sile/mlmd

ml-metadata implementation for Rust

mlmd

A Rust implementation of ml-metadata.

This crate supports the schema version 8 used in ml-metadata-v1.7.0 or later.

Examples

use mlmd::MetadataStore;
use mlmd::metadata::EventType;
use tempfile::NamedTempFile;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    // Creates metadata store.
    let db_file = NamedTempFile::new()?;
    let sqlite_uri = format!("sqlite://{}", db_file.path().to_str().unwrap());
    let mut store = MetadataStore::connect(&sqlite_uri).await?;

    // Creates an artifact.
    let artifact_type_id = store.put_artifact_type("DataSet").execute().await?;
    let artifact_id = store.post_artifact(artifact_type_id).uri("/foo/bar").execute().await?;

    // Creates an execution.
    let execution_type_id = store.put_execution_type("Training").execute().await?;
    let execution_id = store.post_execution(execution_type_id).execute().await?;

    // Links the above execution with the artifact.
    store.put_event(execution_id, artifact_id).ty(EventType::Input).execute().await?;

    // Gets executions.
    let executions = store.get_executions().execute().await?;
    assert_eq!(executions.len(), 1);
    assert_eq!(executions[0].id, execution_id);

    Ok(())
}

Limitations

The following features are not supported yet:

• gRPC client
• input_type and output_type fields of Execution

The following features are not planned to be supported:

• gRPC server
• Database schema migration

ml-metadata References

• GitHub
• Guide
• API Docs