Anthos support for multi-cloud is generally available
cloud.google.com
cloud.google.com
Terrible blog post. Had no clue what I was being sold until you pointed it out.
You can also join clusters with Istio just for admin and network traffic, i.e. one cluster with Istio in Azure and another cluster with Istio in AWS, talking via the mesh infrastructure. https://istio.io/pt-br/docs/setup/install/multicluster/gatew...
Unfortunately not even Googlers are immune to the draw to elite signalling by drowning your copy in industry jargon without ever getting to the point.
I see this a lot on hastily-composed GitHub Readmes and "rocket-launch" startup landing pages, but it's a shame it also happens to a company whose name is globally recognized.
I don't feel like that's why we all sometimes do this (though that might be an inside view and I might really be signalling along with everyone else.)
I feel like the internal thought-process that motivates this kinds of inside-baseball writing is twofold (depending on the stage of the project):
- early on: "the only people who could possibly care about something this new are people with the exact problem it solves, who stumbled onto it when jamming google full of enough keywords that there are zero other results; they already know what this does before reading; and they just want to confirm that it isn't fatally flawed. They're deploying this project in anger. (Of course they are; if they didn't desperately need it, they'd never rely on something so early in development!)"
- later on: "We're selling this to customers. We've picked a specific target market that has specific use-cases and needs, and we try to think about things the way they think about things 100% of the time, because talking to those people is what gets us money. We don't do any inbound marketing—we aren't trying to passively educate anyone outside our target market to get them interested. We're having enough trouble capturing the pre-qualified part of our funnel with in-person conversations. Converting people who haven't even fallen into the funnel yet? Who cares! They might get converted by accident if someone who already uses us takes the time to explain our product to them; but otherwise, the only time they'll hear about us is when we reach out to slowly warm them up for a million-dollar enterprise deal."
Being an developer myself however, yeah it's difficult to to get at the technical substance of what any of this means, as there is certainly lots of vaporware out there <cough>... <watson>... <cough>..
Haha! Good one. Google practically invented elite signaling.
- Anthos GKE: managed Kubernetes across GCP, AWS, on-prem via VMware vSphere
- Anthos service mesh: Istio + SRE practices (Cloud Operations)
- Cloud Run: Knative serverless
- Anthos Config Mgmt: policy & security automation
- Marketplace: 3rd party applications for Anthos
Disclaimer: work for Google, but nowhere near an expert on Anthos. Just going off of some notes I took last week.
My understanding is that Anthos is a multi cloud platform which essentially enables you to run managed Kubernetes on-prem, in GCP, and other public clouds. You can then run other Google Cloud services, such as Cloud Run, on your own hardware or in other clouds in addition to GCP
If the former, does the service itself depend on Google being up, or could Google go bankrupt and my on-prem hardware keep running Cloud Run?
Technologically both is possible.
It allows you to manage a Kubernetes cluster from GCP that can have resurces mixed from both on-prem and google services (now you can also have pods on AWS).
For on-prem you'll need a WMWare hypervisor on your servers.
Why this is useful?
- Regulation: some workloads can be run only on-prem due to regulation (for instance, in Spain you can only process gambling data in the same region/comunidad autonoma you have the business).
- Cloud offloading of K8s workloads - you can workloads onprem off-peak and mixed loads cloud+onprem on peak times.
- Google Cloud Marketplace : you can deploy solutions from GCP Marketlplace on Anthos (that is, on your hardware, now also on other clouds).
- (now) Onprem + Multicloud workloads: Multicloud is hard. Not just having to know well how stuff works on each cloud, but also having to deal with rough interactions between clouds (even with k8s).
Disclaimer: I work for a Google cloud partner.
For me that's one of the biggest adoption barriers because unless you already have that VMware licences, you'll have to burn a lot of CapEX to run Anthos.
I hope they don't make us pay for both the bare metal license and the by-cpu license (or it is a _very_ small amount).
Insane.
However, I do want to note that having had the opportunity to sit in a number of customer meetings with very large enterprise customers and listen to our VP in conversations with their senior executive leadership, her article is intended to address concerns in a way that resonates with them. This article targets executive decision makers who need to commit enormous resources to long term strategies and it is important for them to hear in a language they understand that they will have enough flexibility to solve existing as well as evolving and shifting needs in the long run.
Many customers are trying to navigate their way to a successful modernization path that gives them the ability to run their workloads where they need to, helps them to apply modern best practices, and gives them unified management, monitoring, and compliance and auditing controls. Many of them are on prem and have VMware licenses right now, or have already begun implementing some kind of hybrid cloud strategy, or are even considering multi-cloud approaches.
For those of you looking for more practical information, we hear you. We were already preparing to launch new content directly targeted for developers to give you a direct, hands-on experience with Anthos using live clusters and a running application as part of Next.
Unfortunately due to changing conditions, we had to take a phased approach to communicating this as part of moving everything to completely digital formats. However, the good news is that you will be able to explore Anthos on Google Cloud for yourself and I'm really looking forward to sharing those details very soon.
I know of at least one company where they had to blacklist AWS and GCP on corp cards because developers were still trying to use them despite the company having signed a multi-hundred million dollar deal with Azure.
Looks like an Oracle style approach (couldn't find that famous Larry's quote that developers don't grasp the concerns what he is discussing with the C-suites), and this GCP offering is the first big one under the new GCP leadership transplanted from Oracle. This is probably the new look and feel of GCP going forward.
Hybrid cloud is definitely a thing as you scale up. For a big enterprise with a significant base load you're paying 5-8 times the cost of on prem for that in a public cloud, so being able to run your own infra for production base load but scale up to the public cloud for dev and burst is a real win.
Anthos and OpenShift are trying to create a 'PaaS-lite', or maybe Orchestration++ is a way to think about it.
Alot of the K8s vendors are trying to build a complete experience: for both operators and developers. You can use the CNCF's capabilties (K8s, Prometheus, Istio, etc) to build a container orchestration and developer experience.
(Disclaimer - work in this area). Maybe the difference is that OpenShift is quite tightly bound, whereas Anthos is more modular. The complexity is how all the elements are plugged together and the overall life-cycle.
NOTE: Weaveworks builds application platforms built from CNCF technology so I definitely have 'opinions' in this space
was expecting 1st paragraph to be 'anthos is a'
But I like this too, it's like the philosophy cult ads on the subway
the first customer they quote is a 200 yr old bank. 'in my day kubernetes was a lunch place on stone st'
And I thought I was being tinfoil-hat-y about Google's true motives in pushing Kubernetes so hard and the benefits of it being so popular. It enabled the first true cross-cloud experience, that Google just happens to be the creator of and have control over (no matter how much you hear that it's "community driven", the only project I've seen that is actually lives up to that is Rust). Yeah, Kubernetes is massively useful and I'm grateful, but that doesn't stop me from being very cynical about it -- the complexity and learning curve is a consultant's dream (disclaimer: I'm a consultant) and at this point it looks like a 5d chess move to improve their cloud offering by basically backdooring themselves into every single other cloud and the headspace of millions of developers.
Condolences to Crossplane[1]. Hyperbole yes, but I do not envy any companies that have to try and compete with Google in the current regulatory (captured) marketplace. If they didn't buy you but entered your space it's gotta be bad news on some level.
Luckily, the fact the kubernetes is open source and at least in some part community driven (i.e. the wide ecosystem of plugins) means that companies like Crossplane can still exist if they work hard to out-sell and out-market google, and there is a slim possibility that the barrier to creating a "cloud platform" being lowered so much might mean that all the little mom & pop shop VPS providers and bare metal providers will get in on it, and force the bigger guys to compete on price somewhat (we all know big enterprise is going to go straight to AWS/GCP/Azure anyway).
[EDIT] - I just realized that while I was harsh on Crossplane, they arguably just became an extremely valuable acquisition target for every other large cloud provider that wants to buy their way in. Probably not Amazon but Azure/IBM/Orcale/etc probably.
Also, if the point of multi-cloud was vendor lock-in, aren't you now GCP locked-in when you use Anthos? It's not open source and may die any time like other Google projects.
Disclaimer: I am actively contributing to Crossplane.
Correct me if I'm wrong, but Crossplane introduces CRDs that paper over the differences between the provider (ex. `OmniDatabase`), rather than just offering CRDs that correspond to a platform's offering (ex. `RDSDatabase`) though I assume you can access those too, and I don't see why Google won't do both. That's the thing about being a monopoly player in the space -- unless a smaller player is doing something so next level that you can't copy it, you can just offer underpriced mediocrity until they fold (forgive the cynicism and hyperbole).
Crossplane has a huge head start and probably the least buggy implementation of this concept, and was one of the first to innovate in this field (I saw a few talks from previous kubecons) so wishing you good luck in the future!
Redhat, Microsoft and vmware are major contributors to the product. Few of the founders work for Google anymore. Google probably has the largest influence but it is very much a community.
For example the decision to avoid donating Istio to the CNCF[0].
> That was Google's standard response about Istio for years. But people familiar with Istio's behind-the-scenes discussions said that Google had changed its mind late last year, driven by a faction within the company that believed donating Kubernetes to a foundation was a strategic mistake. When contacted by Protocol earlier this year, Google declined to clarify its plans for Istio.
Google is not completely evil, but it is not my friend, it's a corporation that isn't even a not-for-profit nor non-profit.
[0]: https://www.protocol.com/google-cloud-kurian-istio-foundatio...
I'll try to translate this into something direct.
Anthos is an umbrella term for things in the combined ecosystem of Kubernetes, Istio, Anthos Config Management, Traffic Director, and so on.
Anthos [GKE] on {GCP, AWS, Azure, On-prem} is intended to be a consistent version of this deployed thing in each place.
There is an important historical wrinkle where traditional GKE already exists, but was "just Kubernetes" and sometimes "with Istio Add-on".
I do not agree with the lack of public pricing.
tl;dr: For most people here this is GKE "but wherever you want" (this announcement is "on AWS"), including Istio and friends.
Thanks for the info. I’ve been pretty happy with the designs (K8S, Istio) and services so far (GKE, Cloud Run).
Anthos is a managed Kubernetes platform: you outsource management of the control plane to Google. If you look at the case studies and the diagrams it's for complex use cases.
Altenatives, might be something like Giant Swarm (https://www.giantswarm.io/), Platform9 or the company I work for Weaveworks (https://www.weave.works/product/enterprise-kubernetes-platfo...)
If you want to run k8s on-premise and have someone support you then Rancher is popular. Most of the specialist consulting teams will provide support: Jetstack, Fairwinds, Controlplane etc
There's lots of options out there - apologies to anyone I missed.
(If you criticize my comment, there's a decent change the marketing is gonna have a discussion with my boss about my employment. I'll try not to use as many marketing-y terms as the article.)
At a high level, if you're a solo developer or a small company, you probably don't need to understand everything going on there, many Anthos features/products are actually available a la carte, possibly under different names.
First, I’ll agree that the blog post is written in a way that might be hard to grasp for hands-on practitioners (assuming I know HN enough, since I visit it 30 times a day). I'll start with an architecture diagram: https://cloud.google.com/anthos/docs/concepts/overview
You might still find this diagram difficult to understand (unless you live with the "cloud-native" stuff day to day) so I'll break this down for you:
- Anthos GKE: This is basically the GKE (Google Kubernetes Engine) you know. But it is now capable of running clusters not just on Google Cloud. If you ever thought "ahh, X cloud doesn't have good Kubernetes support", well Google can bring GKE quality to that cloud (AWS support is now GA, Azure is in preview) or to your on-prem datacenter, or AWS/Azure account. The GKE you know is still available with pay-as-you-go model.
I must note that this GKE on-prem/hybrid capabilities are probably the most critical part of Anthos stack.
- Anthos Service Mesh: This is Istio (the open source service mesh), set up and managed for you. It helps you (1) connect your services, even if they're across different datacenters or clouds (2) automatically export telemetry like traces, metrics and set up SLOs/alerts with them (3) set policies for your RPC traffic at high level e.g. tcp retry policies should be like X (4) enable mTLS automatically without changing your code across ALL your fleet and without having to run a PKI. (If you're still not convinced about service mesh, you probably don't have a RPC-heavy infra like some companies, maybe you don't need it.) You can technically install/manage Istio yourself across your clusters, but with Anthos, Google does that for you.
- Cloud Run for Anthos: This is actually hosted Knative (open source serverless stack on Kubernetes) for you. If you like rapid request-based autoscaling containers, you can have it anywhere (GCP, AWS, on-prem). You can install Knative anywhere, but Google does it for you and ensures it works properly with Istio and Kubernetes versions. I've written a blog post explaining what Knative does here: https://ahmet.im/blog/knative-better-kubernetes-networking/ Basically, this works like a CaaS (containers as a service) or build your own opinionated in-house opinionated PaaS with this.
We also offer Cloud Run as a fully-managed serverless product (not running on GKE, runs on Google’s infra directly) if you're into that: https://cloud.run/
- Anthos Config Management: This is available to GKE users as "GKE Config Sync". It's basically a GitOps tool (though not a complete suite like some stuff like Weaveworks Flux etc). You point it to a git repo/branch, and it will go create Kubernetes manifests in that directory across your clusters. It has a pretty neat model where leaf directories correspond to Kubernetes namespace and there's inheritance to fanout Kubernetes policy objects to multiple namespaces etc. AFAIK many companies using Kubernetes at scale build similar in-house solutions to this. So we worked with them to create a solution for everyone. This product also has some policy enforcement features and monitoring to see if your Kubernetes objects are propagating properly to your clusters.
- Other components: You'll find them on the page I linked. I think they are not as interesting to the day to day practitioners here.
Obviously I've oversimplified many of these, but the docs of each specific product do a decent job explaining, you should read if you're interested.
I'd say that if you are a practitioner, you don't need to worry much about details of Anthos. You can continue focus on the technologies you need to know (choose how low-level you wanna go: containers, Kubernetes, RPC or networking layer/service mesh, DevOps/GitOps) and continue to be successful.
However, if you work at a company which doesn't haven't gone through the cloud native application modernization (I'm hitting my head on the wall for you for saying this), Anthos can actually help your company use cloud-native stack WITHOUT having to go to cloud.
As you might imagine, a ton of companies out there are (perhaps happily) run on infrastructure stacks from another time. However, those looking to change it and try to use Kubernetes, they need to build a lot of in-house tooling, hire talent to manage Kubernetes. Google is decent (maybe more) at running Kubernetes clusters at scale for many customers, so Google can bring this service and all the listed above to you.
I'm not that familiar with pricing, though as you can see from https://cloud.google.com/anthos/pricing if you want to use these features on just on GCP, there's a $30 vCPU/month cost based on GKE nodes you're running. For other stuff, you need to contact support.
I should probably make this a blog post.
I submitted detailed feedback on two user surveys, praising the plugin and asking only for this a few months ago and it miraculously appeared.
Probably not related to my survey answers, but I like to think it was ;^)
(Also if you guys are sleeping on Cloud Run, it's basically the greatest thing since sliced bread. Knative and Serverless Containers are heaven)
Thank you for the praise, I'll pass it on to the team and I'm sure they'll appreciate it :)
I'm glad to hear you're enjoying the new Cloud Run support we've added to Cloud Code for IntelliJ. We'll get it added to Cloud Code for VS Code soon too for anyone else who is interested.
Also, thank you for submitting feedback via the user surveys - I read every single one and survey feedback, yours among others, absolutely was used to prioritize this work. If you have other features you'd like to see, please let us know.
Hi, Thanks! You can fill this form to be on the waitlist for Load Balancer to Cloud Run: https://docs.google.com/forms/d/e/1FAIpQLSd5vTSeU00oSDT5Bxye...
Is it binary-only software or is the source code available anywhere, and if so where?
Can you install it yourself or do you have to give Google systems or employees root access to your servers or your instances on non-Google cloud?
Does it need to communicate with Google servers to keep working or can it be configured to work without? If you go for a hybrid setup, does it always need your datacenter to be up, or can it be configured so that it doesn't?
In case it needs to always communicate with Google servers, what's the point of using a multi-cloud approach if it will go down whenever Google goes down?
Is there any plan to deliver it in a normal fashion (clear distributables and prices) instead of "contact sales"?
All the technical pages still seem to carry the message:
> This product or feature is in a pre-release state and might change or have limited support
Pricing directs you to "contact sales"
If you do want to try it out, it seems you need to fill out a google docs form.
I don't have a great idea about what the product does, but it seems to require use of an active google cloud account.
A multi-cloud platform product, having a hard dependency on a single vendor, seems like a non-starter. Doesn't that defeat the main purpose of going multi-cloud in the first place?
I'm also a bit curious what a multi-cloud platform product not built by a single vendor would look like. Are there any credible OSS offerings in this space?
I am willing to bet, although I have not tried Anthos, that the subset of stuff from your deployment process, identity handling, etc. that will "just work" if you try to move from Anthos to some random vanilla Kubernetes offering will be far too small and that you will practically speaking be locked in to the Anthos platform.
For example, OrbitDB was discussed here only a couple days ago: https://news.ycombinator.com/item?id=22918467
Edit: IPFS and this DB give a decentralized storage solution. There might be a decentralized processing solution as well, that I'm not aware of.
Disclaimer: I am actively contributing to Crossplane.
I’m a cloud true believer and admitted AWS fan (and probably would be an Azure fan too if I had any experience with it), but even Amazon admits that only 5% of Enterprise workloads are in the cloud. “The Enterprise” is leery of the cloud.
Even once you can convince a large company to move some of their workloads off prem, most CTO’s wouldn’t touch GCP with a 10 foot pole. No one ever got fired for choosing AWS and many large companies would trust Azure because they are heavily invested in Microsoft.
Google is not exactly known for its great enterprise support or customer support in general.
I think that's the point of this and similar offerings. You get to run on AWS but with even more handholding than AWS enterprise level support.
Besides, if you are at any type of scale - you’re de facto already “locked in”. Infrastructure has weight. Data migrations alone are a pain let alone migrating your network infrastructure, you’d permissions (IAM$), auditing for security compliance, your DNS entries, etc.
Why add Google to the mix at all if you’re all in on AWS and Azure? I bet few people have ever sat down and work with their project management team or IT people and estimated how long it would take or how much money it would cost to actually migrate to another vendor.
Can't confirm. My employer has been using GCP for quite some time and I've talked to lots of people at various conferences about their experience with GCP and they all praised the quality and responsiveness of the support.
Edit: They're talking about a different company
Odd categorization of Plaid, they (at least primarily) provide an API for linking (logging in to) users' bank accounts and scraping information from the bank websites.
https://cloud.google.com/kubernetes-engine/docs/how-to/ingre...
Who are they marketing to??? I feel like I stumbled onto a well written blog aimed at SEO.
Whoever it is sure ain't a good marketer. They really don't know who they're marketing to - some people respond really negatively to this type of copy.
"Whoever it is sure ain't a good marketer. They really don't know who they're marketing to"
Or maybe they are a good marketer, and you're the one who doesn't know who they're marketing to? I am not in marketing, but products with ad copy that doesn't entice me at all make billions of dollars.
That's the target customer for a lot of the stuff the major cloud providers do, actually. It's the whole focus of IBM's cloud hosting (Softlayer.) It's a majority of what Azure does. Even on AWS, look at any services that aren't pure automagic-IT-for-devs services, but are instead e.g. data-science services (Redshift), or pure ops services (AWS Storage Gateway), or even "IT for non-developers" services (AWS WorkSpaces!), and you'll see the same "CTO-ese."
Any IT service that isn't grown through groundswell support by devs, is grown by top-down partner-choice by a CTO.
"You didn't know you needed it, but this product is for you. It will make your work better. Your life better. You'll probably get a raise. And, while we're not allowed to say this ::whisper:: it will make you more attractive." <button>Buy Now Sight Unseen</button>
It's not that I don't understand they're important to a lot of developers and totally belong here. Time to search for a content filter though.
Edit: I mean look at it! It's basically an advertisement. It feels like Kardashians of coding.