Document the process anti pattern of sending large data

[PragTob]

As per usual, thanks to all of ya for your great work!

Follow up to/extension of #13173

First draft, happy to adjust and extend it in many ways. I had trouble coming up with a simple example as we needed a bunch of data to make sure it's not good. I could have employed the first anti pattern myself and did repeated statistics calculation, but that'd have been worse :sweat-smile:

I remembered José's comment around sending along the conn and figured it's central enough in elixir to not throw anyone off.

If someone has a better example, happy to redo it!

Thanks y'all! :green-heart:

[josevalim]

The original anti-patterns mentioned exactly this, sending data between processes, but it was removed because it is hard to precisely say when you are sending large data. Sometimes you do have to send it and there isn't much we can do.

Therefore, can you please rephrase this as accidentally capturing large data on spawn? I.e. without the sending considerations? Thank you.

[PragTob]

@josevalim 👋 Thanks for the context, I was unaware of the previous version.

Rephrasing to validate my understanding:

• Rewrite section to focus on spawn/Task.async aka where the function captures the large data (to be fully clear, should I write it with spawn or Task.async as I think the latter is more common)
• Remove mention of this problem around send as it's ambiguous what large data is and when it's needed/vs. not needed

Happy to do that! Question though, shouldn't we at least mention send/2? While most people will learn about the copying behavior on their elixir/erlang journey I couldn't find it in the elixir docs (at least where I checked, Process, send/2 and the Process getting started mentions it's "is stored in the process mailbox" which could be more explicit, however a beginner guide is probably not the best space for this). I don't mean the mention of the "anti-pattern" but that messages are deep copied to the receiving process. Wouldn't want to litter everything, but maybe Process docs?

[josevalim]

Hrm. It is fine to mention send or even GenServer.call, but the framing matters. The issue not really sending large data, but sending unnecessary data. Don't send or pass as argument to GenServer.call/GenServer.start_link a large data structure, only to pluck part of it from the server. Pluck the parts that you need on the caller/client.

[PragTob]

Makes sense, thanks for the clarification! I'll get back to it later today!

[PragTob]

Thanks for the feedback and guidance! New version is up, hope that that's the direction you envisioned. Happy to adjust and iterate in any case :) 💚

[Nezteb]

Some minor capitalization/wording suggestions. Regardless, I love this content!

[PragTob]

@Nezteb thanks for taking the time to review & make suggestions!

[PragTob]

And a big thank you to @whatyouhide !

[v0idpwn]

Looks great. I suggest also adding an example (either here or on GenServer docs) about doing this when sending messages. While it's way more obvious, it also happens :)

[josevalim]

💚 💙 💜 💛 ❤️

[PragTob]

@josevalim thanks for the merge and as I don't think I'll open another issue/MR that soon: Happy holidays! 💚