Just about every presentation I went to this year at re:invent was not particularly good. They lacked depth and just seemed to be saying "just use this aws product" and you'll be all set.
You’ll have better luck if you limit yourself to 400-level presentations. Still not perfect, but you’re less likely to get 20 minutes of beginner-level introduction before you get to the good stuff.
But yeah, it’s frustrating. If you’re only going to breakout sessions at re:Invent you’re not getting your money’s worth. You need to have lots of discussions with AWS engineers in the AWS portion of the expo hall, and if your spend is high enough, get private meetings with the service teams you’re using or are interested in.
This is one of my main problems with the AWS everything advocates. I'm ok with using AWS mostly for a few key things like infrastructure, but I want my tooling to be independent and able to adapt to other situations. For example,why should I ever use CloudFormation over Ansible or Terraform when it lacks many features and can't be applied to other providers?
Well, then you get the worse of both worlds. Why use AWS at all if you’re not using their managed services? You’re spending more on raw resources and just as much on maintenance.
As far as Terraform, there is nothing portable about it. Each of its provisioners are cloud specific.
> Why use AWS at all if you’re not using their managed services?
Is this a serious question? I can value the infrastructure the AWS provides, such as regions and availability zones, while not wanting to get tied into their other services including the managed services for reasons such as business requirements for cloud agnosticism, which is for example one reason I tend to use HAproxy instead of ELB even on stuff on AWS. With CI/CD pipelines applied to infrastructure and a move towards no-ops, the maintenance costs of self-managed infra is reducing more and more.
I understand for some shops with a lack of personnel or knowledge to accomplish that self management may want to buy into a holisitic solution, but I think it's awfully intellectually disingenuous to not recognize the potential value of not buying into a holistic solution in certain situations.
Also, common, 'cloud specific'? So if I want to spin up my stack on Google or some other platform, I just have to rewrite a few lines of some .tf files. If I want to do that and I'm tied to CloudFormation.... I'm sol...
Really? You would go as far as managing your own load balancers instead of using a managed, highly available ELB from AWS?
Have you tried using Terraform across providers? It’s far more than just “a few lines of codes”. If you just want cross region availability and still manage things yourself, there are much cheaper ways than AWS.
Yes I would, there are some cases where the featureset is better. Also, it really is just a few lines of code. If you separate out IAC-code and configuration code even if you are running through say a jenkins pipeline it's not much trouble to change a few .tf files around to get the same result on a different provider... You act like writing terraform is hard or convoluted and I don't understand where you are coming from on that.
Do you get the cross AZ redundancy? and you’re managing more servers and probably spending more than just using an ELB. What about the HA from the target servers? If you were using a load balancer + autoscaling group with a min/max of 2. If your server went down or your entire AZ went down, autoscaling would bring another one up. If the server or AZ went down running your load balancer, it would also be redundant?
One of our web servers crashed in the middle of the night once. I got an alert saying it crashed and that another one was brought up. I turned over, went back to sleep and investigated the next morning. If the “server” running an ELB crashed, I wouldn’t have to care.
Do you really think that you’re going to be changing providers Willy nilly once you reach any type of scale?
It’s just like all of the bushy tailed “architects” who use the repository pattern just in case one day their CTO on a whim decides to move their six figure a year Oracle installation to Postgres.
What business value is being added by going through all of the trouble of staying “cloud agnostic” compared to the lack of HA and maintenance.
Btw, we also don’t have to manage Jenkins servers. We have a cloud formation template with parameters and quick create link that a developer can click on to create a CodeBuild project anytime they want to create a new build project. They specify the Github link that triggers the job. You then specify your build steps as a list of bash (Linux) or Powershell (Windows) commands in a yaml file.
We could also do the same with CodePipeline - create a CF template that creates Code Pipeline jobs on the fly, but we do use OctopusDeploy for deployments.
CodeBuild runs Docker containers to run builds. Again no servers to manage but I believe by default it can run 50 builds simultaneously. If a build job needs custom tooling or a custom version of a runtime, the developer can build a custom Docker container to run the build.
You’re doing more work just in the eventuality that you may want to rip out your entire infrastructure and you’re paying more for the privilege.
You've obviously fully bought into AWS and don't want to hear anything to the contrary. That's fine if it works for you, but you aren't interested in honest conversation, so I'm done here.
It’s not about “fully buying into AWS”. I would say the same about Azure or GCP - well not GCP I wouldn’t trust Google for any infrastructure. If you have a static workload and the staff to handle infrastructure, AWS may not be worth it. Use a colo or a cheaper service like Linode. Everything I use managed services for I’ve done on prem.
I’m saying that it is just as much a waste of money and time to use any cloud provider just to duplicate what you could do at a colo.
I would say the same thing about spending extra money for Oracle or SQL Server. Why spend extra on a commercial solution and not take advantage of it. You wouldn’t use Excel just to create CSV files.
I come from places where that have self-managed on-prem datacenters, with self-managed CPU/GPU clusters, and using CloudFormation for internal infrastructure simple is a non-starter, but with some stuff in AWS, why would we do half CF and half terraform? Do it in tf et al for both. Yes I have occasionally used ELB/EFS in front of HAproxy for some of the reasons mentioned, because I recognize some situations where the value-add is worth the tradeoff.
The entire point of my post is that lock-in to AWS can be bad in certain situations, which I have tried to clarify multiple times now and you have ignored every time. I'm not saying it's bad in all situations, just that it can be bad in some. Your extreme defensiveness seems strange to me for that reason.
I look at it like datacenter design. You don't have just one isp. You have multiple ISP lines incoming, often in different forms (aerial/underground/wisp) so that you are never too dependent on one thing. Your metaphorical datacenter may have a OC-3840, and that may be a big feature for you, but you should at least admit to yourself and customers not having other ISP lines might be a weakness, despite the awesomeness of the 3840..
Writing Terraform is easy - having it work after six month of features being delivered, the cloud provider evolving, and working with all your other microservices, isn't particularly. Assuming you can do that and just change a few lines to redeploy in a new cloud environment (imagine that including GKE to EKS) seems naive.
On another note, in the list of top ten things that will add business value and either make the company money or save the company money where does “being cloud agnostic rate”?
What competent CTO would go through a migration from one provider to another on a whim? The chance of regression and business interruption is too high and the benefits too low to make it worth it if you’re at any type of scale.
At a certain scale, being cloud agnostic is actually very financially beneficial. The negotiating leverage leads to substantially better private pricing deals with your cloud provider.
At a certain scale are you really going to move your entire infrastructure over to another provider? The time, migration cost, and business risks are usually not worth it.
The time you’re spending migrating, testing, validating for correctness and often compliance is time you’re not spending on creating business value. On top of that, you’ve migrated all of the practices from being at a colo without any of the time to market , lower headcount, and flexibility or availability that you get from using native cloud services.
You’re investing time and not taking into account any of the opportunity cost. Businesses are so dependent on vendors these days - from payroll, to Office Suites, to database vendors, to even their physical facilities it’s silly to optimize for one part of your infrastructure at the expense of other opportunities in the rare case that you might change vendors.
Do you also have a migration plan for all of your other dependencies? Are all of your documents in an open source format? What about your project management system? Your AD? Your device management system? Are all your desktops running Linux? Are you using Microsoft Exchange? What are you using for time tracking? Expense reimbursement? SSO? Alerting and monitoring?
When your cloud costs reach the high hundreds of millions to billions of dollars per year, it is worth it. At that point you probably want to run your infrastructure across cloud providers anyway.
At that point, you’re going to need the in house expertise anyway and you would probably be better off managing your own data center.
Even if you are “cloud agnostic”, at that size just your networking infrastructure, compliance, auditing, and data migration is going to cost a boatload of money and it would take years to save enough to make the migration worth it. Not to mention the retraining and risks that come from the migration.
You probably also have a few direct connections to the cloud provider.
Cloud agnostic in terms of using the best products from each service: sure. It makes sense to leverage the strengths of each cloud provider.
Having all your infra written with shim layers to move it freely between providers sounds like the scale you need to start thinking about hosting your own "cloud provider" or stop using public cloud.
I’ve been slowly coming around to this. Also there is no reason why something like salt cannot be adopted into the k8s world, it would bring better base image management capabilities by forcing larger organizations into a common image. A common base image would be easier to patch, upgrade, and maintain. Larger k8s deployments tend to not be so great at image management, which is a cost paid in operations dollars that were supposed to be saved by k8s.
What it comes down to is tooling which separates the application enough from the platform where I can move between metal, virtual, and containers without much effort. If this goal is achieved, moving between providers is achieved in tandem.
I enjoyed the talk about scaling Pokemon GO + migrating from DBshards to Aurora PostgreSQL which went into what they tried, what failed and what succeeded rather than just saying "We used X service and it was great"
Almost all presentations at big tech events are like that. I usually watch the technical backgrounds and deep dives to get information that isn't already on the marketing website.
Can confirm, it looked like this was the whole point of a SageMaker presentation, for instance. (On the other hand, no one has been fired for buying from IBM... err, Amazon!)
Just about every presentation I went to this year at re:invent was not particularly good. They lacked depth and just seemed to be saying "just use this aws product" and you'll be all set.