Fix parallel preloading with 'fake' transaction behaviour

[harmon25]

Hi @evadne!
Was playing around with this and tried running multiple preloads in parallel Repo.preload([:employee, :customer]) and it was failing with the following error:

function Etso.Adapter.in_transaction?/1 is undefined or private

Took a look and realized the adapter is missing the Ecto.Adapter.Transaction behaviour.
I presume this is because there is no transaction mechanism we can use for :ets?
Anyway - figured we could trick Ecto into thinking we are never in a transaction - and the parallel preload now works as expected.

Might be some side effects I have not run into, thoughts?

• add transaction behaviour to adapter
• add a test for parallel preloading

[evadne]

Hello @harmon25

It is probably better to not fake support for transactions, and instead use:

Repo.preload(…, in_parallel: false)

Just my opinion here of course.

ALTERNATIVELY… Something like this might work as well

def transaction(_, _, _) do
  {:error, :not_implemented}
end

def in_transaction?(_) do
  false
end

def rollback(_, _) do
  raise NotImplementedException
end

[harmon25]

I didnt love the idea of faking it either, but being able to preload in parallel is nice - and it should work in ets, right?

I like your alternate suggestion - so if someone tries to run Repo.transaction/3 / Repo.rollback/2 they get some good feedback - while still allowing parallel preloading.

I will update the PR and add another test or two!

[evadne]

@harmon25 Indeed, since the requirement of parallel loading is “not in a transaction” (although I do not necessarily agree), our lack of support of transactions with ETS should not be an issue at all, IMO.

[evadne]

@harmon25 Hey, how have you been? 🧠

[harmon25]

Hi @evadne, just a bit busy. Not been working on the project that relies on Etso - but I did update the PR to address your feedback! Hope you are well.

This did not work though, and broke all tests that are writing to the repo.

def transaction(_, _, _) do
  {:error, :not_implemented}
end

So kept that fake transaction behavior to avoid the compilation warning when it is not defined.

Attempting to rollback now raises an exception.

[evadne]

OK, then in that case I think we need to do something else here

[evadne]

I think we can’t cheat so have to do it properly, I found a paper from 10 years ago and I will try to look for something newer.

🤔

Paper is:

Erlang ETS Tables and Software Transactional Memory: How Transactions Make ETS Tables More Like Ordinary Actors — Patrik Nyblom, Ericsson AB [2011]

[evadne]

With Ecto 3.8 this has become much easier to do w/o needing to fake transactions. Have verified on develop that :in_parallel actually executes from various pids now