Sqldef: Idempotent schema management tool for MySQL, PostgreSQL, SQLite

https://sqldef.github.io/

by Palmik • 6 months ago

263 61 6 months ago

sqldef.github.io

davidkwast• 6 months ago

Wow. I saved the link for emergencies. And I sent it to all my team. It is like a diff tool but to advance the schema.

evanelias9• 6 months ago

Personally I've always called this style "declarative schema management" since the input declares the desired state, and the tool figures out how to transition the database to that state.

sqldef is really cool for supporting many database dialects. I'm the author of Skeema [1] which includes a lot of functionality that sqldef lacks, but at the cost of being 100% MySQL/MariaDB-specific. Some other DB-specific options in this space include Stripe's pg-schema-diff [2], results [3], stb-tester's migrator for sqlite [4], among many others over the years.

The more comprehensive solutions from ByteBase, Atlas, Liquibase, etc tend to support multiple databases and multiple paradigms.

And then over in Typescript ORM world, the migrators in Prisma and Drizzle support a "db push" declarative concept. (fwiw, I originated that paradigm; Prisma directly copied several aspects of `skeema push`, and then Drizzle copied Prisma. But ironically, if I ever complete my early-stage next-gen tool, it uses a different deployment paradigm.)

[1] https://github.com/skeema/skeema/

[2] https://github.com/stripe/pg-schema-diff

[3] https://github.com/djrobstep/results

[4] https://david.rothlis.net/declarative-schema-migration-for-s...

tln• 6 months ago

I like how Drizzle provides several options for the migrations.

1. DB is source of truth, generate TS from DB 2. TS to DB direct sync, no migration files 3. TS source, Drizzle generates and applies SQL 4. TS source, Drizzle generates SQL, runtime application 5. TS source, Drizzle generates SQL, manual application 6. TS source, Drizzle outputs SQL, Atlas application

stemchar2• 6 months ago

Personally I've always called this style "declarative schema management" since the input declares the desired state, and the tool figures out how to transition the database to that state.

Personally I've called it a mistake, since there's no way a tool can infer what happened based on that information.

canadiantim3• 6 months ago

What do you think of the approach that pg_roll takes?

Seems like the top contenders, at least for Postgres, are:

sqldef: https://news.ycombinator.com/item?id=46845239

pgschema: https://github.com/pgschema/pgschema

pg_roll: https://github.com/xataio/pgroll

atlas: https://github.com/ariga/atlas

grate (minimal SQL-based migrations): https://grate-devs.github.io/grate/

pg-schema-diff: https://github.com/stripe/pg-schema-diff

results: https://github.com/djrobstep/results

https://david.rothlis.net/declarative-schema-migration-for-s...

dewey• 6 months ago

I'm always in the market for new sql tooling, but I'm wondering what's the use case there?

Isn't it much quicker to write a one line migration vs copying the DDL, then adapting it to the desired state and then getting getting the migrations from that tool? Or am I misunderstanding something?

evanelias9• 6 months ago

Be sure to look at the actual sqldef command-line tool, not the trivial copy-and-paste demo on their website. Declarative schema management is best used combined with a Git repo.

In the big picture, declarative schema management has lots of advantages around avoiding/solving schema drift, either between environments (staging vs prod) or between shards in a sharded setup (among thousands of shards, one had a master failure at an inopportune time).

It's also much more readable to have the "end state" in your repo at all times, rather than a sequence of ALTERs.

There are a bunch of other advantages; I have an old post about this topic here: https://www.skeema.io/blog/2019/01/18/declarative/

It's also quite essential when maintaining nontrivial stored procedures. Doing that with imperative migrations is a gateway to hell. https://www.skeema.io/blog/2023/10/24/stored-proc-deployment...

montroser• 6 months ago

Yes, we've used skeema for this for many years, and it is just plain lovely. Putting into source control your desired end state is so much more intuitive and understandable than accumulating migrations. In a way it's like the difference between jQuery and React -- you just say how you want it to look like in the end, and the computer does the work to figure out how to make it so.

netghost• 6 months ago

Out of curiosity, the post you linked mentions that it won't work for renames. What's the approach for these and other types of procedural migrations, such as data transformations (ie: splitting a column, changing a type, etc.)

With a declarative model, would you run the migration and follow immediately with a one off script?

Pxtl• 6 months ago

MS' Sql Server Data Tools is such an abominable garbage fire that I have no interest in these kind of tools. Besides being a buggy mess, it's very often insufficient - you end up having to maintain migrations anyways because often you have to inject data into new tables or columns, or rename columns, etc.

branko_d• 6 months ago

often you have to inject data into new tables or columns

No tool can help you with that, simply because this kind of data migration depends on your particular business logic that the tool has no way of knowing about.

While SQL Server Data Tools has its warts, it has been immensely useful for us in making sure every little detail gets handled during migration. That doesn't usually mean that it can do the entire migration itself - we do the manual adjustments to the base tables that SSDT cannot do on its own, and then let it handle the rest, which in our case is mostly about indexes, views, functions and stored procedures.

After all that, SSDT can compare the resulting database with the "desired" database, and reliably flag any differences, preventing schema drift.

edoceo• 6 months ago

Anyone with real-world usage of this one vs atlas. I've used atlas but it seems to be moving away from FOSS (paywalled-fearures). Its also struggled with some migrations that were a bit more than not trivial (like trying to play changes in the wrong order)

nodesocket• 6 months ago

Looks pretty cool. Lately I've been using ChatGPT to generate SQLite schema migrations and it works shockingly well. I give it the original schema and new schema and it generates the statements including caveats and gotchas to watch out for.

canadiantim3• 6 months ago

Looks great! Could this work with duckdb?

drdaeman• 6 months ago

Looks like this is only useful for empty databases. Which severely limits possible use cases.

Schema management is only a small part of the problem, and I don’t think this tool handles data migrations. E.g. if I reshape a JSONB column into something more structured, I don’t think it would be able to handle that. Or if I drop a column the backwards migration it generates ADD COLUMN … NOT NULL, which is obviously unusable if the table has any data in it already.

wener• 6 months ago

I use https://github.com/ariga/atlas for this, migration based and schema base both has there good and bad side, I prefer both even in one project, schema based can make dev faster, eaiser, migration based make feel reliable.

a8m• 6 months ago

Ariel from Atlas here. This setup is pretty common (declarative locally, versioned in real environments). Since migrations are auto-generated (by Atlas) on PRs, most developers never touch the versioned workflow directly.

See: https://atlasgo.io/concepts/declarative-vs-versioned#combini..., and https://github.com/ariga/atlas-action?tab=readme-ov-file#ari...

gwking• 6 months ago

I wrote a very rudimentary schema and automatic migration system for SQLite. One problem that I ran into recently was deploying changes that spanned two migrations, because the tool doesn’t know how to step through commits to do successive auto-migrations between schema versions. I guess there are rather obvious ways to handle this if you generate and then commit the full migration sql for each schema change. Nonetheless I’m curious if this is a problem you have had to think about, if you find it interesting or if it sounds like a bad path to go down, and if atlas does anything smart in this department. Thanks in advance!

Palmik• 6 months ago

Yeah, after looking more into sqldef and alternatives I stumbled on Atlas too and I like the explicit support for migration based flow for exactly the same reasons. I want to know exactly what kind of migration will be applied to my prod database beforehand.

panzi• 6 months ago

This is very cool!

Also means I can stop with my hobby project that was supposed to do the same. Wasn't far along and haven't worked on it in months anyway.

So I'll spend my time on another hobby project then that also solves something that is already solved 100 times over, but I don't like the other solutions (simple log monitoring including systemd and sending error emails if something is found).

stemchar2• 6 months ago

I renamed a column and it added a new one.

andrewg• 6 months ago

From the docs: https://github.com/sqldef/sqldef?tab=readme-ov-file#renaming...

You tell it what’s being renamed with a special comment.

systems• 6 months ago

is there anything (open source) similar to microsoft database project but that would work for postgresql

i like the following about it 1. database schema is regular code 2. make schema change declaratively 3. packaging (.daspac) and deployement script

most open source tools , seem to be after the fact tools, that do diffs ms db project, handle the code from the start in a declarative, source code managed way

waynenilsen• 6 months ago

This post from 2022 is the one I keep going back to

https://david.rothlis.net/declarative-schema-migration-for-s...

chmaynard2• 6 months ago

tianzhou• 6 months ago

If someone is looking for a more comprehensive coverage for Postgres, please check out https://github.com/pgschema/pgschema.

I built this last summer and thought it was fairly complete. And it turns out I was naive. 6 months later, I have resolved ~100 issues reported by the users.

quantike• 6 months ago

Incredible tool, definitely going to get a working poc for this with my team. Curious how this can apply across many databases, say in a cluster, checking for discrepancies etc. Wondering if that is a supported feature?

Nice work, really happy to have found this today.

cjonas• 6 months ago

This looks really cool for schema migrations but how does it handle updates/inserts if you need to move actual data as part of the migration?

ForHackernews• 6 months ago

Reminds me of Migra: https://github.com/djrobstep/migra

canadiantim3• 6 months ago

pg_roll from xata also a worthy consideration

magicalhippo• 6 months ago

We've rolled our own variant of this, which uses an XML file as input. Just easier to parse than SQL schema definitions. The tool compares the schema defined by the XML with the DB, and applies changes as needed.

We used Sybase SQLAnywhere, and a complication there was that if you had materialized views against a table, you had to drop and recreate that view when adding or removing columns in the table. And when you recreate it, you of course have to remember to recreate the indexes on that materialized view.

Tracking this in case you have multiple materialized views touching multiple tables became a bit tricky, and you don't want to do the "dumb" thing to just always drop and recreate, or per-column, since some of them might take an hour or so to recreate and reindex.

The tool has some built-in safeties like never dropping columns if it's missing (have to add explicit element for that in the XML), and only performs safe column definition changes, ie integer to varchar(50) is safe, integer to varchar(3) is not, etc.

It really made database migrations very painless for us, great since we've have hundreds of on-premise installations. Just modify the XML, and let the tool do its job.

And since its our tool, we can add functionality when we need to.

yearolinuxdsktp• 6 months ago

How is this better than Liquibase? Isn’t there a risk of missing schema evolution steps when you only diff current vs target? (Because current can be v1 and target v3 and you might have lost column migration logic from v2)

rswail2• 6 months ago

How does this compare to Entity Framework (for dotnet) migrations or sqitch/liquibase.

I get the declarative nature of this, but migration of schemas is not purely declarative, particular on large production databases. An ideal schema manager would understand the costs of particular migrations (perhaps by using the table stats and EXPLAIN) and apply that back to the migration strategies so that downtime is minimized/eliminated.

Adding or remove columns or indexes can trigger major database table scans and other problems, especially when partitioning conditions change.

exceptione• 6 months ago

The bigger problem is that data can be part of a migration. A diff is far too rudimentary.

If I split a Fullname into FirstName and LastName, a diff will only tell half of the story. In EF Core, you will adjust an Up and a Down generated method to make the change reversible, plus you deal with data transformation there.

So I would love to know how people handle that without an explicit notion of migrations.

morshu90012• 6 months ago

Yeah this sort of thing has to be manual, unless you want to roll the dice with an LLM

morshu90012• 6 months ago

Pleasantly surprised this post was about a neat little tool and not another migration manager. This always seemed like a weak spot in regular SQL. Wish it were declarative like Spanner DDL.

When using Postgres, I just try to keep my schema script idempotent and also work with any version of the DB. Like I start with a CREATE TABLE IF NOT EXISTS, then if I add a new col, it goes into there and also a separate ALTER. But at some point it gets cluttered and I delete the ALTERs once things are stable. Maybe something could hit the fan, I have to restore an old backup from before the last purge, and this tool helps me make it compatible quickly.

uwemaurer• 6 months ago

This is great! It would be a cool feature to integrate with SQG. there you currently need to provide the sequence of migration steps to get the full schema.

https://github.com/sqg-dev/sqg/

froh42• 6 months ago

It seems to try to parse the DDL itself, right? I tried only with a few of the PostgreSQL things we use in our schema and and failed parsing them

e.g. CREATE TABLE books2 (LIKE book INCLUDING ALL);

So I assume it is not Progresql DDL it uses but "something close".

nopurpose• 6 months ago

Is there a good tool for background migrations?

For example add temporarily nullable column to a large table, deploy new code which starts writing to the new column, in background populate that column for existing rows in batches and finally alter column to be mandatory non-nullable.

Another example of non-trivial schema management case is to make schema change after new version rollout completes: simple migration at the start of the container can't do that.

It must be a solved problem, but I didn't see a good tool for it which would allow expressing these imperative changes in a declarative way which can be comitted and reviewed and tested along the app code. It is always bunch of adhoc ugly scripts on a side and some hand waving deployment instructions.

tracker12• 6 months ago

I tend to prefer to hand-roll schema migrations... but I use grate[1] for the most part. That said, I've created similar tooling for different scenarios.

1. https://grate-devs.github.io/grate/

Pretty easy to setup/use in a dev environment as well... see docker-compose.yaml and run/dbup script.

https://github.com/tracker1/FastEndpoints-SqlJobQueues

anymouse123456• 6 months ago

I was looking for a dead simple migration tool some years ago and didn’t want a random runtime (and version) dependency to maintain my database.

Found shmig and it’s been really fantastic.

https://github.com/mbucc/shmig

chmaynard2• 6 months ago

Minor quibble: sqlite3 is not a database, it's the CLI for SQLite 3.x

DANmode• 6 months ago

I’ve seen it written this way in older sqlite management tools.

lun3x2• 6 months ago

Tried it out now with SQLite to see what it does with difficult migrations, like adding a foreign key constraint to an existing table. Looks like it just generates invalid migrations, eg:

`ALTER TABLE books ADD CONSTRAINT fk_books_author FOREIGN KEY (author_id) REFERENCES authors (id)`

Which is not valid in SQLite (https://www.sqlite.org/lang_altertable.html)

Akronymus• 6 months ago

So you have to drop and re-add a column when adding foreign keys?

lun3x2• 6 months ago

The SQLite docs have a section about it, the idea is: 1. Create new table with foreign key constraint 2. Copy data from old to new 3. Drop old table 4. Rename new table to old name

gavinray• 6 months ago

ralferoo• 6 months ago

Sadly, it doesn't handle dropping columns well in sqlite. Actually, it was news to me that DROP COLUMN has been added into sqlite somewhat recently, but certainly the version you'll get on most devices doesn't include it.

But anyway, from the example for sqlite, I added "x integer not null," to the original table, and was greeted with "-- Skipped: ALTER TABLE books DROP COLUMN x". Then ticked the "Enable DROP" and got the same, except with the line uncommented.

Such a shame, this is one thing that would make a lot of difference. FWIW, the standard way of doing this is creating a temporary table with the new data, dropping the original table, and recreating the new table, copying the data across and deleting the temporary. It's kind of a shame that it doesn't automate that, and this is one of those things that's incredibly easy to mess up if you're not careful, especially if there are constraints involved.

If it just does the easy stuff, you might as well just do it by hand.

Sqldef: Idempotent schema management tool for MySQL, PostgreSQL, SQLite

Related Stories