Hello everyone ) Are there any plans to add MySQL support?

Not in the near future. For versioned MySQL you can check out Dolt or Dolthub for a managed service. Alternatively, you can move your data to an object storage and use lakeFS :-).

Yes... I can, but I want to use lakeFS c TiDB .

I didn't know TiDB. Let me check it out and see what the options may be. I'll probably get to it tomorrow.

Super. ! ) Thank you.

Hi Denis, IIUC you would like to add a MySQL kv implementation? We love PRs! This should be straightforward based on the existing Postgres implementation. But, because this is a core component, I worry about correctness and about ensuring that it continues to work. We will need a full understanding of things like isolation levels used, and also comprehensive testing. Most of the work will have to happen there.

Hi Ariel, Yes, I understand... But... In this case, the bottleneck is Postgres. I want to remove it. Postgres is very difficult and always has downtime and problems when the master crashes. There is no master-master replication or for a lot of money. Of course, I would like to add and conduct any necessary tests. If someone can coordinate me on this, then I would like to provide normal support for TiDB.

Just to be clear: I want to be that person, and there's a fair chance that I will be. I'm worried about diluting the team effort to support additional kv solutions. I will talk with some people, I want to work out how we support something like this over time. My gut feeling is that we will definitely need a running system test setup ("esti" in the repo). But I hope you understand that we will have to understand what commitments we end up making.

Yes, I understand... Then I'm waiting for the "final decision".

As a first measure, the kV package has an acceptance test suite that can be used to check basic implementation logic

Good news: I talked to the team & we're good to go 🙂 . Let me know when and how you want to proceed! Obviously this is your call. I suggest a VC or just chat.

@Денис Егоров Just my personal opinion, but I prefer you go with the Galera cluster than with TiDB.

Great! ) What are our actions? Am I writing code following the example of Postgres or...? You can communicate, but my English is very bad. The best way is to chat. Yes, we can start with Galera cluster.

Great, let's get started! I think that, since we are talking about adding an SQL-based KV backend, it will be a lot like the Postgres backend. So I would start with reading the Postgres implementation: it lives here.

ok... Is this driver for muscle?

I would guess most of the operations there will be almost the same? Apart from the code that creates a table the first time lakeFS runs, and of course

Open

. The first you probably don't need at first. Also the whole business with the advisory lock looks like needless complexity. The second is to open a pool of connections to the DB; you probably know more about MySQL than I do.

yes, it is used here.

I think what we need to get started it to write mysql/store.go and mysql/store_test.go. The important bit about store_test.go is where it calls kvtest.DriverTest. This runs the existing test of KV functionality. If that passes, we feel much happier about the KV implementation.

To begin with, everything is there and clear. I’ll take a look and it will be clear later whether it’s clear or not...)) Thank you.