Connect SlateDB to S3

This tutorial shows you how to connect SlateDB to S3. We’ll use LocalStack to simulate S3.

Create a project

Let’s start by creating a new Rust project:

cargo init slatedb-playground
cd slatedb-playground

Add dependencies

Now add SlateDB and the required dependencies to your Cargo.toml:

cargo add slatedb tokio --features tokio/macros,tokio/rt-multi-thread
cargo add object-store --features object-store/aws
cargo add anyhow

Note

If you see “object_store::path::Path and object_store::path::Path have similar names, but are actually distinct types”, you might need to pin the object_store version to match slatedb’s object_store version. SlateDB also exports slatedb::object_store for convenience, if you’d rather use that.

Setup

You will need to have LocalStack running. You can install it using Homebrew:

brew install localstack/tap/localstack-cli

localstack start -d

For a more detailed setup, see the LocalStack documentation.

You’ll also need the AWS CLI:

brew install awscli

Initialize AWS

SlateDB requires a bucket to work with S3.

Create your S3 bucket:

# Create S3 bucket
aws --endpoint-url=http://localhost:4566 s3api create-bucket --bucket slatedb --region us-east-1

Write some code

Stick this into your src/main.rs file:

main.rs

use object_store::aws::S3ConditionalPut;
use slatedb::Db;
use std::sync::Arc;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    let object_store = Arc::new(
        object_store::aws::AmazonS3Builder::new()
            // These will be different if you are using real AWS
            .with_allow_http(true)
            .with_endpoint("http://localhost:4566")
            .with_access_key_id("test")
            .with_secret_access_key("test")
            .with_bucket_name("slatedb")
            .with_region("us-east-1")
            .with_conditional_put(S3ConditionalPut::ETagMatch)
            .build()?,
    );

    let db = Db::open("/tmp/slatedb_s3_compatible", object_store.clone()).await?;

    // Call db.put with a key and a 64 meg value to trigger L0 SST flush
    let value: Vec<u8> = vec![0; 64 * 1024 * 1024];
    db.put(b"k1", value.as_slice()).await?;
    db.close().await?;

    Ok(())
}

Note

Our code example writes a 64 MiB value to object storage to trigger an L0 SST flush. This is just to show the compacted directory in your bucket.

Run the code

Now you can run the code:

cargo run

This will write a 64 MiB value to SlateDB.

Check the results

Now’ let’s check the database path in the bucket:

% aws --endpoint-url=http://localhost:4566 s3 ls s3://slatedb/tmp/slatedb_s3_compatible/
                           PRE compacted/
                           PRE compactions/
                           PRE manifest/
                           PRE wal/

There are four folders:

• manifest: Contains the manifest files. Manifest files define the state of the DB, including the set of SSTs that are part of the DB.
• wal: Contains the write-ahead log files.
• compacted: Contains the compacted SST files (may not appear in short examples).
• compactions: Contains the compaction-state snapshots.

Let’s check the wal folder:

% aws --endpoint-url=http://localhost:4566 s3 ls s3://slatedb/tmp/slatedb_s3_compatible/wal/
2024-09-04 18:05:57         64 00000000000000000001.sst
2024-09-04 18:05:58   67108996 00000000000000000002.sst

Each of these SST files is a write-ahead log (WAL) file, and each WAL file can contain many RowEntry values. They get flushed based on the flush_interval config. The last WAL file is 64 MiB because it contains the value we wrote.

Finally, let’s check the compacted folder:

% aws --endpoint-url=http://localhost:4566 s3 ls s3://slatedb/tmp/slatedb_s3_compatible/compacted/
2024-09-04 18:05:59   67108996 01J6ZVEZ394GCJT1PHZYY1NZGP.sst

Again, we see the 64 MiB SST file. This is the L0 SST file that was flushed with our value. Over time, the WAL entries will be removed, and the L0 SSTs will be compacted into higher levels.