[flagged]
If you don’t understand how the OS works at a very fundamental level, you will write sub-optimal code. This is usually fine, because people are smart, compilers are good, and hardware is absurdly fast. However, the fact remains that it’s probably giving up some bits here and there.
More germane to my point, though, the amount of libraries and tooling and abstractions being spun up to avoid having to run a server seems like more effort than its worth. Serverless sometimes has cost merits, but as a counterpoint, you can get a 1 year RI micro EC2 for something like $3/month. If your app is struggling to run on that, either you have the scale to justify more cash outlay, or your app has performance issues – and now we’re full circle back to the previous paragraph.
- Flame still assumes you have a server somewhere- this is purely for elastic workloads.
- Dynamically scaling horizontally is not something "linux" can do for you: It's a legitimately hard problem because of all the usual distributed systems difficulties.
- Often, there are workloads that are only needed occasionally but also have very different resource needs than the rest of the system. Being forced to not only create a new version of your application, but additionally a bespoke server setup and dedicated VM/Container image for that case takes something that could be trivial in the Flame model, and makes it take hours even for someone experienced.
Basically, you really should've read the article first.
It can, depending on your definition of horizontal, and whether you accept that tools like HAProxy fall under the realm of “Linux administration.”
systemd can spawn N services from a single definition file, which can then register with the load balancer. A small shell script could monitor those service cgroups for basic metrics like memory or CPU utilization, and then command systemd to scale up/down. This is horizontal from the app’s perspective.
I’m not suggesting the above is a great idea, and you’d probably be better off with container orchestration of some kind – but tbf, all that’s doing is encapsulating cgroups and monitoring consumption to command scaling.
Everything has trade-offs. You don’t generally get good, fast, and cheap.