Initial article: April 28th 2023 - see below for an update
I got intrigued by Deno KV, which describes itself as "a global database for global apps". It's a key/value store for Deno applications which bundles some kind of worldwide distributed/replicated database service.
The code example looked like this:
const kv = await Deno.openKv();
Wait, that looks like a core language feature? Are they shipping a client for their own proprietary hosted cloud database as part of their core language?
They're not - at least not in the open source implementation of Deno. I dug in and I think I understand what they're doing.
This is a new feature that's still hidden behind Deno's --unstable
flag. It landed it Deno 1.32 on 22nd March 2023.
The getting started documentation included this confusing text:
A database can be opened using the
Deno.openKv()
function. This function optionally takes a database path on disk as the first argument. If no path is specified, the database is persisted in a global directory, bound to the script thatDeno.openKv()
was called from. Future invocations of the same script will use the same database.
What does "global directory" mean - is this the cloud service they're talking about?
It's not. I ran the following script and used Activity Monitor to see what it was doing:
import { sleep } from "https://deno.land/x/sleep/mod.ts";
const kv = await Deno.openKv();
// Persist an object at the users/alice key.
await kv.set(["users", "alice"], { name: "Alice", age: 44 });
// Read back this key.
const res = await kv.get(["users", "alice"]);
console.log(res.key); // [ "users", "alice" ]
console.log(res.value); // { name: "Alice" }
console.log('About to sleep for a minute')
await sleep(60);
I ran it like this:
% deno run --unstable hello.js
[ "users", "alice" ]
{ name: "Alice", age: 44 }
About to sleep for a minute
This gave me a 60s window to open up Activity Monitor on my Mac, find the deno
process, click the information icon and click on Open Files and Ports:
The crucial file was this one:
/Users/simon/Library/Caches/deno/location_data/82469e8b266758412fd6bbd0058abaee6712cadb9c64024473af6afcff9eba6f/kv.sqlite3-shm
So that's what it means by "a global directory" - it's talking about the ~/Library/Caches/deno/location_data
folder, which appears to hold directories with hashes for names (possibly the hash of the path to the Deno script) which can contain extra data.
And kv.sqlite3
is a SQLite database!
Here's the schema of that database:
sqlite-utils dump kv.sqlite3
BEGIN TRANSACTION;
CREATE TABLE data_version (
k integer primary key,
version integer not null
);
INSERT INTO "data_version" VALUES(0,5);
CREATE TABLE kv (
k blob primary key,
v blob not null,
v_encoding integer not null,
version integer not null
) without rowid;
INSERT INTO "kv" VALUES(X'0275736572730002616C69636500',X'FF0F6F22046E616D652205416C696365220361676549587B02',1,5);
CREATE TABLE migration_state(
k integer not null primary key,
version integer not null
);
INSERT INTO "migration_state" VALUES(0,2);
CREATE TABLE queue (
ts integer not null,
id text not null,
data blob not null,
backoff_schedule text not null,
keys_if_undelivered blob not null,
primary key (ts, id)
);
CREATE TABLE queue_running(
deadline integer not null,
id text not null,
data blob not null,
backoff_schedule text not null,
keys_if_undelivered blob not null,
primary key (deadline, id)
);
COMMIT;
This is interesting. Clearly they've come up with their own atomic key/value primitive, then designed a SQLite schema for storing serialized versions of those objects on disk.
It looks to me like they've designed a data structure that will work well with the new hosted, FoundationDB, global infrastructure they've been building for their Deno Deploy cloud service, then figured out a way to support the same set of operations locally on top of SQLite.
It looks like the magic part is that if you write code that uses Deno.openKv()
without any extra arguments, your script will use a SQLite database in that location_data
directory when it runs locally... but once deployed to Deno Deploy it will switch to using a FoundationDB-backed cloud database, replicated around the world.
I find this particularly interesting in terms of open source business models: they're baking a core feature into their framework which their SaaS platform is uniquely positioned to offer as a global-scale upgrade.
I tried to copy this from the KV marketing landing page, and ended up having to tweak it a bit to get it to work.
Saved as counter.js
:
import { serve } from "https://deno.land/std/http/server.ts";
const kv = await Deno.openKv('count.db');
serve (async () => {
await kv.atomic().sum(["visits"], 1n).commit();
const res = await kv.get(["visits"]);
console.log(res);
return new Response(`Visits: ${res.value.value}`);
});
Run like this:
deno --unstable run counter.js
✅ Granted read access to "count.db".
✅ Granted write access to "count.db".
✅ Granted net access to "0.0.0.0:8000".
Listening on http://localhost:8000/
Now http://localhost:8000/
serves a blank page with a "Visits: 3" number that increments on every hit.
The console logs this out:
{
key: [ "visits" ],
value: KvU64 { value: 19n },
versionstamp: "00000000000000130000"
}
{
key: [ "visits" ],
value: KvU64 { value: 20n },
versionstamp: "00000000000000140000"
}
{
key: [ "visits" ],
value: KvU64 { value: 21n },
versionstamp: "00000000000000150000"
}
And the relevant parts of the SQLite database dump look like this:
CREATE TABLE data_version (
k integer primary key,
version integer not null
);
INSERT INTO "data_version" VALUES(0,22);
CREATE TABLE kv (
k blob primary key,
v blob not null,
v_encoding integer not null,
version integer not null
) without rowid;
INSERT INTO "kv" VALUES(X'0276697369747300',X'1600000000000000',2,22);
CREATE TABLE migration_state(
k integer not null primary key,
version integer not null
);
INSERT INTO "migration_state" VALUES(0,2);
I had to upgrade to Deno 1.32.5 to get this to work (brew upgrade deno
) as the kv.atomic().sum(...)
feature is brand new.
Deno KV is now in open beta, with an intriguing new feature. You can now connect to a remote Deno KV database by setting a DENO_KV_ACCESS_TOKEN
environment variable and then doing this:
const kv = await Deno.openKv(
"https://api.deno.com/databases/<database-id>/connect",
);
On first glance this looked to me like an even deeper intrusion of a proprietary extension into their open source core... but actually it's not. The protocol they are using for this is called KV Connect and is described in enough detail in their documentation that anyone else could build a backend that supports this same feature.
I think this is really neat.
Today Deno announced Deno Queues - a mechanism built on top of Deno KV that lets you send at-least-once delivered messages to a queue and read them back off again.
As with other KV features it works locally using SQLite and uses Foundation DB if you deploy using Deno's cloud service.
One surprising detail: there isn't currently a way to have multiple queues for a single KV instance. If you want to have multiple types of message they all have to go through the same queue - or you need to run against a separate SQLite KV store for each one (locally, I'm not sure what the Deno Deploy version of that looks like).
I got queues working by upgrading to Deno 1.37 using brew upgrade deno
, then putting this in a demo.js
file:
const db = await Deno.openKv('hello.db');
db.listenQueue(async (msg) => {
console.log(msg);
});
await db.enqueue({ channel: "C123456", text: "Slack message" }, {
delay: 10000,
});
This queues a message when the script starts, marking it to run 10s later.
I did this so I could Ctrl+C the script before the ten seconds and see what the schema in SQLite looked like.
I ran the script like this:
deno run --unstable demo.js
I quit before it displayed the message, then used sqlite-utils dump hello.db
to inspect the database. The relevant tables are these ones:
CREATE TABLE queue (
ts integer not null,
id text not null,
data blob not null,
backoff_schedule text not null,
keys_if_undelivered blob not null,
primary key (ts, id)
);
INSERT INTO "queue" VALUES(
1695836430899,
'ab7ac84f-7ceb-4871-b78e-03e1805fd607',
X'FF0F6F22076368616E6E656C220743313233343536220474657874220D536C61636B206D6573736167657B02',
'[100,1000,5000,30000,60000]',
'[]'
);
CREATE TABLE queue_running(
deadline integer not null,
id text not null,
data blob not null,
backoff_schedule text not null,
keys_if_undelivered blob not null,
primary key (deadline, id)
);
CREATE INDEX kv_expiration_ms_idx on kv (expiration_ms);
Also in their codebase: some constants containing SQL queries that are used to interact with that table:
const STATEMENT_QUEUE_ADD_READY: &str =
"insert into queue (ts, id, data, backoff_schedule, keys_if_undelivered) values(?, ?, ?, ?, ?)";
const STATEMENT_QUEUE_GET_NEXT_READY: &str =
"select ts, id, data, backoff_schedule, keys_if_undelivered from queue where ts <= ? order by ts limit 100";
const STATEMENT_QUEUE_GET_EARLIEST_READY: &str =
"select ts from queue order by ts limit 1";
const STATEMENT_QUEUE_REMOVE_READY: &str =
"delete from queue where id = ?";
const STATEMENT_QUEUE_ADD_RUNNING: &str =
"insert into queue_running (deadline, id, data, backoff_schedule, keys_if_undelivered) values(?, ?, ?, ?, ?)";
const STATEMENT_QUEUE_REMOVE_RUNNING: &str =
"delete from queue_running where id = ?";
const STATEMENT_QUEUE_GET_RUNNING_BY_ID: &str =
"select deadline, id, data, backoff_schedule, keys_if_undelivered from queue_running where id = ?";
const STATEMENT_QUEUE_GET_RUNNING: &str =
"select id from queue_running order by deadline limit 100";
The hex data
value decodes to this:
\xff\x0fo"\x07channel"\x07C123456"\x04text"\rSlack message{\x02
That appears to be some kind of custom encoding scheme, but it's clearly the data that we put in the queue.
I found ext/kv/codec.rs which looks like it might be the custom encoding.
Created 2023-04-28T15:02:26-07:00, updated 2023-09-28T06:23:25-07:00 · History · Edit