Live data from Hacker News

Building PlanetScale with PlanetScale

planetscale.com

21–23 of 23 posts

Re: Building PlanetScale with PlanetScale

#21
post #4

At Tiqets we have a different repo with SQL migrations using Flyway... ITs as simply as branching that repo, creating your migration. Getting the pipeline green and merging it back. All migrations are automatically executed on production, pre-tested. Reviewed. No DB Administrator, No deployment together with the app. It works quite well.. However, personally I really like not having to deal with schema changes. For e…

This works fine for a lot of cases, but you can run into issues on larger databases. Things like default or calculated columns and indexes can seem like innocuous SQL but end up being incredibly expensive and even acquire locks that block other transactions on bigger databases.

We have some "medium sized" Postgres databases around 1-10TB and something like adding an index to a big table (100s millions of rows) can be quite expensive resource wise

(I'd say "big" is something like 20-50TB+ and small <1TB)

Re: Building PlanetScale with PlanetScale

#22
post #5

Personally, migrations have never been painful enough to warrant this sort of product. The idea of branching databases on the other hand is really cool. Is each branch a full copy? Seems like that would lead to huge storage needs for larger databases.

The database branching idea seems to be popping up more and more. IIRC, TermniusDB supports it too - https://terminusdb.com/ I've not played with PlanetScale yet, but I am planning on playing with TerminusDB in the near future.

Yes - branching both schema and data in terminus - becoming more JSON oriented too with a new document API

Re: Building PlanetScale with PlanetScale

#23
post #5

Personally, migrations have never been painful enough to warrant this sort of product. The idea of branching databases on the other hand is really cool. Is each branch a full copy? Seems like that would lead to huge storage needs for larger databases.

In the case of TerminusDB - which was mentioned in another comment - a branch is just a pointer, so isn't expensive. You basically get free clones, but don't have the risk of lots of messy copies of data everywhere
Post reply on HN