2 ms·
What's notably absent are descriptions of problems with versioning interfaces, failures from network unreliability, or problems managing connecting infrastructu
by karmakaze 7y ago
What's notably absent are descriptions of problems with versioning interfaces, failures from network unreliability, or problems managing connecting infrastructure, or poor delineation of service boundaries leading to undesired change dependencies--which is to say it seems like they executed well. There's no sign they fell into common pitfalls.
There are definitely some good insights here that I don't often read about. The idea that with a sufficient number of microservices (say 50+) you not only treat your instances as-cattle-not-pets you have to treat the service types en-masse as-cattle-not-pets. This requires more automation and organized management as pointed out by the need for tuned autoscaling rules. This requires continued investment into automating things you would do manually if you had 50- services.
The other thing to consider is that going to microservices and back to a monolith is not necessarily a failure. Microservices are good for periods of high change velocity, once a platform is mostly built requiring much less new development consolidation completely makes sense. At all points, we're solving for impedance mismatch, whether that's the org structure, velocity of changes, or numbers of developers vs numbers of deployed units.