Ably blog claims company doesn't need Kubernetes to scale, surge in traffic takes down entire website

A post on why using Kubernetes to scale would mean "doing mostly the same things but in a more complicated way" was so popular that the site hosting the article went down due to the sheer volume of traffic. Maik Zumstrull is writer of code at Ably, a company which runs pub/sub (publish and subscribe) messaging services. He …

[FIA]

"Now live with your bad/good design patterns on your own for the life cycle of your product or service, you get none of the shared knowledge and advancement through the k8s community or platform,"

This is an interesting argument, but it does suppose that the 'community' always comes up with the best solution to a given problem. A good look at standardisation processes over the years shows it can just produce the least worse one that works.

How do you avoid the echo chamber effect? People doing things because others did them before, then that becoming the 'way'? (You see this a lot with thinks like Stack Overflow... there's many cut and pasted examples of things where a few vestigial config options, or code lines have crept in as the cut and paste cycle continues, or all the websites that happily tell you the rules to drastically reduce the brute force attack space on their passwords.

Personally, I always find the companies that do things differently to be a bit more interesting. So long as you can give well thought out reasons, it often shows you've considered the problem deeper than just getting the off the shelf solution. Sometimes that can allow you to provide the differentiation between your competitors.

Mind you, get it wrong, and you'll just be re-inventing the wheel a lot. :D

[HammerOn1024]

"... standardisation processes over the years shows it can just produce the least worse one that works."

I couldn't agree more. Standards are good to a point. Unfortunately it's like voting for lizards; one has to vote for ones own lizard just so the other guy's lizard does not get elected; to paraphrase Douglas Adams. Plus people use standards as a shield for doing really, really, dumb things.

A case in point; the height above the floor of the average display screen in a video conferencing system is defined by a specification. The problem is, it's way too close to the floor. This leads to the display being blocked by most of the participants sitting around a table. When I pointed this out to the IT folks installing my companies systems they pointed to the specification, installed so that no one could really see the screen without playing "gopher" and started walking away. The result: No one uses the expensive video system.

I asked one of the IT guys that if the specification had said to mount the screen on the ceiling, would they have done that? My boss, who was in the room at the time face palmed, because I was right and he knew I was about to add another notch to my "Idiot IT lizard" belt. As you might expect, the IT guy froze, got mad, red in the face, and walked out, he wasn't able to give the correct response, which would have been "no" since he was specification locked.

If a solution doesn't make sense for an organization, it doesn't make sense for THAT organization, and that's ALL it means.

[teknopaul]

silver bullets

word. And the idea that even if k8s is not a good fit now you should use it anyway incase it is in the future is pretty crazy. There exists a small possibility k8s is never the right solution for a coffee machine.

I am highly sceptical of anyone who thinks _any_ software is a silver bullet.

[heyrick]

"A good look at standardisation processes over the years shows it can just produce the least worse one that works.

Hello, WiFi Alliance, are you listening?

[lostcolony]

All the kids are above average

I always enjoy the argument of using standards. "Look, if you don't do things like everyone else you won't get to benefit from the wisdom of the crowds".

Being like everyone else is the literal definition of mediocre, non-exceptional, etc. You obviously shouldn't deviate just to deviate, but the logic that you'll do things better by just doing them like everyone else is not obviously true, and the idea that you'll somehow be better than average by trying to be like everyone else is obviously false.

[JWLong]

Who says What

One guy says "yes, we use this", then another guy says "no, we don't use this"?

Either way, I wouldn't trust these people to pour piss out of a boot, even if the directions to do so we're printed on the bottom of said "boot".

[Piro]

Re: Who says What

"pour piss out of a boot". Nice. Not heard that one before. The colourful language one finds on El Reg.

[disgruntled yank]

Re: Who says What

@piro

"with instructions printed on the heel." Said to have been said by LBJ of an unsatisfactory client state, 50-odd years ago.

[Anonymous Coward]

Re: Who says What

Must be a PHB's boot then. I hear they are mostly full of piss and wind....

[Claptrap314]

On the surface, K8s has some...disturbing features

Like default networking policy of "allow all". Like "secrets" being entered in base64. And so on. Things like this suggest that there are likely to be very serious problems deep in the system.

In the end, it still feels like it's intended to be used by people whose job is something other than "software engineer". I expect that many, many companies are likely to improve their situation by moving to it. I expect that the bulk of them would do better to have a real software engineer come in and design and build a system for service management.

I'm reminded of a project I owned at IBM. The build requirements were so complex that I was directed to file a software patent for the makefile. By the time that patent went through, I had abandoned make and moved to custom scripts. I see the same kind of thing happening here.

[Duncan10101]

Keep walking

So ... you interview a candidate who seems "Interesting." Then they discover you don't use the shiny-shiny-du-jour. They take this as a signal that something is wrong with your company and choose not to continue. I don't want to rant about the obvious logical fallacy at the heart of this candidate's thinking ... suffice it to say you have had a lucky escape.

[Anonymous Coward]

Re: Keep walking

I was piloting some Oracle data warehousing products many years ago, Oracle supplied a contract 'consultant' who came in, insuted my DBA, refused to engage with my sysops team leader and spent explained to me that the entire approach to the pilot was wrong.

Nobody was allowed to treat my team members badly so I picked up the phone and called my Oracle account manager who dismissed her lack of interpersonal skills as she was such a technical guru.

When I pointed out that she had also stated that we were wasting our time with Oracle Products and should be using Tera data for the pilot he closed the call. Hearing her mobile ring and seeing the look on her face as she was fired was a magical moment. I still had to take the DBA and team lead to the pub for lunch to make up for it though

[yetanotheraoc]

Re: Keep walking

Was she right about using Tera data?

[I Am Spartacus]

Nothing to see here

The beauty and evil of standards is that there are so many to choose from.

We don't use K8 or Docker here either. It just seemed to us to be overly complex for the deployment of a simple JAR. We use a script to build the jar and deploy to a staging area. It works well. We are in total control, and there are no unnecessary layers.

Of course, as with any projects, the STO reserves the right to change his/her mind at any future point in time, with out notice or explanation. So by the time you read this we could be on AWS using Elastic Beanstalk. Exciting being in devops.

[Tom 38]

Re: Nothing to see here

Docker solves a lot of problems to do with environments. There's never just a JAR file, there's everything that goes around it, all the libraries it uses, and the interpreter running the whole thing*. Docker allows you to easily manage all those things. Docker is well worth the effort in my opinion.

k8s is more meh. It's great, we use it for all our deployments, but it is so complex we have a team of 6 managing it and assisting making changes to the clusters. There is a lot of new shiny around the tooling on it, since no-one wants to write manifests by hand, and there's always a new thing coming up - for instance we switched from helm to tanka/ksonnet - not trivial.

Plus, k8s sucks as a developer. Yes, you can use minikube or k3s or a bunch of different solutions and apply manifests on your local machine, but its vastly more complex than just docker-compose. The very best solution for live editing a project in a docker container involves rsyncing files to a running container. So. Dumb. Or you start adding more and more services like Ambassador, Telepresence, Skaffold or Tilt into the mix. So. Complex.

Developers want to run their code in some local setup that's simple to setup and use. Docker containers do that, docker-compose is simple for orchestrating that in development, so we develop in one system and deploy in another. Not so great.

If I interviewed somewhere and they didn't deploy their code using containers, or do local development using containers in some way, I also probably would not take the job (unless of course, the job is helping them move to such a setup and they wanted to pay me lots of money)

* I might be getting some things wrong here, been a while since I did things in Java

[BOFH in Training]

Use the best tool for the job

Simple, as the title says.

Whatever is the new shiny may not be the best tool for your job always.

Just have a process to identify when a new tool is available which will help the job or when an older tool is updated and becomes suitable for the job.

[Plest]

6 of one and half a dozen of the other!

As a veteran 35 years in IT, I've seen so many "fads" just be implemented because those on the project wanted to learn it! Not implemented 'cos is was a good technical fit but simply because the project team all got to bugger off on 6 paid for courses on the latest trendy thing to hit the job market.

I still recall when REST API was the lastest thing, a great technology but when you watch the company dev team do every single process trigger through loads of distinct calls and you have then buy in a scheduling tool to be able to manage 278 REST API calls from a REST call just for copying a file from one UNC path to another to another totally unique REST call that ran a major data load, you have to ask who the hell thought this was a good idea?!

Now don't get me wrong, k8s is a stunning technology and very capable from my limited use of it but so is a toilet brush. A toilet brush is a wonderful invention but you wouldn't use a toilet brush to blend cake mix before cooking it simply because it can be done.

[PM from Hell]

Development fads

I was tech support manager for an organisation where the developers were seen as magicians and the ops / tech support teams as drudges. We were an Oracle shop (Oracle having been specified by the Development Manager) and my team had built up a very high skill level supporting Oracle. We been supporting bought in apps for a number of years before the development team started building bespoke apps. Their initial deliveries were pitiful and my team had to spend a lot of time giving them tuning advice, you can imagine my reaction when the Dev manager announced they were changing development platforms for the next big thing (I can't even remember what he proposed). He had just about got the go ahead to do this when I found out. His business case was based entirely on the fact that the licencing was cheaper than Oracle when I pointed out that most of our apps were Oracle based and my team would be running Oracle for the next decade his face was a picture, just to twist the knife I pointed out that it had taken his team 2 years to get productive on Oracle and the training had cost nearly £100,000. Needless to say that initiative was canned.

[DM2012]

If you think k8s is OTT shiney shiney trendy...

Wait till you lay eyes on Istio

[tip pc]

ACI & SDN

It’s not just server containerisation that is subject to the latest fads. So many “Software Defined Networking” vendors all promising sliced bread, all doing it differently. You then need some kind of orchestration tool to interconnect the differing systems.

How many are running different sdn’s for VMware, data centre, wan and lan?

SDN for VMware makes sense, doesn’t really make sense for the other environments or would be better to use 1 for all (2 if you need ACI)