Or is there still too much conflict for the truly shareable objects?
It is perfectly conceivable that an alien species with different brain characteristics would draw their microservice boundaries at very different points, possibly varying by a whole order of magnitude in either direction.
I don't agree that there are organizational causes even though that's oft repeated as one of the main reasons. I have never seen a one-to-one relationship of microservices and teams irl. Most teams own many services and some services get touched by many teams. In big companies, there are also microservices with zero owners.
What could be a sensible architecture for large orgs has become more of a cult-like behavior that drags everyone down.
I draw the opposite conclusion here. In monoliths, someone else might want the code to stay in the codebase. I don't want to ask everyone about it. Everyone's responsibility is no-one's responsibility, and the code stays as is.
In a microservice, I publish an interface, and I'm free to tear up all the carpet behind it as long as it keeps doing its job as advertised. Knowing whether public routes are still in use is still a problem, so you need a minimum level of metrics or logging before you can retire an endpoint.
How many HTTP calls are made during one user request and how long does it take? (This concern brought down one of the early microservice projects with a 10-minute page load time)
What are the race conditions? User removes billing info at the same moment as making a payment - what happens? Is the payment marked as successful but not actually billed?
> Knowing whether public routes are still in use is still a problem, so you need a minimum level of metrics or logging before you can retire an endpoint.
> What are the race conditions?
Everywhere. Just fucking everywhere. I ran a workshop for my team (4-5 backenders) with a simple, pared down example with reads and writes within a 'transaction'. I then demonstrated that LINQ breaks its transactionality, and running in READ-COMMITTED (which everyone does) also breaks transactionality. Either of these conditions is enough. My team walked away understanding this.
But we left the idea there to die, because you cannot change these things politically. No monolith owner is going to even pretend to demonstrate thread safety.
The lines of defence are:
* He will assume it's thread safe
* He will argue against change "for code consistency"
* He will argue that it makes the code simpler and easier to understand
* He will introduce a change which he assumes will fix it, but clearly only makes the race condition harder to diagnose
* He will argue that it doesn't matter because race conditions are rare anyway.
It's all academic anyway - in a monolith you won't get past the first line of defence anyway.
type ThingDoer interface { DoTheThingIDontCareHow() (result, error) }
func NewAmazingThingIMade(theThingINeedButIDontCareHowItDoesIt ThingDoer) { // You receive the superduper thing and dont care about it in any other way }
func (a *Amazing) DoSomeThing() (result, error) { return a.DoTheThingIDontCareHow() }
Nothing about "monoliths" prevents or hampers this development style. Once someone decides to do the Thing in another fashion, they are free to do so and you wouldn't have to change a thing.
All problems you experience are organizational, not technical. It's psychology we should be studying, not computer science.
Edit: In fact you introduced a network boundary which can fundamentally only complicate matters, not make them simpler.
Also a monolith may (actually, must) have internal public interfaces through which components must communicate, and those contracts are more enforceable, amenable to static checks, and much higher performance that microservices’ network interfaces.
Then they can retrieve it from source control? Leaving dead code in the codebase because "someone else might want it" sounds like nonsense.
> This is complete lack of experience on the authors part
Actually I'm going to turn the accusation back on yourself: if you think reasoning about a system's entire graph of behaviours from runtime traces is trivial, I don't believe you've worked on a truly complex system
The range of what a program can do is much larger than what a program does do in your two week Datadog aggregation window
You can't capture the full range of behaviours and edge cases unless you are willing to ingest billions of events without lossiness
Metrics have limitations of cardinality. Logs need to be reduced or else you are in archive rehydration hell
There will be code paths that have never been triggered that you still need to reason about as part of analysing a system, eg in threat modelling
> I get why you think I might not have worked on large systems
I didn't, necessarily. But it looked like an excessively strong claim, combined with a challenge to the author's competence. I wanted to call that out.
Static analysis doesn't go away. Metrics just covers the gap of "is this used by external callers", and nothing more than that. You need that to identify dead routes in a monolith anyway.
If you like static analysis, you should be able to spin up new code in a language more amenable to it than the incumbent code. Not doable in a monolith.
* You can't change anything of substance. The more people touching the same codebase as you statistically puts you way lower in the pecking order. Changes become political, not technical.
Sometimes the forces of stupidity cancel each other out. It was a 10+ year-old C# monolith, and the top dog wanted to switch to Python because "AI works better with it". I didn't have to waste any time worrying or arguing, since him keeping the system as a monolith made this change impossible.
There are two things monoliths can't easily offer:
* Using different frameworks, languages, etc. But in my experience, it's pretty rare for a team to use many programming languages at once. Usually, it's just a few highly performance-sensitive services that need to be written in another language (say, a proxy in Rust while the rest is in Python). For that, I prefer an architecture with one main monolith plus a few high-performance satellite services. No problem there.
* More optimized scaling in certain scenarios. Say I have a module that processes files and can use all available CPU. I might want to put it in a separate container on another node so that the processing doesn't destabilize the core web server. Technically, monoliths support this too, just run the monolith in a different mode (say, behind an `--image-process` flag of sorts), and you can schedule it on another node in the same way. The only downside is that it may use more RAM than necessary for the extra binaries or scripts that won't be used
What else am I missing?
* NUMA optimization - on multi-CPU architectures, I/O and/or memory bottlenecked software takes a pretty significant performance hit if you let is run on both CPUs, and this has knock on effects that also degrade the performance of other software running on the same machine.
* Network namespaces - if you're e.g. running an internet crawler, and want to assign one host machine multiple public IP addresses, this is much easier if you use something like ipvlan.
* Stateful long running process lifecycles (e.g. crawling) that coexist with the need for live redeployments of other parts of the system.
Teams sure, but your whole org? Nothing wrong with all teams using same tools and language, nothing wrong with them choosing their poison. Of course your monolith can have multiple build systems.
But I think this is beside the main point of the article anyway, and it’s the part about how microservices help create boundaries that mirror organizational boundaries.
There’s probably more off the top of my head. Not that I necessarily advocate for the micro in services. Just the sliding scale of macro-micro services.
These are more problems in larger companies.
Genuinely useful method for abstraction, concern/dependency separation, scaling, and so forth, for teams & projects that genuinely need what they offer.
Davids talk makes some good arguments, but does not fully acknowledge low-maintenance infrastructure at scale. =3
"10 tips for failing at microservices" (David Schmitz)
In our software project, we have had many issues with OOM and golang panics in a goroutine killing our critical endpoint even though the fault was most times caused in the administrational endpoints. By separating the critical endpoint from our administrational ones in two separate services, we reduced the criticality of those issues and therefore also the number of hotfixes we needed.
I thought that was a "microlith".
For a big software company, you don't have to ship a single behemoth monolith, but you don't have have microservices so small that a team of 3 has to manage 10 services.
You can have "right size" services. I don't like to call them micro because they can be fairly big and do multiple things, as long as it makes sense to have them in a single service.
Sometimes you want to have some shared infra abstracted behind a single interface and a single auth model.
Having a service gives you that single isolated unit/abstraction.
In practice it’s always a scale. Some services will be bigger and some will naturally be split out.
I think also having the right server platform that makes it trivial to deploy/standup infra quickly makes this much easier.
https://github.com/GoPlasmatic/Orion
Design principles: * Isolate business logic from service framework using declarative coding * Make the logic representation machine-writable so AI is a first-class author * Treat business logic as versioned data, not compiled services * Ship governance with the platform, not with each service * Invoke services in-process, not over the network * Modular monolith — export and scale a service independently when needed * Push integrations into declarative connectors, keep logic pure
The real—and the one thing you can't replicate—is data isolation.
Of course there are other benefits, but it's data isolation is the differentiator of micro-services and other types of architecture.
Inside a monolith it's easy to answer questions like "Which dependencies are we shipping?"; or "Is this piece of code still used?” Once the code is spread across dozens of independent services, those answers become much harder to obtain.
It turns out, every team thinks they got the smartest people, and they all hate the idea of collaborating with the inferior teams.
Yeah, you will need microservices to bandaid your dysfunctional org. You'll keep going, slowly.
Literally the only reason I can imagine for microservices.
"Microservices" is[0] a tech industry fashion trend. It can be a real solution to a problem, too, but it's like how throwing out all your shit is the solution to depression now because a nice lady in a book said that you probably don't want all that stuff anyway. The solution is applied/suggested because it's trendy, not because it is necessarily the best fitting solution.
Like many buzzwords, 'microservice' is difficult to define. "You know it when you see it" is a euphemism for something devoid of meaning[1]: a null concept. This is a feature and not a bug. And where the term does have meaning, it's always relative -- so everyone can have microservices, and argue about who does.
[0] I do mean 'is', as opposed to 'are'
[1] it can also be the case that the speaker just doesn't know what they're talking about, and perhaps that they know it has meaning to others, social constructs, &c. To avoid spending the next three years working on this comment, I'm assuming that's not the case above.
[x] Conspiracy-theory rabbit-hole: buzzwords are like conspiracy theories in that they rely on vagueness or being undefinable. I guess my point is that concrete ideas aren't really suited to being buzzwords. Only the vague/relative/broad ones get selected (as in natural selection, not by the shady cabal of buzzword pickers), just like how only the best impossible contradictory stories tend to become popular conspiracy theories. ... actually, mostly I just thought "conspiracy-theory rabbit-hole" was a funny heading for a tangent about conspiracy theories.
[1]https://www.makonea.com/en-US/blog/how-did-microservices-arc...
So you might be right.
Microservices isolate them quite literally and lets them work in peace and quiet.
The system as a whole is completely batshit crazy of course, but for some teams, this can turn an unworkable situation into "we might survive this" which is when push comes to shove all that matters.