[Bug]: Entire database wipe when deploying proxy

[aguadoenzo]

What happened?

This is a duplicate of #3035 but it's still happening and too big of an issue to just leave a comment on a closed issue.

Pointing LiteLLM at your database will result in data loss when you re-deploy the LiteLLM proxy.

Here are some logs I found:

⚠️ There might be data loss when applying the changes: 
• You are about to drop the `migratehistory` table, which is not empty (4 rows).
• You are about to drop the `tenant` table, which is not empty (1 rows).
• You are about to drop the `user` table, which is not empty (1 rows).

This is due to LiteLLM blindly updating the schema without taking prisma's behavior and warning into consideration.
See migration script

This is extremely dangerous and will (but more likely already has) result in production data being lost.

The other issue suggests pointing LiteLLM at a different database. While the advice is sound and valid, this is not enough, especially considering none of this is mentioned in the docs

Relevant log output

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

latest

Twitter / LinkedIn details

No response

[ishaan-jaff]

cc @krrishdholakia I believe your prisma migration work will resolve this issue

[krrishdholakia]

Hi @aguadoenzo thank you for raising this issue. We've added support for using prisma migration here

litellm --use_prisma_migrate

# runs `prisma migrate deploy` 

We've also added the initial migration baseline + new updates here - https://github.com/BerriAI/litellm/tree/3ca34a181cae83b8b3c73e5937f07e36509a7f35/deploy/migrations

By end of week, we will have these 2 components connected, so you can just use the safer prisma migrate flow.

I will update this ticket once it's done e2e.