Meanwhile, I have worked with a company that had to use MSSQL and they ended-up using two Ecto MSSQL adapters at the same time so they could cover all use cases, which is obviously far from ideal.
It is one of these things that, if we had updated it as we went along, it wouldn't have been so much work, but because it fell behind, the amount of work becomes quite big.
However, there are no plans for a built-in sqlite3 adapter, so your observation still holds true. Sometimes you will stumble upon a part of the ecosystem that is not quite there, and then you need to make a decision between waiting, developing it, or using something else.
I run a SaaS product where the backend is Elixir (it's been going for 10 years, so originally written in pure Erlang) and part of it needs to do a lot of http requests. It's a bit of a unusual use case, as I need to do 1000s of requests per minute to different servers. It's not feasible to keep that many connections open, so the connection needs to be recreated each time. Using hackney (the #1 Erlang http client wrapper) I found that some requests often took much longer on average than other requests. I narrowed it down to being a problem somewhere on the client side, but Erlang doesn't provide any tooling for digging deeper into the lifecycle of a HTTP request. I figured maybe it's a problem with TLS and then found Erlang (I think its been improved in v22) had very poor performance when verifying SSL certificates, so I had to disable that. I switched to mint which is a wrapper written in pure Elixir which helped a lot, but still it a left a lot to be desired.
I wrote up a quick test in Go and found doing the same requests, it didn't have any performance issues or timing oddities, and the code was much simpler. I've hbeen working on extracting that part of the system into a Go app, called via gRPC. Unfortunately there is no official gRPC library for Erlang, and the only third party one has limited features that I've had to work around. So it's a bit painful, but don't get me wrong I love using Elixir and Erlang :-)
What if you called it via Port instead of gRPC?
We are also using Elixir at our SaaS for more than 2 years in production. There are times (although rare) when Elixir performance is just not enough. For those cases we tend to use Rustler, which makes integration with Rust code pretty safe and straightforward. Did you consider Rust instead of Go?
As for SQLite, we use it internally a fair bit for ETL processes, and as a starting point for most of our applications. 90% of the business apps that we end up creating never need anything more than SQLite -- the 10% that do either end up on MSSQL (if they have to tie in to our main databases) or on Postgres (for everything else).
Also, as a note, my first production Elixir service didn't talk to a relational DB at all, and was basically a thin API wrapper around ElasticSearch... others I know have basically used it just for realtime features connecting to browsers with WebSockets for push via Phoenix Channels (or now LiveView)... so even without any of those database adapters you may well find a use case it's a perfect fit for, and where lack of those adapters is not an impediment! Either way, good luck!
In one different project, there was a desire to deploy a Phoenix API natively to Windows clients. After poking and spending a few days looking at years-old Stack Overflow posts with no activity, we ended up pushing back on the requirement and going with a containerized solution.
Elixir/Erlang/BEAM is a wonderful stack and I probably use it for 50%+ of my software, but I would not choose it in an environment where interop with MS technology was a requirement.
Having been down this path, without the time to dedicate fixing/working on a driver, I wouldn't even consider a non-"blessed" language in combination with MSSQL anymore. It was too much pain.
Ecto v3 has been out for more than a year now, and there are still only two supported adapters.. MySQL and Postgres. In my environment, I use a lot of MSSQL and SQLite in addition to Postgres, and those adapters haven't been successfully ported over to Ecto v3 yet.. in looking at some of the threads related to porting efforts, it appears that it must be a fairly daunting process.
I always get to that point in my technical evaluation and wonder whether I ought to sit back and wait for a while yet, until the database stuff catches up and solidifies. I realize Phoenix can still use Ecto v2, but that just seems like it would add legacy dependency issues.. In actual practice, is most everyone using Elixir and Phoenix just sticking with Postgres/MySQL? Is there some other mitigating method that folks are using that makes Ecto support a non-issue?