Salut tout le monde, bienvenue à la fête de la fête de la podcast.
Google est un podcast sur la compétition de l'engineering et de la production de la software.
Je suis votre host, Steve McGee.
Cette fête est de nos amis et de nos taux de la France.
C'est tout pour ce qui s'est passé dans le space de la SRE.
D'une nouvelle technologie, des processus modernisés.
Et bien sûr, la partie la plus importante est la fête que nous avons faite.
Donc, bonheur et souvenirs, j'espère que ce n'est pas une stratégie.
Vous avez mis le téléphone sur le niveau.
Salut !
Salut ! Je suis très excité pour cet épisode très spécial, Steve.
C'est un épisode très spécial.
On est un peu double-dip dans ici, c'est génial.
Donc, nous allons faire un nombre de travail pour deux outils d'outre,
ce qui est, je pense, fantastique.
Je suis Steve et je suis très sûr que vous êtes Catherine.
Oui, peut-être que nous devons nous introduire,
car les gens ne nous connaissent pas.
C'est un bon point.
Nous avons deux sets d'audience, et les deux n'ont pas de quoi se défendre.
Donc, c'est parfait.
Pourquoi n'est-ce pas une première, Catherine ?
Qui êtes-vous ?
Aller !
Bonjour à tous ceux qui écoutent le podcast,
et peut-être ne pas écouter le podcast de Kubernetes,
régulièrement,
ou si vous êtes juste en train de tomber,
où vous pouvez le entendre.
Je suis un de les co-hosts de la podcast de Kubernetes de Google.
Et je fais tous les choses,
Kubernetes Cloud Native,
containeri-relérité.
Et je suis excité à notre épisode aujourd'hui,
pour parler de plateforme d'ingénierie,
ce qui va vous en un moment.
Mais, à l'abord, Steve.
Oui, je suis Steve McGee.
Je suis un de les co-hosts de la longue période.
Et je suis en dévarelle maintenant,
ce qui signifie que je vais parler de podcasts et de choses,
sur la reliant,
et de la sérénité.
Mais aussi, les choses qui sont relatives.
Donc, je suis sur un équipe,
avec Ben, en fait, notre guest,
sur une équipe de devops,
et nous avons vécu dans le espace de plateforme.
Et je t'ai hosté le podcast,
ce qui est un podcast que vous devez définir,
si vous êtes un de les podcast de Kubernetes de Kubernetes,
vous devez probablement s'inscrire à tous les temps,
pour le reste de votre vie.
C'est une bonne idée.
D'accord.
Oui.
Mais il y a un bon nombre d'objets,
je dirais, dans le espace de reliant,
et le espace de plateforme,
et le espace de Kubernetes.
Donc, comme vous le verrez aujourd'hui,
je pense que nous allons parler de...
Un de mes préférés, Kelsey,
les lines de l'high tower,
c'est comme, vous savez,
Kubernetes étant une plateforme pour plateformes.
Et ici, on a quelqu'un qui parle de plateformes,
c'est grand.
Donc,
le guest de la plus vécue, Mr. Good,
s'il vous plaît d'introduire,
qui êtes-vous?
Je suis heureux.
Mon nom est Ben Good.
Je ne m'apprécie pas de podcast.
Oh.
Je suis vraiment en train de me faire partie.
Je devrais changer ça dans le futur.
Donc, on va dire que c'est bon.
Mais je suis un architecte de solution cloud.
Ou Google,
qui est un moyen de dire
que je prends les différents produits Google,
et non, les produits Google,
et les mettre ensemble
en solution
pour que, tu sais,
on s'en ai peut-être solué les problèmes customaires
et que ça puisse être plus facile
pour tous les gens qui tentent de faire des choses dans le cloud.
Je l'ai fait pour
pas mal de années à ce point.
Je perds la track et j'ai besoin de la math.
Mais pas besoin.
Beaucoup de choses de fun.
Et je l'ai fait pour la formation de la formation.
C'est de l'air.
Ou, en fait,
pour un peu de temps, en fait.
Ouais.
À ce point.
Tu as fait ça,
tu as fait ça l'ensemble du temps, Ben.
C'est incroyable.
Oui.
C'est un truc de nouveau.
C'est un truc de nouveau.
Près à Google,
j'ai fait des opérations
pour les start-ups,
et des différents start-ups
dans l'arrière de Denver,
dans l'arrière de Boulder,
où je suis allé.
Et je faisais un plan de formation
avant,
et j'ai fait un call-up
pour le plan de formation.
Quand les start-ups sortent,
c'était des devops.
Et maintenant,
c'est le plan de formation.
Mais avec un peu de différents takes.

j'ai fait ça pour un peu de temps,
je n'ai pas pu le parler
ou se réaliser
ce qui se passe.
Je suis vraiment excité
de se faire
ce...
n'oubliez pas le podcast.
C'est ce que nous sommes là pour.
On parle de plate-formes d'engineering
et de Kubernetes.
Ce fameux
Quote de Kelsey Hightower
que vous avez mentionné, Steve,
Kubernetes est une plate-forme
pour les plantes de formation.
Est-ce que c'est vrai, Ben?
Est-ce que Kubernetes
est tout le monde qui doit construire les plate-formes?
Grâin, Kelsey ne dit pas
que c'est la seule chose qu'il faut.
Mais...
C'est
un outil
et un outil
et c'est un gros outil
quand il s'agit de la base.
Kubernetes
donne beaucoup de constructeurs
et des capacités
qui font un tout petit plus facile
pour construire les plate-formes.
Donc, à mon avis,
c'est un des outils
et des outils pour construire la plate-forme
et faire-le successful.
Mais ce n'est pas
la seule chose.
En tant que
quand j'ai dit que c'était
une plate-forme d'engineering
avant qu'il s'agit de plate-formes d'engineering,
il y avait une company
que j'ai travaillé pour
qui aurait été construite
ce qu'il y avait été appelée
le Portal
dans la langue d'aujourd'hui.
Mais c'était un moyen
d' الذي la Chambre
subsidies
qu'en fait ce qui était
dans la plate-forme
ou des pays
pour les plateformes
qu'on avait conintage
ou l'ad Strom
et rangeank

Matfield
les
container abstract de la vm, des choses.
So Kubernetes is super helpful, but it's not the only thing that you need to make a
platform.
In my two cents.
So I'd love to know more about what else goes into it these days.
I know, you know, last year, two years ago, when was it that we were all excited
about backstage as a tool for like, when I think about platform engineering, I'm
imagining teams at various companies who are taking the underlying infrastructure
and making it something that developers and other technologists throughout the
organization can consume.
So what does that look like for you today?
So Kubernetes as the base of managing the underlying compute infrastructure, but
there's so many other types of infrastructure that you have to deal with.
And then you have to deal with also, how are people interacting with all of that?
So what kinds of tools are you using to make that happen?
And does that describe what you do?
Yeah, that's a very good description of what it is.
So the way I think about it is when you're doing platform engineering, you're
taking the underlying technology, like you say, Kubernetes for compute, maybe
managed databases, you know, for your database infrastructure, maybe, maybe
self-managed databases, you know, all the things that you need to make your
applications run and run at scale.
Those are the things that you're building interfaces to and you're applying
automation to.
And I really think that platform engineering is the process that you go
through to glue all that stuff together.
So from a technology standpoint, you're seeing lots of this typical, like,
a lot of bash and a lot of scripting and a lot of terraform and a lot of
yaml and it's all stuck together with some sort of automation tool in the
background to run it and orchestrate it.
And then that interface is fungible.
I've recently been working on a project where the interface is a
document in Firestore.
It's not a fancy UI.
It's write a document in the proper, you know, document format into Firestore.
And then automation kicks off in Magic App.
And then that is the interface to it.
You can get more advanced or user-friendly with a tool like Backstage,
but it doesn't have to be that.
It's just got to be some sort of well-defined interface.
Et c'est vraiment ce qu'il vient de faire.
Et ça doit être facile pour vos utilisateurs ou vos ingénieurs
qui l'adopteront et les faire.
Oui, il y a une phrase que je sais, qui va sur le plan de la
planterie, c'est l'idée des paths de la Goude.
C'est un, vous savez, si vous allez faire ce truc, si vous
allez faire ce truc sur cette route, ne vous arrêtez pas de
le faire, parce qu'il y a beaucoup de branches que vous pouvez
faire, vous savez, il y a beaucoup de clés d'or.
Ou des mauvaises choses qui se passent.
Vous pouvez falloir dans un bon ou quelque chose.
Je ne sais pas où ça va.
Mais l'idée est, si quelqu'un a déjà pris ce path pour vous et
a laissé des brics derrière eux, disant que c'est le
moyen, juste follow ce truc, vous êtes plus
probablement à réussir.
Et souvent, c'est ce que j'ai vu.
Ça tend à être, c'est ce que vous avez dit,
que ce soit un document ou un portal ou quelque chose.
C'est vraiment juste, à l'end of the day, un formule d'abstraction.
Donc, vous voulez que les développeurs soient comme,
je veux juste aller au bout de cette route pour que je puisse
helper vraiment lesoptons d'avoir un donne.

Je suis répris.



par exemple, pour ce que tout le monde pense, c'est que c'est super facile et comme
un wizzy-wig et tout ça.
Non, le path gold
peut être aussi simple que un document
qui fait 3, 4, 5 pièces que vous avez à faire
pour accomplir un task.
Et c'est un exemple de path gold.
Vous pouvez le faire un tout le plus grand
que juste un documentaire si vous voulez.
Mais ça n'a pas de problème.
Il a juste de rencontrer
le pire d'ingénieur où ils sont et les tasks qu'ils tentent de accomplir.
Et c'est un endroit où le plan de l'ingénieur plateforme
est vraiment très compliqué, je pense,
parce que vous avez eu
so many different ways
que vous pouvez trouver de votre côté
dans cette plage.
Et donc, l'une des choses claires
que les ingénieurs plateformes
font dans une organisation
est comprendre
comment différents parts de l'organisation
ont besoin d'utiliser cette infrastructure.
Et j'aime que vous avez donné l'exemple
de utiliser un document de feu
comme l'interface
pour le travail
de faire ce que ce soit.
Parce que je pense que
backstage offert une vision
d'une interface unifiée
de ce que vous demandez
de la compétition de différents sources
et des choses comme ça.
Mais
ça peut être beaucoup plus simple que ça,
selon le cas de l'utilisation.
Juste donner un document de feu
que ils peuvent mettre le truc
et ça va.
Je l'aime.
Oui, exactement.
Et dans le cas du document de feu,
vous pouvez le faire
avec un petit CLI,
vous pouvez le faire
par backstage.
Il n'y a pas de raison
qu'il ne peut pas faire ce call
par backstage.
Il y a
millions de différentes manières
que vous pouvez obtenir
ce document en feu.
Ce n'est pas vraiment pas important
comment ça se passe
pour que c'est facile
pour les
les gens qui tentent
de utiliser ce système.
Flexible mais structureux
plateformes d'engineering.
Bien dit.
Ça ressemble à un bon tagline.
On doit le mettre là-bas.
Oui, ça doit être.
Donc
je dois mettre ma sre de la mienne maintenant.
Quand vous avez
passé
l'phase
dans votre plateforme,
fera,

méthodologie que vous utilisez,
ou que vous avez abstracté
un peu de
these inconvenient truths
about infrastructure,
don't worry about it,
just follow this path.
You're good.
You're getting the user,
the developer
through the provisioning stage
to the initial deployment stage.
It turns out there's more stages.
We're not done yet.
This is where I come in.
This is where the SREs come in.
It's like, okay, cool.
We've deployed the thing.
It mostly works.
What's next?
What do you have to do next
in terms of the life cycle?
Sometimes we call this day two operations
or day two, whatever.
Just observability.
What are all the things that you would now want to
make sure that your developers have access to
maybe through a portal,
maybe through another method
beyond just that startup process?
What happens in the operate phase
of this life cycle?
Yes.
I think if you lump all that together,
you need to provide visibility
into what's happening
with the workloads
that are running on the platform.
That takes lots of different shapes
and form factors.
If I think about the traditional
in-air quotes,
operations metrics,
what's my latencies?
What are my failure rates?
Or did my deployment succeed?
Those kinds of things,
that's a level of visibility
that you can expose back up
to those engineering teams
using this platform engineering concept.
Make it easy to get
for those developers and engineers
to see what's happening.
But I think that there's more to it,
to your point than that.
There's visibility into
is the application working
in the way that the users expect it to,
so as the features and functionality
happening there,
that's something the platform can provide up.
Cost controls can come up through there.
Visibility into different security
and compliance
regimens
or that they have to adhere to.
Those kinds of things
that need to come up into the platform.
And those are, to your point,
all day-to-things.
Great, you got the first couple releases out.
You got some users.
Now you have to keep it up and going.
And the platform can provide
those ways and those golden paths
to make that easy to see.
Yeah, sometimes they're referred to
as non-functional requirements.
Like this has nothing to do with widget selling.
But you still have to do it, sorry.
The worry that teams have
when they're not building the stuff
through a platform is like,
now that I've deployed my widget selling device,
now I have to do day two,
but it's really more like day 222.
Like it's going to take so long
to like, which regulation are we subject to
and how do you generate these artifacts
for which lawyers to look at?
Like what are you talking about?
So like, what are some patterns
that you've seen customers of ours adopt to?
How do you do this in such a way
that it's like part of the platform
and not just like a thing you have to do.
It's like a tax on your time.
How do you do this properly?
Yeah, I think that the platform
can provide those things for you.
So if we go back to the,
using Kubernetes as an example,
there's things that you can do
in your Kubernetes deployment
around making sure that the right policies
are in place, the right RBAC,
all those kinds of things are in there,
such that when you get a namespace,
that those policies just kind of come along with it.
And then that's an example
of where you're getting those things
via the platform for free.
You can do those same things
with metrics and observability and logging.
Like you can plumb those up through
such that the application teams
can go in and see their logs
without having to do any sort of extra work
or magic to make that happen.
Or log into another tool or something like that.
Like, oh, I need the credentials
for the logging app now.
Great.
Yeah.
So it's not like, those aren't super,
those are examples.
Every implementation about how you go about it
is a little bit different,
depending on the tool stack
and all those kinds of things.
But that's what the platform
can do for application teams
and those other engineers
is just like do that for them.
The term is shifting down.
You might have heard that before.
Like, those are examples
of where you can shift those responsibilities
down into the platform.
And you just get that stuff for free
because your application is running on that platform.
Observability is so important.
I often hear platform engineers
that I talk with
when we ask for feedback
on like features
that we're developing
and things like that.
One of the things that I often hear
is this kind of consideration of
well, that feature is great.
But when I bubble that up to my users,
how am I going to make sure
that they know the pieces of that
that are going to keep them
from spending a ton of money
or running up a crazy amount of compute
or something like that?
I need to make sure that it's all observable
and visible to them
through the platform
that we are building.
So...
Yeah.
I feel like another thing
that we talk about
that is kind of what you're getting
at Gazzelin is like
when you're developing a platform,
like basically,
that platform doesn't just come out of the box
in perfect form
like as Ben has described,
like has all these things.
Like you kind of have to develop it.
And I think this is a
basically per customer problem.
Like per, you know, like entity, like org.
Because like every company
is a little bit different
and has different requirements
and has different...
They're different levels of like
capability
and like just staffing even
and like regulation
and like they're all
slightly different snowflakes.
And I mean that in,
not in the like pejorative sense,
but just like they're all
very different from each other.
So one thing that I've heard
and maybe Ben,
you can back me up on this is that like
whenever a team builds a platform,
they tend to name it.
Like they give it a name
and it's sort of a pet.
And like,
and then they take care of it.
Like it is,
this platform is a pet
and you,
you love it
and you feed it
and you extend it
to exactly what you need
as a team
because it's always going to be
a little bit different.
Does that line up
with your experience?
Oh yeah.
Yeah.
Platforms are very much
bespoke things
to the engineering organization
that they serve.
Because to your point,
every engineering team
operates a different skill level,
a different layer in the stack.
You might have
different teams
that are responsible
for different types of infrastructure
that need to come together
to work on the platform together.
The one that I was,
you know,
likely referencing to earlier,
it was called Divine Spork.
That was the name
that,
that GitHub,
you know,
auto-generated for me.
I'm like,
ah,
that's kind of fun.
Like we're kind of,
you know, spoon feeding,
but it's kind of pokey.
So it's a,
it's a spork
and it's kind of,
you get things bi-magically.
So it's kind of divine.
So that's what,
that's what we called
our platform
at that company
was Divine Spork.
So yeah,
they,
they're very much bespoke.
Things,
they have names,
they have teams
that rally around them,
but yes,
you do care,
you care and feed for them
and love them
and,
and hopefully they love you back
and,
you know,
all that kind of stuff.

but yeah,
every platform is different
and I think it's a bit of an
anti-pattern
to go and look for the platform
in a box
that you can just click button,
install and like,
woo,
I have a platform.
That doesn't typically happen.
Yeah,
as we alluded to before,
Kubernetes being a platform
for platforms.
So I've actually,
I've actually heard of
backstage being a,
not a portal,
but a
system for building portals.
So it's like,
kind of the same idea
of like,
it has all these knobs,

don't use them all,
don't,
don't just like,
click install.
Like,
you got to think about
what it is
you're trying to get out of this.
And,
you know,
you know,
I think we've said in the past,

you want to,
like the abstract idea
of a platform
is just a,
a grouping of capabilities.
And so like,
what are the capabilities
that your team wants to adopt
and then deliver
through this platform,
through,
whether it's through
the Kubernetes side
or through the backstage thing
or through the observability suite
that you,
you know,
have through another company
or through the,
whatever.
Right.
There's,
there's a load of different
things that you could be,
you know,
a lot of stuff you build
yourself too.
Right.
So the,
but at the end of the day,
it's,
it's the set of capabilities
and the fact that they're
working together
in concert,
hopefully,
that makes it beneficial
to your company.
Yeah,
very,
very much so,
like the folks that install
and are successful with backstage,
they have a team
or a group of people
that are,
you know,
modifying and building
plugins for it
and,
and creating
the different templates.
Like,
it,
it isn't just something
that you get to install,
but you kind of,
you build on it,
that,
you know,
expose this,
those critical user journeys
through the portal
in a way that,
that works.
And that takes development effort
to maintenance
and,
and care and love as well.
And this customization
for different organizations
and for different types of users
within organizations.
There's a term that we used
in our notes
in preparing for this interview,
which was deployment archetypes.
And I feel like all of this
lends itself
to that term,
which I hadn't really seen before,
but I,
I kind of really like it.
So,
is that what you all think of
as deployment archetypes
is kind of making these workflows
customized for the environment
that they exist within
and the users that they're serving?
I think the,
what you want to do is,
is you have like a pattern
for ways that you deploy things.
And those patterns
are what you end up
exposing out
through the platform.
So those,
those would be the archetypes.
And you can largely group them
if we've done with Steve
as well aware of like
different reliability characteristics
built into them.
So,
I have this thing
of this shape and size.
I need to deploy it
on this type of infrastructure
with this,
this low balancer
configuration,
this database,
you know,
configuration.
And that supports,
you know,
this shape of user
across these different regions
of the world
or this form factor.
And you provide
in number of those things
in the platform.
And those become
the archetypes,
if you will.
And those can be
genericized
and we've published
some documentation
around those.
But those are the,
that's the type of thing
that you want to
expose out through the platform.
And then making it easy
to pick the right archetype
or change between archetypes
is a task
that the platform
helps you with.
And get you down that journey.
So, a good way to
think about this
for the like
the pure Kubernetes viewpoint.
A good way to think about this
is like when you're trying to
make a system
more reliable
and more robust.
Often what you do is you're like,
we have one of these things.
How about we have two, right?
Like just have a second one.
And that's great.
It's a good idea.
Like that's generally
the method that we use
to make things more robust
is that we have
two of them
instead of one
or n of them, right?
But one thing
that you could do naively
is you're like,
we have a cluster of stuff.
Let's make a second cluster.
Done.
Stamp it out.
Good to go.
Ready, set, go.
Except you put it
in the same zone.
Like you have two clusters
that are in the same zone.
And then when there's
a problem in that zone,
guess what?
You actually only had one.
You did not have two.
Too bad, bummer.
You know, you get a bad day.
So there's this idea
of failure domains.
And this will be like
familiar to a lot of
SREs out in the world.
But like you have,
they're like the
matrice-gadals of like,
you know, you have like
your app
and then you have
like your namespace
and then you have your cluster
and then you have your zone
and you have your region
and then you have like
the universe or something.
And so your failure
can happen at any point
along these things.
And so there's this paper
that came out,
I don't know, years ago
at this point called
deployment archetypes,
which is like how to
apply this understanding
of failure domains
to when you're doing deployments.
And this is where like,
you know, zonal deployments
versus regional deployments
versus global deployments
versus like multi-region
versus multi-zonal
or blah, blah, blah.
Like, oh, there's like these
very abstract ideas
of basically where to put
your app in which clusters
at which time.
And like, it's too much.
Like it's hard to understand
all of it.
And so what we've seen
people do in the past
and Ben has done a really
good job of this
with some customers
that I've seen of,
is that like,
there will just be like
a dropdown in the portal
or in that doc or something.
It's like, do you want this
to be a big one
or a little one?
You know, and like
under the covers,
it's picking one of these
and it's like making
a bunch of choices for you
so you don't make
the wrong choice.
So it's like, instead of
saying, just give me
two clusters in the same zone,
it's like, no, no, no, no.
Would you like a big one
or a small one?
And you say, a big.
And it's like, okay,
I'm going to give you
two clusters in two
different regions
and not tell you about it.
But then also,
the deployment is going to
be able to take advantage
of that
and the observability
is going to be able
to take advantage of that.
And like, you don't even
know the whole time.
It's just like,
kind of magically happens.
That's the kind of the vague
idea that I get.
But Ben, Ben has actually
done it.
So I'm curious
if that's,
if that holds true
when you when you hit,
you know, reality.
Oh, very much so.
Cool.
Yeah, it definitely holds
true.
And the thing that
you're doing
with that little drop down
is you're abstracting away
a lot of the super
nitty gritty detail
what cluster,
what subnet,
what this,
what that,
that you might have to go
and know how to do otherwise.
I think a lot of times
when folks are trying
to build a platform,
you like throw a bunch
of terraform
or throw a bunch of yaml
what folks
and say, okay,
go do the thing
with the yaml
or the terraform
or whatever your
favorite,
you know,
language is.
A lot of the things
that we've talked about
in our conversation today
are kind of evergreen
topics,
reliability,
making sure that the systems
keep working,
making sure that you
understand
how they're working,
building golden paths.
But I think a lot of that
has kind of shifted
recently due to industry trends.
I wanted to know if you've seen
any of that in your day to day
work, Ben,
if the types of golden paths
that you're building these days
are different and how
and all of that kind of stuff.
I think the underlying
technology changes a little bit,
but really
it's still the same thing.
I think that we've been doing
just with slightly different
twists or name
or a change of focus on
how do we do it
at scale,
if you will.
I think the rate of change
in the industry as a whole
is always ever increasing.
Right?

like, I think the things
that we have been doing,
we just have to do more
of those things
and maybe with different
technology,
not doing something different.
Yeah.
In the Kubernetes space,
what I've seen a lot of is
hardware accelerators.
It's a lot of the same kinds
of stuff,
but now we're enabling folks
to use different types of hardware
than we've used in the past
and for applications
that are very particular
about how they're using that hardware.
So Kubernetes has traditionally
allowed folks to
a lot of the management
of that underlying infrastructure
is handled for you
through Kubernetes
and then you add another layer
on top of it,
the platform that users
actually interact with
to further abstract that.
And now folks want
a different type of abstraction
for those types of workloads
that really need careful control
over those underlying hardware resources.
So that's where I've seen
Kubernetes
doing a lot of work recently
to adapt to new changes.
Yeah, no, that's fair.
Like, you know,
back when Kubernetes was young
and when I started using it,
it was CPU and memory.
Like, those were the things
you had to be concerned with.
Now it's,
there's networking
and to your point,
there's GPUs
and how do you go about
getting access to those things?
But yeah,
it's just,
you're beginning to add more to it,
but really,
you're still orchestrating
the access
to the underlying infrastructure
and providing more control
than you were
way back forever when,
but
it's the same process
or the same thing
you're accomplishing,
just a little bit different.
I've heard that also
customers will
have different motivations
for platforms.
Then they used to,
like sort of in the post
zero interest rate
XERP phenomenon,
there's been a lot of
consideration around
cost management, right?
And cost observability,
things like that.
And frankly,
part of the problem
with a,
you know,
kind of a naïve level
way of doing these abstraction
through platforms
is like to just hide everything.
And then like,
you get a bill
and you're like,
oh, whoops,
we did something.
And so the ability to have,
like to expose
usage patterns
and like,
oh, we accidentally,
all the databases,
like early and often,
right?
So you can actually see
in like,
somewhat real time,
like your consumption
of the database
or whatever
the expensive thing
of the month is
and be able to control that
and like make a change.
So like,
if you do a deployment
that accidentally
writes a,
you know,
10 billion times,
you should be able to,
as a developer,
for like,
notice that
and rectify that,
like all on your own.
So like self service
is one of these things
that is really,
really ubiquitous
in the,
like the marketing
of platform engineering,
but often we think of that
just in terms of like,
why won't you make
my service
for the first time?
Like that green field step,
but self service also applies
in the operational context as well.
So does this come up
like as a motivation
when customers
are talking to you
about platform engineering
or I'm making this all up?
Yeah,
like there's some customers
that approach it
from a cost control perspective.
You know,
not many approach it
from a self service standpoint.
I get a lot of customers
that look at it from a
security compliance,
governance perspective.
And those,
those are reasons
to go about it,
but that can't be
the angle that you take
in my opinion.
So we can't take the angle
of what we're going to do
platform engineering
to reduce costs.
And we're going to begin
to make our engineering teams
do something to reduce costs.
We want to improve
compliance
and governance
and those things,
but forcing
engineering teams
onto the platform
doesn't typically
yield
the desired outcome,
if you will.
There's oftentimes
reasons
why that happens
or,
you know,
you want to listen
to those engineering teams
and begin to design
your platform
in a way
that supports
those engineering teams,
but also accomplishes
the security
and compliance
and governance
and cost controls
and,
and those things
come along
when you begin to
actually do
platform engineering.
So,
the motivation
can be there,
but it can't necessarily
be the reason
that you go to engineering teams
and say,
we're going to do
platform engineering
and you're going to use
the platform
because of this
that usually doesn't
yield the best platform.
That opens up
a whole new can of worms
that we could dive into
for a whole other episode
at least,
diving into
security considerations,
regulatory compliance issues
and working all of those
into how we build
platforms
for specific users,
but I think we've covered
some really awesome stuff today
with
making sure that you're designing
platforms
to serve the users
that they need to serve
and that a lot of the underlying
concepts
are the same as they've ever been.
It's about making sure
that your infrastructure
is usable
by the teams
within the organization.
And I'm excited for all of those
platform engineers out there
listening to
do that.
Another thing that we didn't get to,
which I'm amazed Ben,
how did we not mention Dora?
I'm even wearing the shirt
under this jacket.
That's true.
Like, so, I know, we blew it.
But like being able to
track your team's
success through metrics
and being able to show
that these capabilities
are actually making an
improvement
to the lives of the developers
and to the business,
like go to dora.dev
to see what that even
would all those words mean.
But like delivering it
through a platform
is totally like a good move.
I believe you would
you would agree with me on that.
Side note, Ben and I work on the
platform engineering chapter
of the Dora survey every year.
I know he agrees with me.
It's a lot of fun.
I definitely agree.
A fun thing to think about
from a Dora perspective
in your platform engineering endeavors
is using Dora metrics
to understand your
feature velocity
and your application development
practices,
but also your platform
engineering practices
and your platform velocity,
which I think is a fun,
a fun thing to think about.
Yeah, platforms or software too.
Well, thank you Ben.
And well, we'd love to have you back
either on each of our episodes
or maybe another double episode.
Who knows?
Well, it'll be great.
You never know what's going to happen
in the future as we just learned.
This was a lot of fun.
Thank you.
I really appreciate it.
Thank you much.
You've been listening to podcast,
Google's podcast
on site reliability engineering.
Visit us on the web
at sre.google
where you can find papers,
workshops, videos,
and more about SRE.
The podcast is hosted
by Steve McGee
with contributions from Jordan Greenberg,
Florian Rathgeber,
and Matt Siegler.
The podcast is produced by Paul Gulliomino,
Sunny Schau,
and Salim Virgi.
The podcast theme is
Tell-Abot
by Javi Beltran.
Special thanks to MP English
and Jen Petoff.
You missed a page from Televod.