Building Resilient Microservices from the Fallacies of Distributed Computing | Hacker News Reader

Building Resilient Microservices from the Fallacies of Distributed Computing

datawire.io

42 pointsaustingunter7 comments

"Defensive Programming is a well-regarded technique in software engineering."

No.

No, it's not.

Fail early, fail fast, crash loudly, and then react. That, plus deep validation testing, is how you build robust systems. Trying to soldier on when things are wrong just propagates bad data and bad behaviour into a larger surface area that needs cleanup.

Assert everything, even in production. Capture all failures and action each one (turn a 500 crash into a validation failure 400, etc)

Monitor all logs for unexpected/new anomalies.

Be vigilant around testing. Failure cases are part of the spec, too!

That's how robust systems are really built!

In the section that you mention, the author of the article is comparing techniques for dealing with local function calls vs microservices over the network, specifically in regards to the fallacy that the network is reliable.

With respect to local functions, I think most developers would agree that defensive programming is well-regarded (otherwise this browser would crash on half the web pages I visit).

With respect to distributed system, basically the rest of that section agrees with you, concluding with:

"In fact, using a middleware or services layer that forces engineers to think about their resilience strategies in the face of network failures is quite valuable. After all, the engineers are the best people to decide how a system should behave when things go wrong."

So you basically just re-iterated what the author wrote.

Also, isn't Hystrix a good example of defensive programming in a services environment? Seems to work well for Netflix.

Asserting everything is defensive programming. You don't have to survive errors to be defensive. The defence comes from noticing them.

I find that microservice is just splitting a problem into multiple small problems while the overhead of fixing each problem is uniform, so you end up with a huge technical debt, and eventually fall into dependency issues.

Distributed computing is going to be less efficient than centralized system, now you have multiple vectors susceptible attack...

I find the best architecture is a detached standalone-tenancy meaning a copy of the web application distributed as an image running on a different server assets distributed across different web host providers. 1 clone of your app = 1 domain = 1 customer

This way DDOS attack requires knowing all of your customer's domains which runs your web application, and it dramatically increases the cost of launching a successful and prolonged DDOS attack. The attacker even with a huge bandwidth rate now has to spread it thin across hundreds of your customers website.

Sure your own website hosted on amazon s3 could take a hit but your customers are still able to run their business without drama from foreign state actors or xbox players.

Microservices are good, when you have a big team. Microservices are bad, if you have a small team.

Microservices is about management, not about code.

The reality of microservices, is that we are all forced to use them whether or not the local app is a monolith. Any app dependent on a 3rd party library (analytics, database, webhook) is typically subject to the article's listed network issues. I've seen the issue in 3rd party libraries from each of these categories. So if our app is already architected to defend against errors in 3rd party libs, it's not much of a stretch to apply the same techniques to a couple local microservices.

As another commenter mentioned, Netflix designed Hystrix to guard against this specific scenario. There are important lessons here regardless of our local app design.

Reply on news.ycombinator.com