It is also possible to encode JSON documents directly as a serialized B-tree. Then you can construct iterators on it directly, and query internal fields at indexed speeds. It is still a serialized document (possible to send over a network), though now you don't need to do any parsing, since the document itself is already indexed. It is called the Lite³ format.
This is super cool! I've always liked Rkyv (https://rkyv.org) but it requires Rust which can be a big lift for a small project. I see this supports binary data (`lite3_val_bytes`) which is great!
Thank you. Having a native bytes type is non-negotiable for any performance intensive application that cannot afford the overhead of base64 encoding. And yes, Rkyv also implements this idea of indexing serialized data. The main differences are:
1) Rkyv uses a binary tree vs Lite³ B-tree (B-trees are more cache and space efficient).
2) Rkyv is immutable once serialized. Lite³ allows for arbitrary mutations on serialized data.
3) Rkyv is Rust only. Lite³ is a 9.3 kB C library free of dependencies.
4) Rkyv as a custom binary format is not directly compatible with other formats. Lite³ can be directly converted to/from JSON.
I have not benchmarked Lite³ against Rust libraries, though it would be an interesting experiment.
That second point is huge – Rkyv does have limited support for in-place mutation, but it is quite limited!
If you added support for running jq natively, that would be very cool. Lite³ brings the B-trees, jq brings the query parser and bytecode, combined, you get SQLite :P
Yes, in fact the name Lite³ was chosen because it is lighter than SQLite.
I thought about implementing something like jq or JSON query, and this is very possible. It is like sending a mini-database that can be queried at speeds thousands of times faster than any JSON library is able to parse.
One interesting effect of being a zero-copy format is that the 'parsing speed' can exceed the memory bandwidth of the CPU, since to fulfill a query you do not actually need to parse the entire dataset. You only walk the branches of the tree that are actually required.
I've talked to some other people that have also shown interest in this idea. There doesn't really seem to exist a good schemaless single-file format that supports advanced queries. There is only SQLite and maybe HDF5.
How does Lite^3 compare to PG's JSONB? PG's JSONB is also a serialized, indexed data structure. One of the key things about JSONB is that for arrays (and so objects) it encodes first their lengths, then the values, but every so many elements (32 is the default IIRC) it encodes an offset, and the reason for this design is that when they encoded offsets only the result did not compress well (and if you think about it it will be obvious why). The price they pay for this design is that finding the offset to the nth element's value requires first finding the offset of the last entry before n that has an offset, then adding all the lengths of the entries in between. This way you get a tunable parameter for trading off speed for compressibility.
EDIT: Ok, I've looked at the format. Some comments:
- Updating in place is cool but you need to clear unused replaced data in case it's sensitive, and then unless you re-encode you will use up more and more space -- once in a while you need a "vacuum". Though vacuuming a Lite^3 document is quite simple: just traverse the data structure and write a new version, and naturally it will be vacuumed.
- On the whole I like Lite^3 quite a bit. Very clever.
- JSONB is also indexed as encoded, but IIUC it's not in-place updateable (unless the new items are the same length as the old) without re-encoding. Though I can imagine a way to tombstone old values and replace them with offsets into appended data, then the result would also need a "vacuum" once in a while.
- I'm curious about compressibility. I suspect not having long runs of pointers (offsets) helps, but still I suspect JSONB is more compressible.
I love the topic of serialization formats, and I've been thinking for some time about ASN.1 compilers (since I maintain one). I've wanted to implement a flatbuffers / JSONB style codec for ASN.1 borrowing ideas from OER. You've given me something to think about! When you have a schema (e.g., an ASN.1 module) you don't really need a B-tree -- the encoded data, if it's encoded in a convenient way, is the B-tree already, but accessing the encoded data by traversal path rather than decoding into nice in-memory structures sure would be a major improvement in codec performance!
The main difference between Lite³ and JSONB is that JSONB is not a standalone portable format, and therefore is not suitable for external interchange. Its purpose is to be an indexable representation of JSON inside a Postgres database. But sending it as standalone messages to arbitrary consumers does not really make sense. JSONB can only be interpreted in a Postgres context. This is different from for example BSON, which can be read and constructed as a standalone format without Mongo.
Another difference is that JSONB is immutable. Suppose you need to replace one specific value inside an object or array. With JSONB, you would rewrite the entire JSONB document as a result of this, even if it is several megabytes large. If you are performing frequent updates inside JSONB documents, this will cause severe write amplification. Despite the fact that offsets are grouped in chunks of 32, Postgres still rewrites the entire document. This is the case for all current Postgres versions.
On the other hand, Lite³ supports replacing of individual values where ONLY the changed value needs updating. For this to work, you need separate offsets. Postgres makes a tradeoff where they get some benefits in size, but as a result become completely read-only. This is the case in general for most types of compression.
Also JSONB is not suited to storing binary data. The user must use a separate bytea column. Lite³ directly implements a native bytes type.
JSONB was designed to sacrifice mutability in favor of read performance, but despite this, I still expect Lite³ to exceed it at read performance. Of course it is hard to back this up without benchmarks, but there are several reasons:
1) JSONB performs runtime string comparison loops to find keys. Lite³ uses fixed-size hash digests comparisons, where the hashes are computed at compile time.
2) JSONB must do 'walking back' because of the 32-grouped offset scheme.
3) Lite³ has none of the database overhead.
Again, the two formats serve a different purpose, but comparing just the raw byte layouts.
I agree that Lite³ is almost certainly better than JSONB on every score except compressibility, but when Lite³ is your database format then that doesn't matter (you can always compress large string/blob values if need be). Compressibility might matter for interchange however, but again, if your messages are huge chances are there are compressible strings in them, or if they're not huge then you probably don't care to compress.
Rkyv is basically the last thing you mentioned already?
It's basically a code gen for deriving serialized structures that can be accessed for read with the exact same API and functionally almost identical (but not quite; in the differences lies much of the special sauce) ABI.
I love SQLite and this is in no way I'm making a point devaluing SQLite, Author's method is excellent approach to get analytical speed out of SQLite. But I am loving DuckDB for similar analytical workloads as it is built for such tasks. DuckDB also reads from single file, like SQLite and DuckDB process large data sets at extreme speeds. I work on my macbook m2 and I have been dealing with about 20 million records and it works fast, very fast.
Loading data into DuckDB is super easy, I was surprised :
SELECT
avg(sale_price),
count(DISTINCT customer_id)
FROM '/my-data-lake/sales/2024/*.json';
and you can also load into a JSON type column and can use postgres type syntax
col->>'$.key'
duckdb is super fast for analytic tasks, especially when u use it with visual eda tool like pygwalker. it allows u handles millions of data visuals and eda in seconds.
but i would say, comparing duckdb and sqlite is a little bit unfair, i would still use sqlite to build system in most of cases, but duckdb only for analytic. you can hardly make a smooth deployment if you apps contains duckdb on a lot of platform
depending on the size and needs of distributed system or application im kind of really excited about postgres + pg_lake. postgres has blown my mind at how well it does concurrent writes at least for the types of things i build/support for my org, the pg_lake extension then adds the ability to.. honestly work like a datalake style analytics engine. it intuitively switches whether or not the transaction goes down the normal query path or it uses duckdb which brings giga-aggregation type queries to massive datasets.
someone should smush sqlite+duckdb together and do that kind of switching depending on query type
It's not an index, it's just (probably parallel) file reads
That being said, it would be trivial to tweak the above script into two steps, one reading data into a DuckDB database table, and the second one reading from that table.
I thought this was common practice, generated columns for JSON performance. I've even used this (although it was in Postgres) to maintain foreign key constraints where the key is buried in a JSON column. What we were doing was slightly cursed but it worked perfectly.
If you're using postgres, couldn't you just create an index on the field inside the JSONB column directly? What advantage are you getting from extracting it to a separate column?
CREATE INDEX idx_status_gin
ON my_table
USING gin ((data->'status'));
Yes, as far as indices go, GIN indices are very expensive especially on modification. They're worthwhile in cases where you want to do arbitrary querying on JSON data, but you definitely don't want to overuse them.
If you can get away with a regular index on either a generated column or an expression, then you absolutely should.
It works until you realize some of these usages would've been better as individual key/value rows.
For example, if you want to store settings as JSON, you first have to parse it through e.g. Zod, hope that it isn't failing due to schema changes (or write migrations and hope that succeeds).
When a simple key/value row just works fine, and you can even do partial fetches / updates
Doesn't sound very cursed, standard normalized relations for things that need it and jsonb for the big bags of attributes you don't care to split apart
This is the typical practice for most index types in SingleStore as well except with the Multi-Value Hash Index which is defined over a JSON or BSON path
Hilariously, I discovered this very technique a couple weeks ago when Claude Code presented it out of the blue as an option with an implemented example when I was trying to find some optimizations for something I'm working on. It turned out to be a really smart and performant choice, one I simply wasn't aware of because I hadn't really kept up with new SQLite features the last few years at all.
Lesson learned: even if you know your tools well, periodically go check out updated docs and see what's new, you might be surprised at what you find!
As others mention, you can create indexes directly against the json without projecting in to a computed column... though the computed column has the added benefit of making certain queries easier.
That said, this is pretty much what you have to do with MS-SQL's limited support for JSON before 2025 (v17). Glad I double checked, since I wasn't even aware they had added the JSON type to 2025.
Exclusively using computed columns, and never directly querying the JSON does have the advantage of making it impossible to accidentally write a unindexed query.
i.e. something like this: CREATE INDEX idx_events_type ON events(json_extract(data, '$.type'))?
i guess caveat here is that slight change in json path syntax (can't think of any right now) can cause SQLite to not use this index, while in case of explicitly specified Virtual Generated Columns you're guaranteed to use the index.
Yeah, you can use index on expression and views to ensure the expression matches, like https://github.com/fsaintjacques/recordlite . The view + index approach decouples the convenience of having a column for a given expression and the need to materialise the column for performance.
> slight change in json path syntax (can't think of any right now) can cause SQLite to not use this index
It's pretty fragile...
-- Just changing the quoting
select * from events where json_extract(data, "$.type") = 'click';
-- Changing the syntax
select * from events where data -> '$.type' = 'click';
Basically anything that alters the text of an expression within the where clause
When you write another query against that index a few weeks later and forget about the caveat, that slight change in where clause will ignore that index.
I'm not sure 2015 counts as new, but that's same release that first introduced the JSON extension. There isn't a version of SQLite with JSON expressions but without indexes on expressions. Also, the JSON extension wasn't enabled by default until 2022, so most people using SQLite with JSON have got a version much newer than 2015.
Opening an article on HN, seeing one of my comments quoted at the top, and then finding out the whole article is about that one comment: that's a first!
I wish devs would normalize their data rather than shove everything into a JSON(B) column, especially when there is a consistent schema across records.
It's much harder to setup proper indexes, enforce constraints, and adds overhead every time you actually want to use the data.
* The data does not map well to database tables, e.g. when it's tree structures (of course that could be represented as many table rows too, but it's complicated and may be slower when you always need to operate on the whole tree anyway)
* your programming language has better types and programming facilities than SQL offers; for example in our Haskell+TypeScript code base, we can conveniently serialise large nested data structures with 100s of types into JSON, without having to think about how to represent those trees as tables.
You do need some fancy in-house way to migrate old JSONs to new JSON in case you want to evolve the (implicit) JSON schema.
I find this one of the hardest part of using JSON, and the main reason why I rather put it in proper columns. Once I go JSON I needs a fair bit of code to deal with migrartions (either doing them during migrations; or some way to do them at read/write time).
Since OP is using Haskell, the actual code most likely won’t really touch the JSON type, but the actual domain type. This makes migrations super easy to write. Of course they could have written a fancy in-house way to do that, or just use the safe-copy library which solves this problem and it has been around for almost two decades. In particular it solves the “nested version control” problem with data structures containing other data structures but with varying versions.
I find that JSON(B) works best when you have a collection of data with different or variant concrete types of data that aren't 1:1 matches. Ex: the actual transaction result if you have different payment processors (paypal, amazon, google, apple-pay, etc)... you don't necessarily want/care about having N different tables for a clean mapping (along with the overhead of a join) to pull the transaction details in the original format(s).
Another example is a classifieds website, where your extra details for a Dress are going to be quite a bit different than the details for a Car or Watch. But, again, you don't necessarily want to inflate the table structure for a fully normalized flow.
If you're using a concretely typed service language it can help. C# does a decent job here. But even then, mixing in Zod with Hono and OpenAPI isn't exactly difficult on the JS/TS front.
Yeah document formats (jsonb) are excellent for apps etc that interface with the messy real world. ecommerce, gvt systems etc, anything involving forms, payments etc
tryna map everything in a relational way etc - you're in a world of pain
For very simple JSON data whose schema never changes, I agree.
But the more complex it is, the more complex the relational representation becomes. JSON responses from some API's could easily require 8 new tables to store the data in, with lots of arbitrary new primary keys and lots of foreign key constraints, your queries will be full of JOIN's that need proper indexing set up...
Oftentimes it's just not worth it, especially if your queries are relatively simple, but you still need to store the full JSON in case you need the data in the future.
Obviously storing JSON in a relational database feels a bit like a Frankenstein monster. But at the end of the day, it's really just about what's simplest to maintain and provides the necessary performance.
And the whole point of the article is how easy it is to set up indexes on JSON.
When a data tree is tightly coupled (like a complex sample of nested data with some arrays from a sensor) and the entire tree is treated like a single thing by writes, the JSON column just keeps things easier. Reads can be accelerated with indexes as demonstrated here.
I fully agree that's wrong (can't imagine the overhead of some larger tables I have if that had happened), that said, often people want weird customizations in medium-sized tables that would set one on a path to having annoying 100 column tables if we couldn't express customizations in a "simple" JSON column (that is more or less polymorphic).
Typical example is a price-setting product I work on.. there's price ranges that are universal (and DB columns reflect that part) but they all have weird custom requests for pricing like rebates on the 3rd weekend after X-mas (but only if the customer is related to Uncle Rudolph who picks his nose).
If you replace JSON with XML in this model it is exactly what the "document store" databases from the 90s and 00s were doing -- parsing at insert and update time, then touching only indexes at query time. It is indeed cool that sqlite does this out of the box.
I was looking for a way to index a JSON column that contains a JSON array, like a list of tags. AFAIK this method won't work for that; you'll either need to use FTS or a separate "tag" table that you index.
Yeah, SQLite doesn't have any true array datatype. I think you could probably do it with a virtual table, but that would be adding a native extension, and it would have to pack its own index.
Tiny bug report: I couldn't edit text in those SQL editor widgets from my iPhone, and I couldn't scroll them to see text that extended past the width of the page either.
I've been coding a lot of small apps recently, and going from local JSON file storage to SQLite has been a very natural path of progression, as data's order of magnitude ramps up. A fully performant database which still feels as simple as opening and reading from a plain JSON file. The trick you describe in the article is actually an unexpected performance buffer that'll come in handy when I start hitting next bottleneck :) Thank you
Thanks for the feedback. The document model in mongo was slopped together by a junior engineer, so perhaps an unorthodox approach. It is basically flat, and already used in a pseudo-relational manner via in-app join to the existing sqlite store. This blog post inspired me to think, what if we just chucked all the json from mongo into sqlite and used the generated indices? Then we can gradually "strangler fig" endpoint by endpoint
Very cool article. To really drill it home, I would have loved to see how the query plan changes. It _looks_ like it should Just Work(tm) but my brain refuses to believe that it's able to use those new indexes so flawlessly
In the 2nd section you're using a CREATE TABLE plus three separate ALTER TABLE calls to add the virtual columns. In the 3rd section you're using a single CREATE TABLE with the virtual columns included from the get go.
I think the intent is to separate the virtual column creation out when it’s introduced in order to highlight that it’s a very lightweight operation. When moving onto the 3rd example, the existence of the virtual columns is just a given.
Depending on the amount of inserts, it might be more efficient to create all the indexes in one go. I think this is certainly true for normal columns.
But I suspect with JSON the overhead of parsing it each time might make it more efficient to update all the indices with every insert.
Then again, it's probably quicker still to insert the raw SQL into a temporary table in memory and then insert all of the new rows into the indexed table as a single query.
In 2 they show how to add virtual columns to an existing table, in 3 how to add indexes to existing virtual columns so they are pre-cooked. Like a cooking show.
Particularly with drizzle, it means I can use sqlite on device with expo-sqlite, and store our data format in a single field, with very little syntax, and the schema and queries all become fully type safe.
Also being able to use the same light orm abstraction server side with bun:sqlite is huge.
Generated columns are pretty great, but what I would really love is a Postgres-style gin index, which dramatically speeds up json queries for unanticipated keys.
I've been using this trick for a while, and it actually got me to do quite a bit without an ORM (just hacking a sane models.py with a few stable wrappers and calling it a day)
json columns pretty much obviated the need for ORMs. It used to be that you'd sometimes have a deep nested thing you really only ever query all at once rather than in pieces, so you'd use an ORM to automate that, but now you can just shove it into json. And then use regular SQL for the relations you actually care about.
Regardless of the number of rows, it doesn't really matter, there are useful cases for where you might be consuming json directly, so instead of parsing it out into a schema for your database, why not just keep it raw and utilize the tools of the database.
It's been around for quite while, but DB people hate to explain where they got an idea. For all I know Vertica got it from somewhere else; I think postgres got jsonb around the same time.
Now there’s a name I haven’t heard in 10 years. (I’m only tenuously connected to the kinds of teams that use/would have used that, so it doesn’t mean much.)
LOL what are the odds, I posted in `Show HN` about Marmot today https://github.com/maxpert/marmot/releases/tag/v2.2.0 and in my head I was thinking exact same thing for supporting MySQL's JSON datatype. At some level I am starting to feel, I might as well be able to expose a full MongoDB compatible protocol that let's you talk to tables as collections, solving this problem once it for all!
Dude what? This is incredible knowledge. I had been fearing this exact problem for so long, but there is an elegant out of the box solution. Thank you!!
Disclaimer: I am working on this.
https://github.com/fastserial/lite3
reply