DBConnection Implementation via HTTP

Hello!

I am currently working on an implementation of an Ecto Adapter for XTDB v2.x. 2.x is not a production-ready version, but I just wanted to try to make this project as a way to learn more about Elixir and XTDB as well.

So first of course I would need to create a DB driver for Elixir. XTDB has an HTTP API so I figured I will implement the driver using that as the protocol.

To create a driver, it seems standard that you should implement the DBConnection.Connection behavior. I just have a few struggles with this.

1.) From looking at other DBConnection implementations (MyXQL, Postgrex), they all seem to implement their database-specific protocols via TCP. Then DBConnection is somehow able to keep these TCP connections alive (via pinging?). My question is, how would one implement a DBConnection for an HTTP based protocol, specifically for the ping callback. To my understanding, HTTP is stateless, so I don’t think there is a realistic way to implement ping for this. Should I just leave it undefined and force :idle_interval = 0? Which brings me to my next question…
2.) Is there anything to be gained from implementing DBConnection for an HTTP based protocol? The main value-gain from implementing DBConnection seems to be that it is able to pool and maintain open TCP connections. I don’t know if there are other benefits, but if this is not applicable to HTTP, then should I bother with implementing it in the first place?

Sorry if these are ill-informed questions, as my knowledge of network protocols is very low. Any advice is appreciated, thank you!

There’s “keepalive” in HTTP/1.1 that lets a connection be used for multiple sequential requests.

You’d need to check if XTDB supports it, though.

The goal of connection pooling for this is largely around reducing the number of SSL handshakes required - every time a new connection is opened it takes a bunch of network round-trips and cryptographic calculations.

2 Likes

:wave: @zackattackz

I implemented an HTTP driver for ClickHouse and DBConnection was very useful.

  1. You can GET /status for handle_ping, it would probably be enough to keep the connection alive.

  2. In addition to pooling and keeping connections alive, DBConnection provides a way to make them re-connect with backoff in case things go bad.

Just make sure to use an HTTP client without a pool!

2 Likes

Interesting, I will need to see if it is supported. Thank you!

This is awesome, thank you!! I was looking into using Mint as well, so this is a huge help :smiley:

Looking more into this it seems that HTTP 1.1 keeps connections alive by default, so pinging some known endpoint periodically, like @ruslandoga does in their Clickhouse driver, should be enough to keep the connection alive.