flow VS fuse

You can actually have "background jobs" in very different ways in Elixir.

> I want background work to live on different compute capacity than http requests, both because they have very different resources usage

In Elixir, because of the way the BEAM works (the unit of parallelism is much cheaper and consume a low amount of memory), "incoming http requests" and related "workers" are not as expensive (a lot less actually) compared to other stacks (for instance Ruby and Python), where it is quite critical to release "http workers" and not hold the connection (which is what lead to the creation of background job tools like Resque, DelayedJob, Sidekiq, Celery...).

This means that you can actually hold incoming HTTP connections a lot longer without troubles.

A consequence of this is that implementing "reverse proxies", or anything calling third party servers _right in the middle_ of your own HTTP call, is usually perfectly acceptable (something I've done more than a couple of times, the latest one powering the reverse proxy behind https://transport.data.gouv.fr - code available at https://github.com/etalab/transport-site/tree/master/apps/un...).

As a consequence, what would be a bad pattern in Python or Ruby (holding the incoming HTTP connection) is not a problem with Elixir.

> because I want to have state or queues in front of background work so there's a well-defined process for retry, error handling, and back-pressure.

Unless you deal with immediate stuff like reverse proxying or cheap "one off async tasks" (like recording a metric), there also are solutions to have more "stateful" background works in Elixir, too.

A popular background job queue is https://github.com/sorentwo/oban (roughly similar to Sidekiq at al), which uses Postgres.

It handles retries, errors etc.

But it's not the only solution, as you have other tools dedicated to processing, such as Broadway (https://github.com/dashbitco/broadway), which handles back-pressure, fault-tolerance, batching etc natively.

You also have more simple options, such as flow (https://github.com/dashbitco/flow), gen_stage (https://github.com/elixir-lang/gen_stage), Task.async_stream (https://hexdocs.pm/elixir/1.12/Task.html#async_stream/5) etc.

It allows to use the "right tool for the job" quite easily.

It is also interesting to note there is no need to "go evented" if you need to fetch data from multiple HTTP servers: it can happen in the exact same process (even: in a background task attached to your HTTP server), as done here https://transport.data.gouv.fr/explore (if you zoom you will see vehicle moving in realtime, and ~80 data sources are being polled every 10 seconds & broadcasted to the visitors via pubsub & websockets).

Elixir has a few interesting abstractions for that: GenStage, Flow, Broadway.

https://github.com/dashbitco/flow

Indeed. I do wish there were a few more "extras" in OTP, because "let it crash" needs some more details in some circumstances.

For instance if you have a system with a user interface and some various components, like say, a database, and the database becomes unavailable, you don't want the entire system to crash. You want it to display an error message to the user and maybe go into some kind of diagnostic mode or other "things are not normal" state.

Something like https://github.com/jlouis/fuse is one approach.

> If you configure it well (the defaults are not always optimal) you can have your invisible mesh of services survive extended outages of the 3rd party APIs it depends on.

This is something that annoyed me a bit with OTP. The basic strategies aren't really enough for that, so you need something like https://github.com/jlouis/fuse

I wrote something like that myself, but it hasn't seen a ton of use: https://github.com/davidw/hardcore