apiary_runtime/

node.rs

//! The Apiary node — a stateless compute instance in the swarm.
//!
//! [`ApiaryNode`] is the main runtime entry point. It initialises the
//! storage backend, detects system capacity, creates the bee pool,
//! starts the heartbeat writer and world view builder, and manages
//! the node lifecycle.

use std::collections::HashMap;
use std::sync::Arc;
use std::time::Duration;

use arrow::record_batch::RecordBatch;
use tokio::sync::RwLock;
use tracing::info;

use apiary_core::config::NodeConfig;
use apiary_core::error::ApiaryError;
use apiary_core::registry_manager::RegistryManager;
use apiary_core::storage::StorageBackend;
use apiary_core::{CellSizingPolicy, FrameSchema, LedgerAction, Result, WriteResult};
use apiary_query::ApiaryQueryContext;
use apiary_storage::cell_reader::CellReader;
use apiary_storage::cell_writer::CellWriter;
use apiary_storage::ledger::Ledger;
use apiary_storage::local::LocalBackend;
use apiary_storage::s3::S3Backend;

use crate::bee::{BeePool, BeeStatus};
use crate::behavioral::{AbandonmentTracker, ColonyThermometer};
use crate::cache::CellCache;
use crate::heartbeat::{HeartbeatWriter, NodeState, WorldView, WorldViewBuilder};

/// An Apiary compute node — the runtime for one machine in the swarm.
///
/// The node holds a reference to the storage backend and its configuration.
/// In solo mode it uses [`LocalBackend`]; in multi-node mode it uses
/// [`S3Backend`]. The node is otherwise stateless — all committed state
/// lives in object storage.
pub struct ApiaryNode {
    /// Node configuration including auto-detected capacity.
    pub config: NodeConfig,

    /// The shared storage backend (object storage or local filesystem).
    pub storage: Arc<dyn StorageBackend>,

    /// Registry manager for DDL operations.
    pub registry: Arc<RegistryManager>,

    /// DataFusion-based SQL query context.
    pub query_ctx: Arc<tokio::sync::Mutex<ApiaryQueryContext>>,

    /// Pool of bees for isolated task execution.
    pub bee_pool: Arc<BeePool>,

    /// Local cell cache with LRU eviction.
    pub cell_cache: Arc<CellCache>,

    /// Colony thermometer for measuring system health.
    pub thermometer: ColonyThermometer,

    /// Abandonment tracker for task failure handling.
    pub abandonment_tracker: Arc<AbandonmentTracker>,

    /// Heartbeat writer for this node.
    heartbeat_writer: Arc<HeartbeatWriter>,

    /// Shared world view (updated by the background builder).
    world_view: Arc<RwLock<WorldView>>,

    /// World view builder (kept alive for on-demand cleanup).
    #[allow(dead_code)]
    world_view_builder: Arc<WorldViewBuilder>,

    /// Cancellation channel to stop background tasks on shutdown.
    cancel_tx: tokio::sync::watch::Sender<bool>,
}

impl ApiaryNode {
    /// Start a new Apiary node with the given configuration.
    ///
    /// Initialises the appropriate storage backend based on `config.storage_uri`
    /// and logs the node's capacity.
    pub async fn start(config: NodeConfig) -> Result<Self> {
        let storage: Arc<dyn StorageBackend> = if config.storage_uri.starts_with("s3://") {
            Arc::new(S3Backend::new(&config.storage_uri)?)
        } else {
            // Parse local URI: "local://<path>" or treat as raw path
            let path = config
                .storage_uri
                .strip_prefix("local://")
                .unwrap_or(&config.storage_uri);

            // Expand ~ to home directory
            let expanded = if path.starts_with("~/") || path.starts_with("~\\") {
                let home = home_dir().ok_or_else(|| ApiaryError::Config {
                    message: "Cannot determine home directory".to_string(),
                })?;
                home.join(&path[2..])
            } else {
                std::path::PathBuf::from(path)
            };

            Arc::new(LocalBackend::new(expanded).await?)
        };

        info!(
            node_id = %config.node_id,
            cores = config.cores,
            memory_mb = config.memory_bytes / (1024 * 1024),
            memory_per_bee_mb = config.memory_per_bee / (1024 * 1024),
            target_cell_size_mb = config.target_cell_size / (1024 * 1024),
            storage_uri = %config.storage_uri,
            "Apiary node started"
        );

        // Initialize registry
        let registry = Arc::new(RegistryManager::new(Arc::clone(&storage)));
        let _ = registry.load_or_create().await?;
        info!("Registry loaded");

        // Initialize query context
        let query_ctx = Arc::new(tokio::sync::Mutex::new(ApiaryQueryContext::with_node_id(
            Arc::clone(&storage),
            Arc::clone(&registry),
            config.node_id.clone(),
        )));

        // Initialize bee pool
        let bee_pool = Arc::new(BeePool::new(&config));
        info!(bees = config.cores, "Bee pool initialized");

        // Initialize cell cache
        let cache_dir = config.cache_dir.join("cells");
        let cell_cache =
            Arc::new(CellCache::new(cache_dir, config.max_cache_size, Arc::clone(&storage)).await?);
        info!(
            max_cache_mb = config.max_cache_size / (1024 * 1024),
            "Cell cache initialized"
        );

        // Initialize heartbeat writer
        let heartbeat_writer = Arc::new(HeartbeatWriter::new(
            Arc::clone(&storage),
            &config,
            Arc::clone(&bee_pool),
            Arc::clone(&cell_cache),
        ));

        // Initialize world view builder
        let world_view_builder = Arc::new(WorldViewBuilder::new(
            Arc::clone(&storage),
            config.heartbeat_interval, // poll at same rate as heartbeat
            config.dead_threshold,
        ));
        let world_view = world_view_builder.world_view();

        // Write initial heartbeat and build initial world view synchronously
        // so that swarm_status() works immediately after start().
        heartbeat_writer.write_once().await?;
        world_view_builder.poll_once().await?;
        info!("Initial heartbeat written and world view built");

        // Create cancellation channel
        let (cancel_tx, cancel_rx) = tokio::sync::watch::channel(false);

        // Start heartbeat writer background task
        {
            let writer = Arc::clone(&heartbeat_writer);
            let rx = cancel_rx.clone();
            tokio::spawn(async move {
                writer.run(rx).await;
            });
        }

        // Start world view builder background task
        {
            let builder = Arc::clone(&world_view_builder);
            let rx = cancel_rx.clone();
            tokio::spawn(async move {
                builder.run(rx).await;
            });
        }

        // Start query worker task poller (for distributed execution)
        {
            let storage = Arc::clone(&storage);
            let query_ctx = Arc::clone(&query_ctx);
            let node_id = config.node_id.clone();
            let rx = cancel_rx.clone();
            tokio::spawn(async move {
                run_query_worker_poller(storage, query_ctx, node_id, rx).await;
            });
        }

        info!("Heartbeat, world view, and query worker background tasks started");

        Ok(Self {
            config,
            storage,
            registry,
            query_ctx,
            bee_pool,
            cell_cache,
            thermometer: ColonyThermometer::default(),
            abandonment_tracker: Arc::new(AbandonmentTracker::default()),
            heartbeat_writer,
            world_view,
            world_view_builder,
            cancel_tx,
        })
    }

    /// Gracefully shut down the node.
    ///
    /// Stops background tasks (heartbeat writer, world view builder),
    /// deletes the heartbeat file, and cleans up resources.
    pub async fn shutdown(&self) {
        info!(node_id = %self.config.node_id, "Apiary node shutting down");

        // Signal background tasks to stop
        let _ = self.cancel_tx.send(true);

        // Allow background tasks a moment to stop
        tokio::time::sleep(Duration::from_millis(100)).await;

        // Delete our heartbeat file (graceful departure)
        if let Err(e) = self.heartbeat_writer.delete_heartbeat().await {
            tracing::warn!(error = %e, "Failed to delete heartbeat during shutdown");
        } else {
            info!(node_id = %self.config.node_id, "Heartbeat deleted (graceful departure)");
        }
    }

    /// Write data to a frame. This is the end-to-end write path:
    /// 1. Resolve frame from registry
    /// 2. Open/create ledger
    /// 3. Validate schema
    /// 4. Partition data
    /// 5. Write cells to storage
    /// 6. Commit ledger entry
    pub async fn write_to_frame(
        &self,
        hive: &str,
        box_name: &str,
        frame_name: &str,
        batch: &RecordBatch,
    ) -> Result<WriteResult> {
        let start = std::time::Instant::now();

        // Resolve frame metadata
        let frame = self.registry.get_frame(hive, box_name, frame_name).await?;
        let schema = FrameSchema::from_json_value(&frame.schema)?;
        let frame_path = format!("{}/{}/{}", hive, box_name, frame_name);

        // Open or create ledger
        let mut ledger = match Ledger::open(Arc::clone(&self.storage), &frame_path).await {
            Ok(l) => l,
            Err(_) => {
                Ledger::create(
                    Arc::clone(&self.storage),
                    &frame_path,
                    schema.clone(),
                    frame.partition_by.clone(),
                    &self.config.node_id,
                )
                .await?
            }
        };

        // Write cells
        let sizing = CellSizingPolicy::new(
            self.config.target_cell_size,
            self.config.max_cell_size,
            self.config.min_cell_size,
        );

        let writer = CellWriter::new(
            Arc::clone(&self.storage),
            frame_path,
            schema,
            frame.partition_by.clone(),
            sizing,
        );

        let cells = writer.write(batch).await?;

        let cells_written = cells.len();
        let rows_written: u64 = cells.iter().map(|c| c.rows).sum();
        let bytes_written: u64 = cells.iter().map(|c| c.bytes).sum();

        // Commit to ledger
        let version = ledger
            .commit(LedgerAction::AddCells { cells }, &self.config.node_id)
            .await?;

        let duration_ms = start.elapsed().as_millis() as u64;
        let temperature = self.thermometer.measure(&self.bee_pool).await;

        Ok(WriteResult {
            version,
            cells_written,
            rows_written,
            bytes_written,
            duration_ms,
            temperature,
        })
    }

    /// Read data from a frame, optionally filtering by partition values.
    /// Returns all matching data as a merged RecordBatch.
    pub async fn read_from_frame(
        &self,
        hive: &str,
        box_name: &str,
        frame_name: &str,
        partition_filter: Option<&HashMap<String, String>>,
    ) -> Result<Option<RecordBatch>> {
        let frame_path = format!("{}/{}/{}", hive, box_name, frame_name);

        let ledger = match Ledger::open(Arc::clone(&self.storage), &frame_path).await {
            Ok(l) => l,
            Err(ApiaryError::NotFound { .. }) => return Ok(None),
            Err(e) => return Err(e),
        };

        let cells = if let Some(filter) = partition_filter {
            ledger.prune_cells(filter, &HashMap::new())
        } else {
            ledger.active_cells().iter().collect()
        };

        if cells.is_empty() {
            return Ok(None);
        }

        let reader = CellReader::new(Arc::clone(&self.storage), frame_path);
        reader.read_cells_merged(&cells, None).await
    }

    /// Overwrite all data in a frame with new data.
    /// Commits a RewriteCells entry removing all existing cells and adding new ones.
    pub async fn overwrite_frame(
        &self,
        hive: &str,
        box_name: &str,
        frame_name: &str,
        batch: &RecordBatch,
    ) -> Result<WriteResult> {
        let start = std::time::Instant::now();

        let frame = self.registry.get_frame(hive, box_name, frame_name).await?;
        let schema = FrameSchema::from_json_value(&frame.schema)?;
        let frame_path = format!("{}/{}/{}", hive, box_name, frame_name);

        let mut ledger = Ledger::open(Arc::clone(&self.storage), &frame_path).await?;

        let sizing = CellSizingPolicy::new(
            self.config.target_cell_size,
            self.config.max_cell_size,
            self.config.min_cell_size,
        );

        let writer = CellWriter::new(
            Arc::clone(&self.storage),
            frame_path,
            schema,
            frame.partition_by.clone(),
            sizing,
        );

        let new_cells = writer.write(batch).await?;

        let cells_written = new_cells.len();
        let rows_written: u64 = new_cells.iter().map(|c| c.rows).sum();
        let bytes_written: u64 = new_cells.iter().map(|c| c.bytes).sum();

        // Remove all old cells, add new ones
        let removed: Vec<_> = ledger.active_cells().iter().map(|c| c.id.clone()).collect();

        let version = ledger
            .commit(
                LedgerAction::RewriteCells {
                    removed,
                    added: new_cells,
                },
                &self.config.node_id,
            )
            .await?;

        let duration_ms = start.elapsed().as_millis() as u64;
        let temperature = self.thermometer.measure(&self.bee_pool).await;

        Ok(WriteResult {
            version,
            cells_written,
            rows_written,
            bytes_written,
            duration_ms,
            temperature,
        })
    }

    /// Initialize the ledger for a frame (called after create_frame in registry).
    pub async fn init_frame_ledger(
        &self,
        hive: &str,
        box_name: &str,
        frame_name: &str,
    ) -> Result<()> {
        let frame = self.registry.get_frame(hive, box_name, frame_name).await?;
        let schema = FrameSchema::from_json_value(&frame.schema)?;
        let frame_path = format!("{}/{}/{}", hive, box_name, frame_name);

        Ledger::create(
            Arc::clone(&self.storage),
            &frame_path,
            schema,
            frame.partition_by.clone(),
            &self.config.node_id,
        )
        .await?;

        Ok(())
    }

    /// Execute a SQL query and return results as RecordBatches.
    ///
    /// The query is executed through the BeePool — assigned to an idle bee
    /// or queued if all bees are busy. Each bee runs in its own sealed
    /// chamber with memory budget and timeout enforcement.
    ///
    /// Supports:
    /// - Standard SQL (SELECT, GROUP BY, ORDER BY, etc.) over frames
    /// - Custom commands: USE HIVE, USE BOX, SHOW HIVES, SHOW BOXES, SHOW FRAMES, DESCRIBE
    /// - 3-part table names: hive.box.frame
    /// - 1-part names after USE HIVE / USE BOX
    pub async fn sql(&self, query: &str) -> Result<Vec<RecordBatch>> {
        let query_ctx = Arc::clone(&self.query_ctx);
        let query_owned = query.to_string();
        let rt_handle = tokio::runtime::Handle::current();

        let handle = self
            .bee_pool
            .submit(move || {
                rt_handle.block_on(async {
                    let mut ctx = query_ctx.lock().await;
                    ctx.sql(&query_owned).await
                })
            })
            .await;

        handle.await.map_err(|e| ApiaryError::Internal {
            message: format!("Task join error: {e}"),
        })?
    }

    /// Return the status of each bee in the pool.
    pub async fn bee_status(&self) -> Vec<BeeStatus> {
        self.bee_pool.status().await
    }

    /// Return the current world view snapshot.
    pub async fn world_view(&self) -> WorldView {
        self.world_view.read().await.clone()
    }

    /// Return swarm status: a summary of all nodes visible to this node.
    pub async fn swarm_status(&self) -> SwarmStatus {
        let view = self.world_view.read().await;
        let mut nodes = Vec::new();

        for status in view.nodes.values() {
            nodes.push(SwarmNodeInfo {
                node_id: status.node_id.as_str().to_string(),
                state: match status.state {
                    NodeState::Alive => "alive".to_string(),
                    NodeState::Suspect => "suspect".to_string(),
                    NodeState::Dead => "dead".to_string(),
                },
                bees: status.heartbeat.load.bees_total,
                idle_bees: status.heartbeat.load.bees_idle,
                memory_pressure: status.heartbeat.load.memory_pressure,
                colony_temperature: status.heartbeat.load.colony_temperature,
            });
        }

        // Sort by node_id for deterministic output
        nodes.sort_by(|a, b| a.node_id.cmp(&b.node_id));

        let total_bees: usize = nodes.iter().map(|n| n.bees).sum();
        let total_idle_bees: usize = nodes.iter().map(|n| n.idle_bees).sum();

        SwarmStatus {
            nodes,
            total_bees,
            total_idle_bees,
        }
    }

    /// Return the current colony status: temperature and regulation state.
    pub async fn colony_status(&self) -> ColonyStatus {
        let temperature = self.thermometer.measure(&self.bee_pool).await;
        let regulation = self.thermometer.regulation(temperature);

        ColonyStatus {
            temperature,
            regulation: regulation.as_str().to_string(),
            setpoint: self.thermometer.setpoint(),
        }
    }

    /// Execute a distributed query (v2 feature - explicit control).
    ///
    /// This method is reserved for v2 when users want explicit control over
    /// distributed execution strategy. In v1, distributed execution happens
    /// transparently within the query context based on query planning.
    #[allow(dead_code)] // Reserved for v2 explicit distribution control
    pub async fn sql_distributed(&self, query: &str) -> Result<Vec<RecordBatch>> {
        // v1: Distributed execution is transparent in query context
        // v2: This will allow explicit control over distribution strategy
        self.sql(query).await
    }
}

/// Summary of the swarm as seen by this node.
#[derive(Debug, Clone)]
pub struct SwarmStatus {
    /// Info for each known node.
    pub nodes: Vec<SwarmNodeInfo>,
    /// Total bees across all nodes.
    pub total_bees: usize,
    /// Total idle bees across all nodes.
    pub total_idle_bees: usize,
}

/// Info about a single node in the swarm.
#[derive(Debug, Clone)]
pub struct SwarmNodeInfo {
    pub node_id: String,
    pub state: String,
    pub bees: usize,
    pub idle_bees: usize,
    pub memory_pressure: f64,
    pub colony_temperature: f64,
}

/// Colony temperature and regulation status.
#[derive(Debug, Clone)]
pub struct ColonyStatus {
    /// Current colony temperature (0.0 to 1.0).
    pub temperature: f64,
    /// Regulation state: "cold", "ideal", "warm", "hot", or "critical".
    pub regulation: String,
    /// Temperature setpoint.
    pub setpoint: f64,
}

/// Best-effort home directory detection.
fn home_dir() -> Option<std::path::PathBuf> {
    #[cfg(target_os = "windows")]
    {
        std::env::var("USERPROFILE")
            .ok()
            .map(std::path::PathBuf::from)
    }
    #[cfg(not(target_os = "windows"))]
    {
        std::env::var("HOME").ok().map(std::path::PathBuf::from)
    }
}

/// Background task that polls for distributed query manifests and executes assigned tasks.
async fn run_query_worker_poller(
    storage: Arc<dyn StorageBackend>,
    query_ctx: Arc<tokio::sync::Mutex<ApiaryQueryContext>>,
    node_id: apiary_core::types::NodeId,
    cancel: tokio::sync::watch::Receiver<bool>,
) {
    use apiary_query::distributed;

    info!(node_id = %node_id, "Query worker poller started");

    let poll_interval = Duration::from_millis(500);

    loop {
        tokio::select! {
            _ = tokio::time::sleep(poll_interval) => {
                // List query manifests
                match storage.list("_queries/").await {
                    Ok(keys) => {
                        // Find manifest files
                        for key in keys {
                            if !key.ends_with("/manifest.json") {
                                continue;
                            }

                            // Extract query_id from path: _queries/{query_id}/manifest.json
                            let parts: Vec<&str> = key.split('/').collect();
                            if parts.len() < 3 {
                                continue;
                            }
                            let query_id = parts[1];

                            // Try to read the manifest
                            match distributed::read_manifest(&storage, query_id).await {
                                Ok(manifest) => {
                                    // Check if any tasks are assigned to this node
                                    let my_tasks: Vec<_> = manifest.tasks.iter()
                                        .filter(|t| t.node_id == node_id)
                                        .collect();

                                    if my_tasks.is_empty() {
                                        continue;
                                    }

                                    // Check if we've already written our partial result
                                    let partial_path = distributed::partial_result_path(query_id, &node_id);
                                    if storage.get(&partial_path).await.is_ok() {
                                        // Already completed
                                        continue;
                                    }

                                    // Execute tasks and write partial result
                                    info!(
                                        query_id = %query_id,
                                        tasks = my_tasks.len(),
                                        "Executing distributed query tasks"
                                    );

                                    let mut results = Vec::new();
                                    let ctx = query_ctx.lock().await;

                                    for task in my_tasks {
                                        match ctx.execute_task(&task.sql_fragment, &task.cells).await {
                                            Ok(batches) => {
                                                results.extend(batches);
                                            }
                                            Err(e) => {
                                                tracing::warn!(
                                                    task_id = %task.task_id,
                                                    error = %e,
                                                    "Task execution failed"
                                                );
                                            }
                                        }
                                    }

                                    // Write partial result
                                    if !results.is_empty() {
                                        if let Err(e) = distributed::write_partial_result(
                                            &storage,
                                            query_id,
                                            &node_id,
                                            &results,
                                        ).await {
                                            tracing::warn!(
                                                query_id = %query_id,
                                                error = %e,
                                                "Failed to write partial result"
                                            );
                                        }
                                    }
                                }
                                Err(_) => {
                                    // Manifest not readable yet or deleted
                                    continue;
                                }
                            }
                        }
                    }
                    Err(e) => {
                        tracing::warn!(error = %e, "Failed to list query manifests");
                    }
                }
            }
            _ = wait_for_cancel(&cancel) => {
                tracing::debug!(node_id = %node_id, "Query worker poller stopping");
                break;
            }
        }
    }
}

/// Helper to wait for cancellation.
async fn wait_for_cancel(cancel: &tokio::sync::watch::Receiver<bool>) {
    let mut rx = cancel.clone();
    let _ = rx.wait_for(|&v| v).await;
}

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

    #[tokio::test]
    async fn test_start_local_node() {
        let tmp = tempfile::TempDir::new().unwrap();
        let mut config = NodeConfig::detect("local://test");
        config.storage_uri = format!("local://{}", tmp.path().display());
        let node = ApiaryNode::start(config).await.unwrap();
        assert!(node.config.cores > 0);
        node.shutdown().await;
    }

    #[tokio::test]
    async fn test_start_with_raw_path() {
        let tmp = tempfile::TempDir::new().unwrap();
        let mut config = NodeConfig::detect("test");
        config.storage_uri = tmp.path().to_string_lossy().to_string();
        let node = ApiaryNode::start(config).await.unwrap();
        node.shutdown().await;
    }
}