks.
And lastly, what you should remember
is that Google Cloud Platform has
everything you need to build serverless Microservices using
Node.js.
And personally, I can't wait to see
what you will build with Node.js on App Engine standard
and all of those products.
[APPLAUSE]
MYLES BORINS: Thank you.
Thank you.
So if you enjoyed the session, you
can go and get your feedback here through the schedule link.
You can use web or Android or iOS, just let
us know what you think.
Here are some resources.
You'll find the docs for the App Engine standard.
You're going to see the functions and the demo
that we were showing you today is all there.
And so just a quick round of applause for everyone
who's watching at home on the live stream.
[APPLAUSE]
And one more quick round of applause.
You don't really see it, but there are some amazing people
at the back of the house and the back
here who have been working all day tirelessly.
Just a round of applause for them.
[APPLAUSE]
And so don't go yet.
STEREN GIANNINI: One more thing.
One more thing.
Look under your seat.
No joking.
Look over there, we have a stand where you can scan your badge
and we will send you invites to try tomorrow,
or as soon as we get your e-mails, Node.js on App Engine
standard environment.
So if you want to give it a try, scan your badge over there.
Thank you.
MYLES BORINS: Thank you.
[APPLAUSE]
[MUSIC PLAYING]