S3 Selectย is an Amazon S3 feature that enables retrieving subsets of S3 Objects content via SQL expressions. You can use clauses like SELECT and WHERE to fetch data from CSVs, JSONs, or Apache Parquet files, even if they are compressed with GZIP and/or server-side encrypted.
S3 Select is simple to use, cost-effective and can drastically improve the performances of your application depending on your query. It is overall a great addition to the Serverless developer toolkit, particularly when:
- You need to access a large amount of data (which would make it unfit for other storage solutions like DynamoDB)
- You need to query and filter it in a complex and/or dynamic way
- You donโt need to use a JOIN clause (as S3 Select can only query a single file) or to update a single record of the data (it is called S3 Select, not S3 Insert ๐)
In this article, weโll learn how to run S3 Select commands in Typescript, and how we can improve our DevX and type-safety using Kysely.
Querying with S3 Select
Letโs say that we have a DB of Pokemons in the shape of a CSV, stored somewhere in a S3 Bucket:
id ; name ; customName ; type ; level ; generation
1 ; pikachu ; ; electric ; 42 ; 1
2 ; charizard ; ; fire ; 54 ; 1
3 ; meganium ; plantyDino ; grass ; 26 ; 2
...
What if we want to retrieve the fire Pokemons from generation 1 and 2? Well, S3 Select let us do that with the following query:
import { S3Client, SelectObjectContentCommand } from '@aws-sdk/client-s3';
export const s3Client = new S3Client({
region: 'us-east-1', // <= Your region here
});
const { Payload } = await s3Client.send(
new SelectObjectContentCommand({
// ๐ Those first params are required but don't mind them
ExpressionType: 'SQL',
OutputSerialization: {
JSON: {
RecordDelimiter: ',',
},
},
// ๐ Those depends on the CSV
InputSerialization: {
CSV: {
FileHeaderInfo: 'USE',
FieldDelimiter: ';',
QuoteCharacter: '"',
},
},
// ๐ Those are the most important
Bucket: 'my-super-bucket-name',
Key: 'pokedex.csv',
Expression: `
select "id", "name", "customName", "type" as "pokemonType", "level"
from "S3Object"
where
"generation" in ('1', '2')
and "type" = 'fire'
`,
}),
);
// Note that this command requires the s3:GetObject permission
This is great and all but we can make it better:
- ๐ย We need to add some custom parsing to actually use the response
Payload
- ๐ย We can use Kysely to produce the SQL expression in a type-safe and devX-friendly way
- ๐ย We can also use it to type the command result
Parsing the query response
The Payload
is not a plain Javascript object but an instance of AsyncInterable
. Now what the hell is an AsyncIterable
you ask me? Well, you could look for the official Async Iterator documentation on MDNโฆ or you can just stop asking questions and use the following helper:
import type { SelectObjectContentEventStream } from '@aws-sdk/client-s3';
// ๐ TextDecoder is a native Node/Browser class
const textDecoder = new TextDecoder();
const parseS3SelectEventStream = async (
s3SelectEventStream:
| AsyncIterable<SelectObjectContentEventStream>
| undefined,
): Promise<unknown[]> => {
if (!s3SelectEventStream) {
return [];
}
const stringifiedJSONOutputs: string[] = [];
for await (const event of s3SelectEventStream) {
if (event.Records) {
const stringifiedJSONOutput = textDecoder.decode(event.Records.Payload);
stringifiedJSONOutputs.push(stringifiedJSONOutput);
}
}
const rows = JSON.parse(
'[' + stringifiedJSONOutputs.join('').slice(0, -1) + ']',
) as unknown[];
return rows;
};
Now we can parse the output like this:
const { Payload: s3SelectEventStream } = await s3Client.send(
new SelectObjectContentCommand({
// ...
}),
);
// ๐ Ta-da!
const myPokemons: unknown[] = await parseS3SelectEventStream(
s3SelectEventStream,
);
Building the query with Kysely
Letโs face it, writing SQL queries by hand is a pain and very error prone. Besides, wouldn't it be nice to have some type error if the CSV suddenly changes shape?
Thatโs where Kysely comes to the rescue: Kysely is a type-safe and devX-friendly typescript SQL query builder. It was designed to work with PostgreSQL and MySQL, but it exposes a few classes that can let us write queries without being connected to an actual relational database.
Letโs begin by designing our CSV type:
enum PokemonType {
Water = 'water',
Grass = 'grass',
Fire = 'fire',
// ...
}
interface PokemonCSV {
id: string;
name: string;
customName?: string;
type: PokemonType;
// ๐ In CSVs everything is a string
level: string;
generation: '1' | '2'; // ...up to 9
}
Next, letโs install Kysely and instanciate a Kysely
database:
# npm
npm install kysely
# yarn
yarn add kysely
import {
Kysely,
DummyDriver,
SqliteAdapter,
SqliteIntrospector,
SqliteQueryCompiler,
} from 'kysely';
interface Database {
S3Object: PokemonCSV;
}
const db = new Kysely<Database>({
dialect: {
createAdapter: () => new SqliteAdapter(),
createDriver: () => new DummyDriver(),
createIntrospector: ($db: Kysely<unknown>) => new SqliteIntrospector($db),
createQueryCompiler: () => new SqliteQueryCompiler(),
},
});
// Thatโs it ๐
Simple, no? Now, letโs write the same query, but while enjoying some type-safety and auto-completion:
const kyselyQuery = db
.selectFrom('S3Object')
.select([
'id',
'name',
'customName',
// ๐ You can rename columns as you like
'type as pokemonType',
'level',
// ๐ฅ Will trigger an error:
'unexistingColumn',
])
// ๐ Every method is type-safe!
.where('generation', 'in', ['1', '2'])
.where('type', '=', PokemonType.Fire);
To protect us from nasty SQL injections, Kysely doesnโt directly provide us with the SQL expression but with a sql
string and parameters
array:
const {
sql, // ๐ SQL query with '?' as placeholders
parameters, // ๐ Array of parameters
} = kyselyQuery.compile();
Because S3 Select doesnโt accept parameters in its API, we have to hydrate the parameters ourselves:
const dangerouslyHydrateSQLParameters = (
sql: string,
parameters: readonly unknown[],
): string => {
for (const parameter of parameters) {
sql = sql.replace('?', `'${String(parameter)}'`);
}
return sql;
};
const { sql, parameters } = kyselyQuery.compile();
// โ๏ธ **BE SURE TO VALIDATE DYNAMIC PARAMETERS FIRST** โ๏ธ
const sqlExpression = dangerouslyHydrateSQLParameters(sql, parameters);
console.log(sqlExpression);
// ๐ We retrieve the same expression as above:
// select "id", "name", "customName", "type" as "pokemonType", "level"
// from "S3Object"
// where
// "generation" in ('1', '2')
// and "type" = 'fire'
We just have to provide it to our S3 Select command and voilร ! Weโre done!
Well almost: Notice that we only typed our SQL query expression! What about the response that comes from S3?
Inferring the response type
Initially, Kysely was not only made to build queries but also execute them. We can just benefit from the inferred type by inspecting the execute
property of our Kysely query. That can be done with the help of some TS wizardry:
import type { SelectQueryBuilder } from 'kysely';
type QueryResponseRow<
// ๐ Add a large type constraint
KyselyQuery extends SelectQueryBuilder<
Record<string, unknown>,
string,
unknown
>,
> =
// ๐ Remove the Promise wrapper
Awaited<
// ๐ Get the return type of the execute method
ReturnType<KyselyQuery['execute']>
// ๐ Unpack the array
>[number];
type Pokemon = QueryResponseRow<typeof kyselyQuery>;
// ๐ Equivalent to:
type Pokemon = {
id: string;
name: string;
// ๐ customName is indeed possibly undefined
customName: string | undefined;
level: string;
// ๐ "type" property has been renamed
pokemonType: PokemonType;
};
Now we can just use this type the response of our parseS3SelectEventStream
util and voilร ! Weโre done! For good this time ๐
Conclusion
Both S3 Select and Kysely are awesome tools. By joining both of them, we can run performant, scalable and cost-effective queries while benefitting from strong and DRY type-safety and inference ๐ฅณ
Note that there is one minor drawback, though: Kysely will add 120KB in your Lambdas bundles (props to for serverless-analyze-bundle-plugin for helping me out with this ๐). It is not a lot, but not negligible either as NodeJS Lambdas bundles above 5MB negatively impacts their cold starts. So you might want to re-evaluate adding Kysely to your bundles if your query is not changing often.
Top comments (1)
EDIT: You can actually use the CAST sql clause to cast the pokemon levels as number. Is is both supported by Kysely and S3 Select ๐ See: