> Note that the system uses a client-server architecture but isn't really distributed.
Yeah, I kind of agree. The thing is, it's orchestrating multiple containers to do the job. I can't figure out why you couldn't just have one container.
Yeah, I kind of agree. The thing is, it's orchestrating multiple containers to do the job. I can't figure out why you couldn't just have one container.
(Of course, it's completely possible to build a single container which runs all of the services in parallel, but then monitoring/scaling/availability/etc are more difficult to handle.)
Having done both, I'm always surprised when people say that it is more difficult to handle running N-copies of the same container than a heterogeneous set of containers. Sure, at really large scale you can glean efficiencies from it (which is why one might do it), but there's a lot of complexity that comes with that.
Modularity of code is a common explanation for using a multi-service architecture, but it's not a particularly strong one (services are just one way of modularizing code). In terms of OSRD services being accessed by external applications, it doesn't seem like that would work particularly well, given the coupling between the pieces. The only pieces that would make any sense to me for outside applications to directly in would be the PostgreSQL database... and if that were the plan, I'd just host the whole thing in PostgreSQL.