spiceai.rs

/*
Copyright 2024 The Spice.ai OSS Authors

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

     https://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

use arrow_flight::decode::DecodedPayload;
use async_stream::stream;
use async_trait::async_trait;
use flight_client::FlightClient;
use flight_datafusion::FlightTable;
use futures::StreamExt;
use futures_core::stream::BoxStream;
use ns_lookup::verify_endpoint_connection;
use secrets::Secret;
use snafu::prelude::*;
use spicepod::component::dataset::Dataset;
use std::borrow::Borrow;
use std::pin::Pin;
use std::sync::Arc;
use std::time::Duration;
use std::{collections::HashMap, future::Future};

use crate::datapublisher::{AddDataResult, DataPublisher};
use crate::dataupdate::{DataUpdate, UpdateType};
use crate::tracers::SpacedTracer;
use crate::{info_spaced, status};

use super::DataConnectorFactory;
use super::{flight::Flight, DataConnector};

#[derive(Debug, Snafu)]
pub enum Error {
    #[snafu(display("Unable to parse SpiceAI dataset path: {dataset_path}"))]
    UnableToParseDatasetPath { dataset_path: String },

    #[snafu(display("Unable to publish data to SpiceAI: {source}"))]
    UnableToPublishData { source: flight_client::Error },
}

pub type Result<T, E = Error> = std::result::Result<T, E>;

#[derive(Clone)]
pub struct SpiceAI {
    flight: Flight,
    spaced_trace: Arc<SpacedTracer>,
    api_key: String,
}

impl DataConnectorFactory for SpiceAI {
    fn create(
        secret: Option<Secret>,
        params: Arc<Option<HashMap<String, String>>>,
    ) -> Pin<Box<dyn Future<Output = super::NewDataConnectorResult> + Send>> {
        let default_flight_url = if cfg!(feature = "dev") {
            "https://dev-flight.spiceai.io".to_string()
        } else {
            "https://flight.spiceai.io".to_string()
        };
        Box::pin(async move {
            let secret = secret.ok_or_else(|| super::Error::UnableToCreateDataConnector {
                source: "Missing required secrets".into(),
            })?;

            let url: String = params
                .as_ref() // &Option<HashMap<String, String>>
                .as_ref() // Option<&HashMap<String, String>>
                .and_then(|params| params.get("endpoint").cloned())
                .unwrap_or(default_flight_url);
            tracing::trace!("Connecting to SpiceAI with flight url: {url}");

            verify_endpoint_connection(&url)
                .await
                .map_err(|e| super::Error::UnableToCreateDataConnector { source: e.into() })?;

            let api_key = secret.get("key").unwrap_or_default();
            let flight_client = FlightClient::new(url.as_str(), "", api_key)
                .await
                .map_err(|e| super::Error::UnableToCreateDataConnector { source: e.into() })?;
            let flight = Flight::new(flight_client);
            let spiceai = Self {
                flight,
                api_key: api_key.to_string(),
                spaced_trace: Arc::new(SpacedTracer::new(Duration::from_secs(15))),
            };
            Ok(Box::new(spiceai) as Box<dyn DataConnector>)
        })
    }
}

#[async_trait]
impl DataConnector for SpiceAI {
    fn supports_data_streaming(&self, _dataset: &Dataset) -> bool {
        true
    }

    /// Returns a stream of `DataUpdates` for the given dataset.
    fn stream_data_updates<'a>(&self, dataset: &Dataset) -> BoxStream<'a, DataUpdate> {
        let flight = &self.flight.client;
        let mut flight = flight.clone();
        let spice_dataset_path = Self::spice_dataset_path(dataset);
        let spice_dataset_name = dataset.name.clone();
        Box::pin(stream! {
            let mut initial_connect = true;
            loop {
                // On connection reset, clear the data by sending an empty overwrite update and re-subscribe.
                if !initial_connect {
                    tracing::info!("Reconnecting to {spice_dataset_path}");
                    yield DataUpdate {
                        data: vec![],
                        update_type: UpdateType::Overwrite,
                    };
                }

                let mut stream = match flight.subscribe(&spice_dataset_path).await {
                    Ok(stream) => stream,
                    Err(error) => {
                    status::update_dataset(spice_dataset_name.clone(), status::ComponentStatus::Error);
                    tracing::error!("Unable to subscribe to {spice_dataset_path}: {error}");
                    return;
                    }
                };
                loop {
                    let result = stream.next().await;
                    status::update_dataset(spice_dataset_name.clone(), status::ComponentStatus::Refreshing);
                    match result {
                    Some(Ok(decoded_data)) => if let DecodedPayload::RecordBatch(batch) = decoded_data.payload {
                            status::update_dataset(spice_dataset_name.clone(), status::ComponentStatus::Ready);
                            yield DataUpdate {
                                data: vec![batch],
                                update_type: UpdateType::Append,
                            };
                        } else {
                            status::update_dataset(spice_dataset_name.clone(), status::ComponentStatus::Ready);
                            continue;
                        },
                    Some(Err(error)) => {
                        status::update_dataset(spice_dataset_name.clone(), status::ComponentStatus::Error);
                        tracing::debug!("Error in subscription to {spice_dataset_path}: {error}");
                        tracing::error!("Error in subscription to {spice_dataset_path}");
                        continue;
                    },
                    None => {
                        status::update_dataset(spice_dataset_name.clone(), status::ComponentStatus::Ready);
                        tracing::error!("Flight stream ended for {spice_dataset_path}");
                        break;
                    }
                    };
                }

                initial_connect = false;
            }
        })
    }

    fn get_all_data(
        &self,
        dataset: &Dataset,
    ) -> Pin<Box<dyn Future<Output = Vec<arrow::record_batch::RecordBatch>> + Send>> {
        let spice_dataset_path = Self::spice_dataset_path(dataset);
        self.flight.get_all_data(&spice_dataset_path)
    }

    fn get_data_publisher(&self) -> Option<Box<dyn DataPublisher>> {
        Some(Box::new(self.clone()))
    }

    fn has_table_provider(&self) -> bool {
        true
    }

    async fn get_table_provider(
        &self,
        dataset: &Dataset,
    ) -> std::result::Result<Arc<dyn datafusion::datasource::TableProvider>, super::Error> {
        let dataset_path = Self::spice_dataset_path(dataset);

        let provider = FlightTable::new(self.flight.client.clone(), dataset_path).await;

        match provider {
            Ok(provider) => Ok(Arc::new(provider)),
            Err(error) => Err(super::Error::UnableToGetTableProvider {
                source: error.into(),
            }),
        }
    }
}

impl DataPublisher for SpiceAI {
    /// Adds data ingested locally back to the source.
    fn add_data(&self, dataset: Arc<Dataset>, data_update: DataUpdate) -> AddDataResult {
        let dataset_path = Self::spice_dataset_path(dataset);
        let mut client = self.flight.client.clone();
        let spaced_trace = Arc::clone(&self.spaced_trace);
        let api_key = self.api_key.clone();
        Box::pin(async move {
            tracing::debug!("Adding data to {dataset_path}");

            if let Err(e) = client.publish(&dataset_path, data_update.data).await {
                if let flight_client::Error::PermissionDenied {} = e {
                    tracing::error!("Permission denied when publishing data to Spice AI at '{dataset_path}' - does the API Key ('{api_key}') match the app that owns the dataset?");
                } else {
                    tracing::error!("Error publishing data to Spice AI at '{dataset_path}': {e}");
                    Err(e).context(UnableToPublishDataSnafu)?;
                }
            };

            info_spaced!(spaced_trace, "Data published to {}", &dataset_path);

            Ok(())
        })
    }

    fn name(&self) -> &str {
        "SpiceAI"
    }
}

impl SpiceAI {
    /// Parses a dataset path from a Spice AI dataset definition.
    ///
    /// Spice AI datasets have several possible formats for `dataset.path()`:
    /// 1. `<org>/<app>/datasets/<dataset_name>` or `spice.ai/<org>/<app>/datasets/<dataset_name>`.
    /// 2. `<org>/<app>` or `spice.ai/<org>/<app>`.
    /// 3. `some.blessed.dataset` or `spice.ai/some.blessed.dataset`.
    ///
    /// The second format is a shorthand for the first format, where the dataset name
    /// is the same as the local table name specified in `name`.
    ///
    /// The third format is a path to a "blessed" Spice AI dataset (i.e. a dataset that is
    /// defined and provided by Spice). If the dataset path does not match the first two formats,
    /// then it is assumed to be a path to a blessed dataset.
    ///
    /// This function returns the full dataset path for the given dataset as you would query for it in Spice.
    /// i.e. `<org>.<app>.<dataset_name>`
    #[allow(clippy::match_same_arms)]
    fn spice_dataset_path<T: Borrow<Dataset>>(dataset: T) -> String {
        let dataset = dataset.borrow();
        let path = dataset.path();
        let path = path.trim_start_matches("spice.ai/");
        let path_parts: Vec<&str> = path.split('/').collect();

        match path_parts.as_slice() {
            [org, app] => format!("{org}.{app}.{dataset_name}", dataset_name = dataset.name),
            [org, app, "datasets", dataset_name] => format!("{org}.{app}.{dataset_name}"),
            [org, app, dataset_name] => format!("{org}.{app}.{dataset_name}"),
            _ => path.to_string(),
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_spice_dataset_path() {
        let tests = vec![
            (
                "spiceai:spice.ai/lukekim/demo/datasets/my_data".to_string(),
                "lukekim.demo.my_data",
            ),
            (
                "spiceai:spice.ai/lukekim/demo/my_data".to_string(),
                "lukekim.demo.my_data",
            ),
            (
                "spiceai:lukekim/demo/datasets/my_data".to_string(),
                "lukekim.demo.my_data",
            ),
            (
                "spiceai:lukekim/demo/my_data".to_string(),
                "lukekim.demo.my_data",
            ),
            (
                "spice.ai/lukekim/demo/datasets/my_data".to_string(),
                "lukekim.demo.my_data",
            ),
            (
                "spice.ai/lukekim/demo/my_data".to_string(),
                "lukekim.demo.my_data",
            ),
            (
                "lukekim/demo/datasets/my_data".to_string(),
                "lukekim.demo.my_data",
            ),
            ("lukekim/demo/my_data".to_string(), "lukekim.demo.my_data"),
            ("eth.recent_blocks".to_string(), "eth.recent_blocks"),
        ];

        for (input, expected) in tests {
            let dataset = Dataset::new(input.clone(), "bar".to_string());
            assert_eq!(SpiceAI::spice_dataset_path(dataset), expected, "{input}");
        }
    }
}