Object storage is all you need
tigrisdata.comThis article doesn’t present much motivation for why you would do this.
To me it reads a bit like, how we built our office without desks: it turns out if you stack two chairs on top of each other, you can balance your laptop on the top and you’ll also have a shelf on the bottom for your things.
> it turns out if you stack two chairs on top of each other, you can balance your laptop on the top and you’ll also have a shelf on the bottom for your things.
But that sounds kinda unstable and doesn't really provide the full benefits we expect from a...
Oh.
> This article doesn’t present much motivation for why you would do this.
For those working with databases, it's easy to come up with a bunch of reasons. ORM is a necessary evil when working with RDBMS, as are things like schema migrations that can easily result in loss of data (I.e., dropping columns or tables).
What if all we need is dumping a big old JSON in a container?
This idea is very enticing. The scale of this whole NoSQL thing is pretty telling.
The blog presents a thought provoking question: what if we don't actually need a full-blown database, and instead we only care about things like unique constraints, transactions, indices, and history tables.
What if we only need a subset of those?
> ORM is a necessary evil when working with RDBMS, as are things like schema migrations that can easily result in loss of data (I.e., dropping columns or tables).
Hmm, neither of those been necessary for me. ORMs is a choice, you can choose not to and still have a proper design and architecture and not suffer from that choice. Requires you have programmers who know how to use SQL, which seems less and less common as the days go by though, so understandable most reach for an ORM.
Schema migrations that can "easily result" in data loss is even easier to avoid though. Don't drop the column/table in the same step you copy data to the new place, or do the migration via application code which gets cleaned up later, to catch most of the stuff before one bigger copy, and then the eventually DB cleanup.
Guess it depends on how careful you want to be, or how careful your agent lets you be, I guess. But it's definitely possible to avoid both of those "necessary evil", which I guess makes them just "evil".
Object databases exist - they don't magically solve data(model|schema) migrations.
See:
Articles like this remind me of that Innovation Tokens article.
Your time and attention is precious as a developer. I'm absolutely sure it's possible to implement uniqueness constraints, transactions, indices, and history yourself, but is that really the most valuable use of your time? There's probably not a need for you to have a unique solution, so you're quite literally just re-inventing something someone already had for not a lot of benefit.
Wouldn't your time be better spent actually solving the problems that whatever you're building is supposed to solve?
Development of video games is the final boss of this one.
If you want to kill someone's dream of publishing a game, encourage them to build their own engine from scratch. I cannot think of a more malicious piece of advice given how effective it is, statistically speaking.
The exceptions to this are so incredibly rare. Virtually all of the in-house engine work at indie scale has been replaced by Godot in recent years. It used to be something like 10-11% of studios were in-house fully custom. Now it's probably closer to 1-2% fully custom and 8-10% on Godot. Having a reasonably stable OSS option has removed the last major argument that I am familiar with for simply using what already exists.
This is frustrating to read. Tigres is built on FoundationDB, but doesn't expose all FoundationDB operations like transactions, range reads, and get mapped range. They go through all sorts of complications to handle these issues, including a database for caching (and they don't consider thundering herd problems).
What if you just ran FoundationDB instead?
I've occasionally pitched the idea of exposing transactions for object storage based on bucket forking. I think the idea is finally starting to land, but it's a really "out there" usecase that is likely hard to conceptualize until it's usable in some form.
> What if you just ran FoundationDB instead?
Would that let you do collaboration blog-posts with another VC-funded startup though?
Priorities...won't people think of the really important priorities. :-)
FoundationDB is only as metadata store.
Also, the database is the last thing you want to re-invent unless your business is explicitly building a db (even then it's best to re-use a db like all the postgresql forks that have existed)
> the database is the last thing you want to re-invent
good news, FoundationDB is already invented, and just slightly more tested than others on the market.
> In practice, when you reach for a database engine you're actually reaching for four basic features: unique constraints, transactions, indices, and history tables.
That might be a very specific assumption. What about serializability? Replication? Materialized views? Procedures? Locking? Access control?
It’s cool to experiment and try new approaches. Neat one here.
Could still end up moving to Postgres.
What about... queries?
Along with putting estimated reading times (21 minutes in this case) can authors please start putting estimated writing times?
Did it take 10 seconds of prompting? Or hours of thought, trial error and revision? Especially when asking people to read for 20+ minutes...
They should put a Summary section at the top with the raw prompt.
> Protobuf field names are forever
If you only have binary encoded protos to worry about (which is typical) then you can rename fields.
There is a simple version of this idea that really resonates with me.
For about a decade, I've been using flat files on disk or object storage for most of my side projects. There's even a python library that handles some of the plumbing for you [1].
If you don't have strong record-level concurrency needs then it's a lot nicer, easier, cheaper than a relational or document database. And if you do need that, you can design your data model around what defines a record.
Absolutely no mention of Iceberg or Delta Lake, which have had many of the same properties for years. I wonder if they are aware of them, or otherwise why they invented something new.
I generally don't read low-quality AI slop though, so its possible I missed an explanation that didn't contain those key words.
The moment a cache becomes load-bearing you have a database again, except it's in RAM, nobody backed it up, and its failure mode is silence.
Article flagged as AI slop.
> Where it actually breaks
I had a sensible chuckle when I got to this part. This kind of article "____ is all you need" is like another case of Betteridge's Law of Headlines. The answer is "that's not true" every time.
Why? This just seems like a bad idea.
Author here. Ampbase is an OpAMP control plane for agent fleets, and this is how it runs without a database: conditional writes for uniqueness and compare-and-swap, one bucket per customer so isolation isn't a WHERE clause somebody has to remember, ULID keys so history is a prefix list.
Two things the post doesn't cover and I'm happy to get into. What closing the cross-region lost update actually took: annotating the RPCs that depend on a compare-and-swap and replaying those to a single region, with a client-side guard. And the read amplification, which we have a plan for but waiting on a clear signal for when it’s needed.
Cross-posted with thanks to the Tigris folks; the original is at ampbase.io.
> [..] so isolation isn't a WHERE clause somebody has to remember, [..] Two things the post doesn't cover and I'm happy to get into. What closing the cross-region lost update actually took: [..]
Set of my Claude alarm bell, and lo-and-behold, Pangram judges this comment to be 100% AI-generated, albeit with limited confidence.
There's even a load-bearing in there as well
And the read amplification, which we have a plan for but waiting on a clear signal for when it’s needed.
This is Claude between the lines admitting none of it was needed. Postgres on a VM would be doing just fine right now.
real people don't write "compare-and-swap" ... or "lo-and-behold" for that matter.
edit. Clearly the joke about "lo-and-behold" didn't land well with the bots...
People who grew up with assembly language or locking primitives are fairly likely to use compare-and-swap because it's literally the commonly understood term for it.
TIL, thanks for the context. What I was getting at still stands, LLM's seem to love a lot of words-with-hyphens which might be grammatically correct, but I've never seen a human actually type.
Compare-and-swap is a hardware instruction, and beyond that a desirable pattern for dealing with concurrent writes.
CMPXCHG
I do.
For compare-and-swap I have even got a lock less queue that I implemented with it.
> and this is how it runs without a database
From the article:
> In practice, when you reach for a database engine you're actually reaching for four basic features: unique constraints, transactions, indices, and history tables. In order to use Tigris' global object storage as a database, we had to implement all of these primitives ourselves.
Not sure why the "without a database" is or isn't so important, why is it mentioned so often and why the article flip-flopping between "we don't have a DB" and "we're effectively building our own DB"?