Dear friend, you have built a Kubernetes (2024)
macchaffee.com
macchaffee.com
The response I always got when suggesting Kubernetes is "you can do all those things without Kubernetes"
Sure, of course. There are a million different ways to do everything Kubernetes does, and some of them might be simpler or fit your use case more perfectly. You can make different decisions for each choice Kubernetes makes, and maybe your decisions are more perfect for your workload.
However, the big win with Kubernetes is that all of those choices have been made and agreed upon, and now you have an entire ecosystem of tools, expertise, blog posts, AI knowledge, etc, that knows the choices Kubernetes made and can interface with that. This is VERY powerful.
I feel the same way about commercial APM tools. Obviously in a perfect world, you would have software so simple and fast that they’re unnecessary. Maybe every month or two someone has to grep some logs that are already in place. Once you’ve gotten yourself in situation where this is obviously not true, having Datadog, New Relic or similar set up (or using k8s instead of 100 unversioned shell scripts by someone who doesn’t work there anymore) will make your inevitable distributed microservice snafu get resolved in hours rather than a longer business-risking period.
The only problem I see in this case is that complexity doesn't come all at once. By the time you reach a problem that k8s is good at solving, you've probably already accidentally made a k8s alongside your piece of software.
In my(quite short) SWE career, I've seen software evolve, even ones with a proper design stage. Maybe I just don't have enough experience to have seen a properly designed project, but I don't know what I don't know after all.
Have they really? I have a few apps deployed on k8s and I feel like every time I need something, it turns out it doesn't do that and I'm into some exotic extension or plugin type ecosystem.
Something as simple as service autoscaling (this was a few years ago) was an adventure into DIY. Moving from google cloud to AWS was a complete writeoff almost - just build it again.
I'm sure it captures some layer of abstraction that's useful but my personal experience is it seems very thin and elusive.
Concretely: Take your app. With one button click, or apt-get install ??? on all your machines, configure k8s. Now, run your app.
The idea that this could work has been laughable for any k8s production environment I've seen, which means you can't do things like write automated tests that inject failures into the etcd control plane, etc.
(Yes, I know there are chaos-monkey things, but they can't simulate realistic failures like kernel panics or machine reboots, because that'd impact other tenants of the Kubernetes cluster, which, realistically, is probably single tenant, but I digress..)
If your configuration is megabytes of impossible to understand YAML, and is also not portable to other environments, then what's the point?
(I understand the point for vendors in the ecosystem: People pay them for things like CNI and CSI, which replace Linux's network + storage primitives with slower, more complicated stuff that has worse fault tolerance semantics. Again, I digress...)
If almost all your configuration is about getting Kubernetes set up, and not about your application setup inside Kubernetes, there probably isn't a point. But being able to use roughly the same config inside different Kubernetes is quite good.
If you just tell kubectl to dump your pod configs, then load them on some other cluster, that definitely won't work.
If you use the management software that generated the pod setup somewhere else, that probably won't work either because the somewhere else is going to be missing the CSI and CNI you targeted. Even if those match, it'll be missing the CRDs. God help you if you want to run two programs on one Kubernetes, and there's a CRD versioning conflict in their two dependency sets.
Yep. Kubernetes is not just kubernetes when moving between clouds, it becomes a very opinionated product (for better or worse) with lots of vendor addons. Could someone that is familiar with one pick up on the other? Sure! But there are gotchas. And then kubernetes on prem adds the hardware lifecycle piece, and potential data locality issues, etc.
We run a bunch of services in two very different cloud vendors (one of which used to be DIYed with kubeadm), and also on dev machines with k3s. Takes a while to figure this out and to draw the kustomize boundaries in the right place, but once you do, it’s actually really nice.
Two things work in our favor:
- we’ve been at this for around 8 years, so we didn’t have to deal with all the gotchas at once
- we aggressively avoid tech that isn’t universal (so S3 is OK, but SQS or DynamoDB is not; use haproxy instead of ingress controllers; etc)
I think this is gradually getting better. Networking with Gateways is better than with Ingress in this sense. Things like autoscaling groups need to get better, as they are (or were a couple of years ago) very bespoke.
Yep! I am now using k8s even for small / 'single purpose' clusters just so I can keep renovate/argo/flux in the loop. Yes, I _could_ wire renovate up to some variables in a salt state or chef cookbook and merge that to `main` and then have the chef agent / salt minion pick up the new version(s) and roll them out gradually... but I don't need to, now!
I've used it every day since then so I have the luxury of knowing it well. So the frustrations that the new or casual user may have are not the same for me.
It really feels like we are drowning in self-imposed tech debt and keep adding layers to try and hold it for just a while longer. Now that being said, there is no reason not to add Kubernetes once a sufficient overlap is achieved.
You have to pick and then configure those components, just like you would have had to pick and configure apps doing those things if you were not using k8s, so the only thing k8s actually brings to the table is a common configuration format (yaml).
And there's zero setup. Just a deployment yaml that specifies exactly what you want deployed, which has the benefit of easy version control.
I don't get why people are so bent on hating Kubernetes. The mental cost to deploy a 6-line deployment yaml is less than futzing around with FTP and nginx.
Kube is the new LAMP stack. It's easier too. And portable.
If you're talking managed kube vs one you're taking the responsibility of self-managing, sure. But that's no different than self-managing your stack in the old world. Suddenly you have to become Sysadmin/SRE.
This made me audibly guffaw. Kubernetes is a lot of things, but "portable" is not one of them. GKE, EKS, AKS, OCP, etc., portability between them is nowhere near guaranteed.
"If you stick to standard C..."
No one does, that's the issue. Helm charts that only support certain cloud providers, operators and annotations that end up being platform specific, etc.
>now give you default storage classes and ingresses
Ingress is being deprecated, it's Gateway now! Welcome to hell, er, Kubernetes.
Do you have any links about Ingress being deprecated?
Official docs here: https://kubernetes.io/docs/reference/kubernetes-api/service-...
There are no mentions about this API being deprecated.
Anyway, Ingress resource been in "Migrate to Gateway" state for awhile.
To clarify, it’s already retired and the repo has been archived since 24th of March.
NGINX Ingress is deprecated, not the Ingress resource itself.
If you're using it after it's dead, you're at risk of further problems of this nature that aren't in the underly nginx reverse proxy but in the code wrapping it.
On EKS I'm using whatever AWS has brewed up to integrate ELB/ALB, but I'll tend to trust it ... though maybe I shouldn't, given all the troubles I have with other integrations like secrets management.
- New cluster setup, time to use gateway! Yay!
- Oh crap, like 80% of the helm chart and other existing configurations I need for the softwares I'm trying to deploy STILL doesn't use gateway, this new API that's been out for... like half a decade at least.
- Even core networking things like Istio/Envoy only have limited gateway support compared to ingress
- Sigh. Ingress again.
It's been like this since gateway's inception and every time I check the needle has moved like 2% towards gateway. So I'm looking forward to year 2050 when I can use gateway!
The problem, as CNCF knows, if they pushed Gateway and deprecated ingress the world would revolt due to the amount of work involved to migrate stuff. Therefore, they leave it up to "the people" to do the extra work themselves, who have no incentive to do so since for many usecases it's not materially better.
Obviously you can lock yourself in if you choose, but I have yet to see third party tools that assume a specific provider (unless you are using tools created BY that provider).
At my previous spot, we were running dozens of clusters, with some on prem and some in the cloud. It was easy to move workloads between the two, the only issue was ACLs, but that was our own choice.
I know they are pushing the new gateway api, but ingresses still work just fine.
Helm charts may not support a cloud platform like Rancher, Azure, etc. or may have platform specific issues. First one I checked: https://docs.jfrog.com/installation/docs/helm-chart-requirem...
"When deploying a JFrog application on an AWS EKS cluster, the AWS EBS CSI Driver is required for dynamic volume provisioning. However, this driver is not included in the JFrog Helm Charts."
"JFrog validates compatibility with core Kubernetes distributions. Some Kubernetes vendors apply additional logic or hardening (for example, Rancher), so JFrog Platform deployment on those vendor-specific distributions might not be fully supported."
I don't you made that argument but could a valid conclusion of your comment be that, because Kubernetes is so ubiquitous, using it frees you from being a Sysadmin/SRE?
Writing this yaml is hours and hours of setup if you can't ctrl+c/v from your last project
It's not obligated to be, but it's also obvious why people would want it to be.
I think I disagree with this, or at least the implication. I think it is true you can solve EACH OF THISE PROBLEMS INDIVIDUALLY in a simpler way than Kubernetes, the fact that you are going to have to solve at least 5-10 of those problems individually makes the sum total more complicated than Kubernetes, not to mention bespoke. The Kubernetes solutions are all designed to work together, and when they fail to work together, you are more likely to find answers when you search for it because everyone is using the same thing.
I think it is fair to say k8s is not a zero cost abstraction, but nothing you use instead is going to be, either, and when you do run into a situation where that abstraction breaks, it will be easier to find a solution for kubernetes than it will for the random 5 solutions you pieced together yourself.
Docker and k8s are just wrappers around namespaces, cgroups, file system ACLs, some essential cli commands, which can also be configured per user.
We may be headed back there. Have seen some experiments leveraging Linux kernels BPF and sched_ext to fire off just the right sized compute schedule in response to sequences of specific BPF events.
Future "containers" may just be kernel processes and threads... again. Especially if enough human agency looks away from software as AI makes employment for enough people untenable. Why would those who remain want to manage kernels and k8s complexity?
Imo its less we agreed on k8s specifically and more we agreed to let people use all the free money to develop whatever was believed to make the job easier; but if the jobs go away then it's just more work for the few left
Docker, yes, but kubernetes is way more than that the instant you have more than one physical machine node. (If you only have one node in any deploy, sure, it's likely overkill, but that seems like a weird enough case to not be worth too much ink.)
If you silently replaced all my container images with VM images and nodes running containers with nodes running VMs, I think the vast majority of all my Kubernetes setup would be essentially unchanged. Heck, replace it all with people with hands on keyboard in a datacenter running around frantically bringing up new physical servers, slapping hard drives in them, and re-configuring the network, and I don't think the user POV of how to describe it would change that much.
huh, but how would bursting work then? Do VMs support it nowadays?
But, honestly, more generally in my head I wasn't thinking much about it since I consider that as a "cost optimization" thing than a "core kubernetes function." E.g. the addition (or not) of limits is just a couple lines, compared to all the rest of the stuff that I'd be managing specification of (replicas, environment, resource baseline, scheduling constraints, deployment mode...) that would translate seamlessly.
(And there are a lot of parts of kubernetes that annoy me, especially around the hoops it puts up to customize certain things if you reaalllly actually need to, but it would never cross my mind in a hundred years to characterize it as just a wrapper around cgroups etc like the OP.)
Inevitably it will be a human replacing it with what they define is the best method
"Kubernetes is beautiful. Every Concept Has a Story, you just don't know it yet... So you use a Deployment... So you use a Service... So you use Ingress... So..."
(Full text at the link)
I do run a k3s cluster for home stuff... But I really wish I could get what it provides in a much simpler solution.
My dream solution would effectively do the same as k3s + storage, but with a much simpler config, zero yaml, zero choices for components, very limited configuration options, it should just do the right ting by default. Storage (both volume and s3), networking, scale to zero, functions, jobs, ingress, etc... should all just be built in.
The main reason I went to k8s, is for the not having to think about what machine will have enough resources to run an app, just throw it at the cluster and it figures out where there's capacity. And, I want hardware failing/getting replaced to be a non issue.
edit: I wanted to add that my hobby is not systems admin, I want it to be as hands off as possible. Self-hosting is a means to an end. I have so far saved over $200/month in subscriptions by replacing subscriptions I was using with self-hosted alternatives. I can now use that money on my actual hobbies.
This comes down to that fact that software architecture is not great in general and very few people care about simplicity.
Once you get used to it, it just makes managing things simple if you always use it for everything. I have a personal harbor service that I run on my local cluster that has all my helm charts and images, and i can run a single script that sets up my one node cluster, then run a helm install that installs cert-manager and my external-dns, and now I can deploy my app with whatever subdomain I want and I immediately get DNS set up and certs automatically provisioned and rotated. It will just work.
2. It allows me to configure everything using standard manifests. I need to provision the cluster itself initially, then everything could be done with gitops of various automation levels. I don't need to upload my pages via FTP. My CI will build OCI image, publish it to some registry, then I'll change image tag of my deployment and it'll be updated.
3. It allows to start simple, and extend seamlessly in the future. I can add new services. I can add new servers. I can add new replicas of existing services. I can add centralized logging, metrics, alerts. It'll get more complicated but I can manage the complexity and stop where I feel comfotable.
4. One big thing that's solved even with the simplest Kubernetes deployment is new version deployment with zero downtime. When I'll update image tag of my deployment, by default kubernetes will start new pod, will wait for it to answer to liveness checks, then redirect traffic to new pod, let old pod to gracefully stop and then remove it. With every alternative technology, configuring the same requires quite a bit of friction. Which naturally restricts you to deploy new versions only at blessed times. With Kubernetes, I started to trust it enough, I don't care about deployment time, I can deploy new version of heavily loaded service in the middle of the day and nobody notices.
5. There are various "add-ons" to Kubernetes which solve typical issues. For example Ingress Controller allows the developer to describe Ingress of the application. It's a set of declarative HTTP routes which will be visible outside and which will be reverse-proxied to the service inside. Simplest route is https://www.example.com/ -> http://exampleservice:8080, but there's a lot more to it, basically you can think about it as nginx config done differently. Another example is certificate manager, you install it, you configure it once to work with letsencrypt and you forget about TLS, it just works. Another example is various database controllers, for example cloudnativepg allows you to declaratively describe postgres. Controller will create pod for database, will initialize it, will create second pod, will configure it as replica, will perform continuous backup to S3, will monitor its availability and switch master to replica if necessary, will handle database upgrades. A lot of moving parts (which might be scary, tbh), all driven by a simple declarative configuration. Another example is monitoring solutions, which allow to install prometheus instance and configure it to capture all metrics from everything in cluster along with some useful charts in grafana, all with very little configuration.
6. There are various "packages" for Kubernetes which essentially package some useful software, usually in a helm charts. You can think about `apt-get` but for a more complicated set of services, mostly pre-configured and typically useful for web applications. The examples above are all installable with helm, but they add new kubernetes manifest types, which is why I called them "add-ons", but there are also simpler applications.
Just for the record, I don't suggest that to everyone. I spent quite a bit of time tinkering with Kubernetes. It definitely brings a lot of gotchas for a new user and it also requires quite a bit of self-restrictions for experienced users to not implement every devops good practice in the world. Sometimes maybe you don't even want to start with ingress, I saw cluster which used manually configured nginx reverse proxy instead and it worked for them. You can be very simple with Kubernetes.
And then they hit all the things that make sense in big company with like 40 services but very little in their context and complain that complex thing designed for complex interactions isn't simple
The database is more complex since there is storage affinity (I use cockroachDB with local persistent volumes for it) - but stateful is always complicated.
Personal projects are one thing, but even the smallest startup wants to be able to avoid data loss and downtime. If you are running everything on one server, how do you do kernel patches? You need to be able to move your workload to another server to reboot for that, even if you don’t want redundancy. Kubernetes does this for you. Bring in another node, drain one (which will start up new instances on the new node and shift traffic before bringing down the other instance, all automatically for you out of the box), and then reboot the old one.
Again, you could do all of this with other tech, but it is just standard with Kubernetes.
Seems true at a glance!
> and downtime.
Maybe less so - I think there’s plenty out there, where they’re not chasing nines and care more about building software instead of some HA setup. Probably solve that issue when you have enough customers to actually justify the engineering time. A few minutes of downtime every now and then isn’t the end of the world if it buys you operational simplicity.
Then I realized giving it root on a $3 VPS is functionally equivalent. If it blows it up, you just reset the VM.
It sounds bad but I can't see an actual difference.
No argument there. The Toyota 5S-FE non-interference engine is a near indestructible 4 cylinder engine that's well documented, popular and you can purchase parts for pennies. It has powered 10 models of Camrys and Lexus and battle proven. You can expect any mechanic who has been a professional mechanic for the last 3 years know exactly what to do when it starts acting up. 1 out of 4 cars on the road have this engine or a close clone of it.
It's not what any reasonable person would use for a weedwhacker, lawnmower, pool pump or an air compressor.
K8s is well suited to dynamically scaling a SaaS product delivered over the web. When you get outside this scenario - for example, on-prem or single node "clusters" that are running K8s just for API compatibility, it seems like either overkill or a bad choice. Even when cloud deployed, K8s mostly functions as a batteries-not-included wrapper around the underlying cloud provider services and APIs.
There are also folks who understand the innards of K8s very well that have legitimate criticisms of it - for example, this one from the MetalLB developer: https://blog.dave.tf/post/new-kubernetes/
Before you deploy something, actually understand what the pros/cons are, and what problem it was made to solve, and if your problem isn't at least mostly a match, keep looking.
It’s well suited to other things as well, people are just in denial about some of them.
“I need to run more than two containers and have a googleable way to manage their behavior” is a very common need.
What's the problem with a single-node cluster? We use that for e.g. dev environments, as well as some small onprem deployments.
> Even when cloud deployed, K8s mostly functions as a batteries-not-included wrapper around the underlying cloud provider services and APIs.
Which batteries are not included? The "wrapper around the underlying cloud provider services and APIs" is enormously important. Why would you prefer to use a less well-designed, more vendor-specific set of APIs?
I seriously don't get these criticisms of k8s. K8s abstracts away, and standardizes, an enormous amount of system complexity. The people who object to it just don't have the requirements where it starts making sense, that's all.
What surprises and gotchas did you have to deal with using k3s as a Kubernetes implementation?
Did you use an LB? Which one? I'm assuming all your onprem nodes were just linux servers with very basic equipment (the fanciest networking equipment you used were 10GbE PCIe cards, nothing more special than that?)
There really weren't any particular surprises or gotchas at that level.
In this context, I've never had to deal with anything at the level of the type of Ethernet card. That's kind of the point: platforms like k8s abstract away from that.
Now the actual CI/CD/thing-doers tools that all suck... I'm still stuck with those.
Same with build.sh and doing it in such a way that I can use all the build.sh in my ci.yml for Github Actions.
Some Kustomize, a little bit of envsubst and we're good to go thank you very much.
I manage 100+ variations on a single helm chart and 50+ such helm charts at work daily for 7 years across 11 datacenters/kubernetes clusters. And I have team members who swear by kustomize. The number of kustomize typo errors and issues that I have deal with is unimaginable. Whereas if I test and deploy a helm chart, I know it will work everywhere in every variation.
Kustomize is just plain terrible and backwards as a solution. It doesn't scale, it is half assed. It tries to basically require you to build your own compiler and parser and transform. With kustomize + envsubst: dear friend, you have built helm.
For the simple use case you're describing, Helm is not required. Plenty of other solutions around.
For use cases where it starts getting useful, we both agree that something has gone terribly wrong.
I still don't know why Helm exists. It's a solution that created lots of problems that didn't exist.
Using one of the most horrible templating languages since ASP. Helm is what happens when a devops team decides to yolo into software development.
What's the issue with kustomize? It works well for us.
cat manifests.yaml | kubectl apply -f - --server-side --field-manager "$FIELD_MANAGER" --prune --applyset "$APPLYSET" --namespace "$NAMESPACE"
Hooks: I want to apply my database migrations and populate the database with static datasets before I deploy my application, without having my CI connect to the database cluster (at places I've worked, the CI cluster and K8s cluster were completely separate).
Regarding hooks: I don't know. All applications that I've used, implemented migrations internally (it's usually Java with Flyway), so I don't need to think about it. One possible approach could be to use flux CD with Job definition. I think that Flux will re-create Job when it changes. So if you change image tag, it'll re-create Job and it'll trigger Pod execution. But I didn't try this approach, so not sure if that would work for you.
kubectl delete -f <manifests.yaml>
kubectl delete -k <kustomization_directory>
A Job feels like a good fit for this. CI deployes the Job without connecting to DB, Job runs migrations using the same connectivity as the application.
You could a) have the app acquire a lock in the db and do its own migrations, or b) create a k8s job that runs the migration tool, but make sure the app waits for the schema to be updated or at least won't do anything bad.
Fun times.
For very simple deployments, you don't need anything at all. Just write manifests and use `kubectl apply`. You can write `deploy.sh` but it'll be trivial.
If you want templating, there are many options. You can use `sed` for the most simple templating needs. You can use `cpp`, `m4`, `helm` or `kustomize`. I, personally, like `kustomize`, but `helm` probably not the worst template engine out there.
Kustomize is even somewhat included into basic kubernetes tooling, so if you want something "opinionated", it is there for you. It works.
Git and Kubetnetes configuration cannot go hand in hand. You cannot go back in past indefinitely because cluster state might not be that reversible. If so, git is useless.
And no, doesn't apply for database migrations. You can mostly run migrations backwards if each migration was written carefully.
I have a "simple" representation of services using CUE, that generates the yaml manifests and flux deploys them.
I hesitated a while before going the k8s route but before that I had a overly error-prone Ansible configuration and I got sick of manual templating (hence the move to CUE for type safety).
There's also the fact that I wanted my services to be as plug and play as possible, so for example automatically generated openid credentials and very easily configurable central SSO, along with the easily configurable reverse-proxy.
If anyone thinks that k8s is not the best tool for this, I'm always interested in advice.
(Also a lot of complexity in my setup is due to self hosting, I have Istio, MetalLB, proxmox CSI, and all other kind of stuff that your cloud provider would already have, and these are the things that take most of the configuration files in my repo)
(I tend to think this one is acceptable in the beginning, but certainly doesn’t scale.)
And that's a very important distinction when it comes to maintaining complex systems. This could've changed with LLMs (I'm still adjusting to what new capabilities mean for various decision-making logic), but before machine intelligence debugging an issue with Kubernetes could've been a whole world of pain.
I'm not doing overlay networks, I'm using a single bare-metal host, and I value the hands-on Linux administration experience versus the K8s cluster admin experience. All of these are reasons I specifically chose not to use Kubernetes.
The second I want HA, or want to shift from local VLANs to multi-cloud overlays, or I don't need the local Linux sysadmin experience anymore? Yeah, it's K8s at the top of the list. Until then, my solution works for exactly what I need.
I run K8s at home. I used to do docker-compose - and I'd still recommend that to most people - but even for my 1 little NUC with 4vcpu / 16Gi Homelab, I still love deploying with K8s. It's genuinely simpler for me.
If anyone's looking for inspiration, my setup:
* ArgoCD pointed to my GitLab repos
* GitLab repos contain Helm charts
* Most of the Helm charts contain open-source charts as subcharts, with versions set like (e.g.) `version: ~0` - meaning I automatically receive updates for all major version until `1`
* Updating my apps usually consists of logging into the UI, reviewing the infrastructure and image tag updates, and manually clicking sync. I do this once every few months
My next little side project: Autoscaling into the cloud (via a secure WireGuard tunnel) when I want to expand past my current hardware limitations
The needed tweaks, the ability to customize things, basically goes to zero because the support staff is technical about the software, but NOT about Kubernetes.
I am not joking: a recent deployment required 3x VMs for Kubernetes, each VM having 256 gigabytes of RAM; then a separate 3x VMs for a different piece. 1.5TB of RAM to manage less than 1200 network devices (routers etc. that run BGP).
No one knew, for instance, how to lower the MongoDB (because of course you need it!) resource usage, despite the fact that the clustered VMware install is using a very fast SSD storage solution and thus MongoDB is unlikely accelerate anything; so over 128GB RAM is being burned on caching the results coming back from SSDs that are running at many-GB/s throughput.
So unless I want to dig into their YAML files that are not documented…
But really this applies to any powerful tool. If you need to measure a voltage, an 4 channel oscilloscope also probably seems too complicated.
Dear friend, you have built a Kubernetes - https://news.ycombinator.com/item?id=42226005 - Nov 2024 (277 comments)
Kubernetes was overkill (I do that all day, 5 days a week); Kamal was too restrictive, so I found myself rolling out Yoink. Just what I need from k8s, but simple enough I can point it to a baremetal machine on Hertzner that can easily run all my workloads.
- using Tailscale SSH is brilliant
- using caddy-docker-proxy for ingress is brilliant
What do you use for:
- service discovery
- secret store (EDIT: Crap you use Infisical. No shade, I just have this horrible foreboding it will end up like Hashicorp. I use Conjur Secretless Broker but am tracking: https://news.ycombinator.com/item?id=47903690)
- backing up and restoring state like in a DB
PS: Have you been having issues with Hetzner the last few weeks?
For secrets, I self-host Infisical on the box -- easy to plug in whatever secret manager, should make it pair nicely with https://github.com/tellerops/teller or something similar
Had no problems with Hertzner so far, just enjoying the raw CPU power of bare metal. The plan is to roll out more boxes across different providers, using Tailscale for the backplane network and Cloudflare to load-balance between them. All in due time What issues have you been having ?
Personally commiting to using Tailscale as a core foundation of my infrastructure and Ionscale is my hedge against getting Hashicorped.
> Service discovery is basically just Docker's internal DNS. Caddy-docker-proxy can use it to find healthy upstreams
Do you have a writeup of this somewhere? I'm unaware of being able to manage Docker's internal DNS over some kind of an API (would appreciate if you know a way to). The only way I know is to manipulate network aliases via Docker Engine API. As a result I use Hickory DNS with RFC 2136. That coupled with Caddy-docker-proxy gets me extremely close.
- People who would prefer their way of doing this, whether that's deployments on VMs, or use some sort of simpler cloud provider.
I had the same opinion a few years ago, but have kind of come to like it, because I can cleanly deploy multiple applications on a cluster in a declarative fashion. I still don't buy the "everything on K8s", and my personal setup is to have a set of VMs bought from a infrastructure provider, setup a primary/replica database on two of them, and use the rest as Kubernetes nodes.
- People who run Kubernetes at larger scales and have had issues with them.
This usually needs some custom scaling work; the best way to work around this if you're managing your own infra[1] is to split the cluster into many small independent clusters, akin to "cellular deployments"[2]/"bulkhead pattern"[3]. Alternatively, if you are at the point where you have a 500+ node cluster, it may not be a bad idea to start using a hyperscaler's service as they have typically done some of the scaling work for you, typically in form of replacing etcd and the RPC layer through something more stable.
- People who need a deep level of orchestration
Examples of such use cases may be to run a CI system or a container service like fly.io; for such use cases, I agree that K8s is often overkill, as you need to keep the two datastores in sync and generate huge loads on the kube-apiserver and the cluster datastore in the process, and it might be often better to just bring up Firecracker MicroVMs or similar yourself.
Although, I should say that teams writing their first orchestration process almost always run to Kubernetes without realizing this pitfall, though I have learned to keep my mouth shut as I started a small religious war recently at my current workplace by raising this exact point.
[1] Notice how I don't say "on-prem", because the hyperscaler marketing teams would rather have you believe in two extremes of either using their service or running around in a datacenter with racks, whereas you can often get bog-standard VMs from Hetzner or Vultr or DigitalOcean and build around that.
[2] https://docs.aws.amazon.com/wellarchitected/latest/reducing-...
[3] https://learn.microsoft.com/en-us/azure/architecture/pattern...
For example:
* For some cursed reasons you want to make sure every single one instance of a large batch job see just one NIC in its container and they are all the same IP and you NAT to the outside world. Ingress? What ingress? This is a batch job!
* Like the previous point, except that your "batch job" somehow has multiple containers in one instance now, and they should be able to reach each other by domain.
Scaling is a sidenote, that it becomes easy is a result of hoisting everything else onto one control plane and a set of coherent APIs.
One example was an engineer wanted to build an API that accepts large CSV (GBs of credit reports) to extract some data and perform some aggregations. He was in the process of discussing with SREs on the best way to process the huge CSV file without using k8s stateful set, and the solution he was about to build was basically writing to S3 and having a worker asynchronously load and process the CSV in chunks, then finally writing the aggregation to db.
I stepped in and told him he was about to build a data warehouse. :P
Basecamp is running successfully with Kamal and surely making more money than all the Kubernetes experts combined on this platform.
Small companies should not behave and pretend like big enterprises. Maybe Kubernetes is justified in those scenarios.
But for 95% of the population, it's an overkill and a constantly moving target.
I see docker as a way to avoid having a standard dev platform for everyone in the company so that the infra team don't have to worry about patch xyz for library abc, only run docker.
But, with all the effort put in place to coordinate docker, k8s and all the shebang, isn't it finally easier to force a platform and let it slowly evolve over time?
Is docker another technical tool that tries to solve a non-technical problem?
K8s is a way to take that and make it scale up for a large number of applications on a large number of hosts in a production business in a way that's automated and resilient to failure.
You can have one template docker-compose.yaml file and separate deployment files for different envs, like: docker-compose.dev.yaml, docker-compose.prod.yaml
I think swarm is really underrated
Yes (if I'm getting your question right). here is an example of nfs volume: https://docs.docker.com/reference/compose-file/volumes/#driv...
Ended up doing a mix. Built on compose for now but in a manner that’ll lift and shift to k8s easily enough. Its containers talking over network either way
If your work is easily googleable/parseable by an AI, why would anyone pay you?
As to the job security idea: the only people who do this are people who aren’t good at creating real value, so they have to try to create niches where they’re needed.
> Any sufficiently complicated C or Fortran program contains an ad-hoc, informally-specified, bug-ridden, slow implementation of half of CommonLisp.
>> The real problem is that programmers have spent far too much time worrying about efficiency in the wrong places and at the wrong times; premature optimization is the root of all evil (or at least most of it) in programming.
-- Donald Knuth, Computer Programming as an Art (1974)
EDIT:
> Except if you quit or go on vacation, who will maintain this custom pile of shell scripts?
Honestly? I don't care. There is a reason why I quit and 99% of time it's the pay. And if the company doesn't pay me enough to bother then why should I? Why should I bother about some company future in the first place?
The people advocating for boring tech generally aren't interested in containers.
You can just run programs.
If your app is just a blob that can be run it is fine, but many languages make it more complicate.
I wonder if just putting app into .appimage + using systemd for some of the separation would be a sweet spot ?
If you had said "unikernels" I would have had no arguments to make.
If you didn't imply that, I apologize.
If you did mean that, I disagree with you precisely because your point works if you only care about dependency management - it falls apart on system state. A static binary is a process on the host and shares the same process space, network stack, and filesystem.
OTOH, a container is a jail (the primary usecase): I can't cgroup a static binary's memory usage or give it a virtual network interface without reimplementing "container lite". Containers aren't just 'statically linked programs' - they allow me to use the kernel as a hypervisor for isolated environments.
What they are though, a messy but practical compromise to Unikernels - which was my last point in our GP.
I didn't mean "containers don't have any advantages compared to statically linked programs".
I have no interest in going back to the hell that is deploying a java app.
And somehow k8s tends to use one of the most valuable chips. The smartest guy who from then on only works with maintaining and updating k8s. Instead of building the best and smartest products.
Kafka also competes with this.