relay-knowledge 1.1.16

use std::{
    path::PathBuf,
    sync::{Arc, atomic::AtomicUsize},
    time::Instant,
};

use crate::{
    api::{
        AgentProtocolStatus, ApiError, ApiMetadata, CodeIndexWorkerRunRequest,
        CodeIndexWorkerRunResponse, EmbeddingProviderProbeResponse, GRAPH_CANVAS_MAX_LIMIT,
        GraphCanvasEdge, GraphCanvasKind, GraphCanvasNode, GraphCanvasRequest, GraphCanvasResponse,
        GraphCanvasSummary, GraphInspectionRequest, GraphInspectionResponse, HealthResponse,
        IndexRefreshRequest, IndexRefreshResponse, IngestRequest, IngestResponse,
        MultimodalExtractionRequest, MultimodalExtractionResponse, ProjectStatusResponse,
        RequestContext, ServiceRecoveryReport,
    },
    clock::system_now_millis_or_zero as current_time_millis,
    domain::{AuditStatus, CodeParseStatusCounts, CodeRepositoryTotals, IndexKind},
    env::EnvironmentConfig,
    model_provider::ModelProviderConfigService,
    observability::ObservabilityRuntime,
    ports::{
        embedding::{EmbeddingProvider, EmbeddingRequest, ProviderRetryClass},
        worker_outbound::WorkerOutboundPort,
    },
    project::{
        LINUX_SERVICE_DEFINITION_FILE_NAME, MACOS_SERVICE_DEFINITION_FILE_NAME, PROJECT_NAME,
        WINDOWS_SERVICE_DEFINITION_FILE_NAME,
    },
    storage::{
        FileIndexDiagnostics, GraphCanvasSelection, GraphCanvasStorageRequest, GraphInspection,
        KnowledgeStore, KnowledgeStoreFactory, NewAuditEvent, StorageError,
    },
};

use storage_provider::StorageProvider;

use super::{
    RuntimeConfiguration, RuntimeConfigurationError,
    knowledge::{
        index_refresh::{
            index_refresh_outcome, metadata_for_indexes, recover_index_kinds, refresh_index_kinds,
        },
        ingest::mutation_batch_from_request,
        multimodal::extraction_ingest_request,
    },
    runtime::{agent_protocol_status, runtime_status, runtime_status_with_model_profiles},
    update::{VersionCheckResponse, check_for_updates},
};

/// Shared application service used by CLI, Web, and future API adapters.
#[derive(Clone)]
pub struct RelayKnowledgeService {
    pub(super) runtime: RuntimeConfiguration,
    pub(super) storage: StorageProvider,
    pub(super) health_cache: Arc<tokio::sync::RwLock<Option<HealthResponse>>>,
    pub(super) watcher: Arc<tokio::sync::RwLock<Option<crate::watcher::WatcherHandle>>>,
    pub(super) code_retention_cursor: Arc<AtomicUsize>,
    pub(super) embedding_provider: Option<Arc<dyn EmbeddingProvider>>,
    pub(super) worker_outbound: Option<Arc<dyn WorkerOutboundPort>>,
}

impl RelayKnowledgeService {
    /// Creates a service from validated configuration and injected runtime adapters.
    pub fn with_runtime_adapters(
        runtime: RuntimeConfiguration,
        factory: Arc<dyn KnowledgeStoreFactory>,
        embedding_provider: Option<Arc<dyn EmbeddingProvider>>,
        worker_outbound: Option<Arc<dyn WorkerOutboundPort>>,
    ) -> Self {
        Self {
            storage: StorageProvider::configured(factory),
            runtime,
            health_cache: Arc::new(tokio::sync::RwLock::new(None)),
            watcher: Arc::new(tokio::sync::RwLock::new(None)),
            code_retention_cursor: Arc::new(AtomicUsize::new(0)),
            embedding_provider,
            worker_outbound,
        }
    }

    /// Creates a service backed by an explicit store and injected runtime adapters.
    pub fn with_store_and_runtime_adapters(
        runtime: RuntimeConfiguration,
        store: Arc<dyn KnowledgeStore>,
        embedding_provider: Option<Arc<dyn EmbeddingProvider>>,
        worker_outbound: Option<Arc<dyn WorkerOutboundPort>>,
    ) -> Self {
        Self {
            runtime,
            storage: StorageProvider::ready(store),
            health_cache: Arc::new(tokio::sync::RwLock::new(None)),
            watcher: Arc::new(tokio::sync::RwLock::new(None)),
            code_retention_cursor: Arc::new(AtomicUsize::new(0)),
            embedding_provider,
            worker_outbound,
        }
    }

    /// Applies network-related settings from a typed environment snapshot.
    pub async fn refresh_network_from_environment(
        &self,
        environment: &EnvironmentConfig,
    ) -> Result<(), RuntimeConfigurationError> {
        self.runtime
            .network
            .refresh_from_environment(environment)
            .map(|_| ())
            .map_err(RuntimeConfigurationError::Network)
    }

    /// Returns the shared observability runtime for interface adapters.
    pub fn observability(&self) -> ObservabilityRuntime {
        self.runtime.observability.clone()
    }

    /// Returns the model provider configuration service rooted in runtime paths.
    pub fn model_provider_config(&self) -> ModelProviderConfigService {
        ModelProviderConfigService::new(self.runtime.paths.clone())
    }

    /// Checks configured release sources without opening graph storage.
    pub async fn check_for_updates(&self, force_refresh: bool) -> VersionCheckResponse {
        check_for_updates(
            &self.runtime.paths,
            &self.runtime.network,
            &self.runtime.updates,
            force_refresh,
        )
        .await
    }

    /// Persists a redacted agent protocol audit event through the durable sink.
    pub async fn record_agent_audit(&self, event: AgentDurableAuditInput) -> Result<(), ApiError> {
        let store = self.storage.get().await.map_err(storage_api_error)?;
        store
            .insert_audit_event(NewAuditEvent {
                operation: event.operation,
                interface: event.interface,
                request_id: event.request_id,
                trace_id: event.trace_id,
                status: event.status,
                actor: event.actor,
                source_scope: event.source_scope,
                graph_version: event.graph_version,
                detail_json: event.detail_json,
                message: event.message,
                now_ms: current_time_millis(),
            })
            .await
            .map(|_| ())
            .map_err(storage_api_error)
    }

    /// Returns the current project status through the unified API contract.
    pub async fn project_status(
        &self,
        context: RequestContext,
    ) -> Result<ProjectStatusResponse, ApiError> {
        let store = self.storage.get().await.map_err(storage_api_error)?;
        let graph_version = store
            .current_graph_version()
            .await
            .map_err(storage_api_error)?;

        let model_profiles = self
            .model_provider_config()
            .profile_summary(&self.runtime.retrieval)
            .await;

        Ok(ProjectStatusResponse {
            project_name: PROJECT_NAME.to_owned(),
            metadata: ApiMetadata::graph_only(&context, graph_version),
            runtime: runtime_status_with_model_profiles(&self.runtime, model_profiles),
        })
    }

    /// Returns runtime diagnostics without opening or migrating graph storage.
    pub fn runtime_diagnostics(
        &self,
        context: RequestContext,
    ) -> (ProjectStatusResponse, AgentProtocolStatus) {
        (
            ProjectStatusResponse {
                project_name: PROJECT_NAME.to_owned(),
                metadata: ApiMetadata::graph_only(&context, crate::domain::GraphVersion::ZERO),
                runtime: runtime_status(&self.runtime),
            },
            agent_protocol_status(&self.runtime),
        )
    }

    /// Commits evidence into graph storage and refreshes all v1 index metadata.
    pub async fn ingest(
        &self,
        request: IngestRequest,
        context: RequestContext,
    ) -> Result<IngestResponse, ApiError> {
        let batch = mutation_batch_from_request(request)
            .map_err(|error| ApiError::invalid_argument(error.to_string()))?;
        let worker_evidence = batch.evidence.clone();
        let store = self.storage.get().await.map_err(storage_api_error)?;
        let receipt = store
            .commit_mutation_batch(batch)
            .await
            .map_err(storage_api_error)?;
        self.queue_worker_tasks_for_evidence(&store, &worker_evidence, receipt.graph_version)
            .await?;
        let (indexes, metadata, index_refresh_error) = match refresh_index_kinds(
            &store,
            IndexKind::ALL,
            receipt.graph_version,
            &self.runtime.retrieval,
        )
        .await
        {
            Ok(outcome) => {
                let metadata =
                    metadata_for_indexes(&context, receipt.graph_version, &outcome.indexes);

                (outcome.indexes, metadata, None)
            }
            Err(error) => (
                Vec::new(),
                ApiMetadata::indexed(&context, receipt.graph_version, None, None, true),
                Some(error.message),
            ),
        };

        Ok(IngestResponse {
            metadata,
            receipt,
            indexes,
            index_refresh_error,
        })
    }

    /// Commits derived multimodal worker output through the same bounded ingest path.
    pub async fn commit_multimodal_extraction(
        &self,
        request: MultimodalExtractionRequest,
        context: RequestContext,
    ) -> Result<MultimodalExtractionResponse, ApiError> {
        let converted = extraction_ingest_request(request).map_err(ApiError::invalid_argument)?;
        let parent_evidence_id = converted.parent_evidence_id;
        let derived_evidence_count = converted.derived_evidence_count;
        let response = self.ingest(converted.ingest, context).await?;

        Ok(MultimodalExtractionResponse {
            metadata: response.metadata,
            parent_evidence_id,
            derived_evidence_count,
            receipt: response.receipt,
            indexes: response.indexes,
            index_refresh_error: response.index_refresh_error,
        })
    }

    /// Returns graph inspection information without exposing storage internals.
    pub async fn inspect_graph(
        &self,
        _request: GraphInspectionRequest,
        context: RequestContext,
    ) -> Result<GraphInspectionResponse, ApiError> {
        let store = self.storage.get().await.map_err(storage_api_error)?;
        let repository_code_totals = store
            .code_repository_totals()
            .await
            .map_err(storage_api_error)?;
        let graph = graph_with_repository_code_totals(
            store.inspect_graph().await.map_err(storage_api_error)?,
            &repository_code_totals,
        );

        Ok(GraphInspectionResponse {
            metadata: ApiMetadata::graph_only(&context, graph.graph_version),
            graph,
            repository_code_totals,
        })
    }

    /// Returns a bounded read-only graph canvas snapshot for the Web workspace.
    pub async fn graph_canvas(
        &self,
        request: GraphCanvasRequest,
        context: RequestContext,
    ) -> Result<GraphCanvasResponse, ApiError> {
        if request.limit == 0 || request.limit > GRAPH_CANVAS_MAX_LIMIT {
            return Err(ApiError::invalid_argument(format!(
                "graph canvas limit must be between 1 and {GRAPH_CANVAS_MAX_LIMIT}"
            )));
        }
        let store = self.storage.get().await.map_err(storage_api_error)?;
        let graph_version = store
            .current_graph_version()
            .await
            .map_err(storage_api_error)?;
        let snapshot = store
            .graph_canvas(GraphCanvasStorageRequest {
                selection: canvas_selection(request.kind),
                source_scope: request.source_scope,
                query: request.query,
                graph_version,
                limit: request.limit,
            })
            .await
            .map_err(storage_api_error)?;
        let node_count = snapshot.nodes.len();
        let edge_count = snapshot.edges.len();

        Ok(GraphCanvasResponse {
            metadata: ApiMetadata::graph_only(&context, graph_version),
            nodes: snapshot
                .nodes
                .into_iter()
                .map(|node| GraphCanvasNode {
                    id: node.id,
                    kind: node.kind,
                    label: node.label,
                    subtitle: node.subtitle,
                    source_scope: node.source_scope,
                    graph_version: node.graph_version.get(),
                    weight: node.weight,
                    status: node.status,
                    details: node.details,
                })
                .collect(),
            edges: snapshot
                .edges
                .into_iter()
                .map(|edge| GraphCanvasEdge {
                    id: edge.id,
                    kind: edge.kind,
                    source: edge.source,
                    target: edge.target,
                    label: edge.label,
                    graph_version: edge.graph_version.get(),
                    confidence_basis_points: edge.confidence_basis_points,
                    evidence_count: edge.evidence_count,
                    details: edge.details,
                })
                .collect(),
            summary: GraphCanvasSummary {
                kind: request.kind,
                node_count,
                edge_count,
                truncated: snapshot.truncated,
                available_kinds: snapshot.available_kinds,
            },
        })
    }

    /// Refreshes derived index metadata up to the current graph version.
    pub async fn refresh_indexes(
        &self,
        request: IndexRefreshRequest,
        context: RequestContext,
    ) -> Result<IndexRefreshResponse, ApiError> {
        let store = self.storage.get().await.map_err(storage_api_error)?;
        let graph_version = store
            .current_graph_version()
            .await
            .map_err(storage_api_error)?;
        let outcome = refresh_index_kinds(
            &store,
            request.kinds,
            graph_version,
            &self.runtime.retrieval,
        )
        .await?;
        let metadata = metadata_for_indexes(&context, graph_version, &outcome.indexes);

        Ok(IndexRefreshResponse {
            metadata,
            indexes: outcome.indexes,
            index_cursors: outcome.cursors,
            diagnostics: outcome.diagnostics,
        })
    }

    /// Probes the configured remote embedding provider without exposing secrets.
    pub async fn probe_embedding_provider(
        &self,
        context: RequestContext,
    ) -> Result<EmbeddingProviderProbeResponse, ApiError> {
        let Some(remote) = self.runtime.retrieval.remote_embedding.clone() else {
            return Ok(EmbeddingProviderProbeResponse {
                metadata: ApiMetadata::graph_only(&context, crate::domain::GraphVersion::ZERO),
                ok: false,
                provider: None,
                model: self.runtime.retrieval.vector_model.name.clone(),
                dimension: self.runtime.retrieval.vector_model.dimension,
                latency_ms: None,
                error_code: Some("remote_embedding_not_configured".to_owned()),
                error_message: Some("remote embedding provider is not configured".to_owned()),
                retryable: Some(false),
            });
        };
        let provider_name = remote.provider.as_str().to_owned();
        let provider = self.embedding_provider.as_ref().ok_or_else(|| {
            ApiError::invalid_argument(
                "remote embedding provider is configured without an embedding adapter",
            )
        })?;
        let started = Instant::now();
        let result = provider
            .embed(EmbeddingRequest {
                inputs: vec!["relay-knowledge provider probe".to_owned()],
                model: self.runtime.retrieval.vector_model.name.clone(),
                dimension: self.runtime.retrieval.vector_model.dimension,
            })
            .await;

        match result {
            Ok(_) => Ok(EmbeddingProviderProbeResponse {
                metadata: ApiMetadata::graph_only(&context, crate::domain::GraphVersion::ZERO),
                ok: true,
                provider: Some(provider_name),
                model: self.runtime.retrieval.vector_model.name.clone(),
                dimension: self.runtime.retrieval.vector_model.dimension,
                latency_ms: Some(duration_millis(started.elapsed())),
                error_code: None,
                error_message: None,
                retryable: None,
            }),
            Err(error) => Ok(EmbeddingProviderProbeResponse {
                metadata: ApiMetadata::graph_only(&context, crate::domain::GraphVersion::ZERO),
                ok: error.code == "rate_limited" && error.retry == ProviderRetryClass::Retryable,
                provider: Some(provider_name),
                model: self.runtime.retrieval.vector_model.name.clone(),
                dimension: self.runtime.retrieval.vector_model.dimension,
                latency_ms: Some(duration_millis(started.elapsed())),
                error_code: Some(error.code),
                error_message: Some(error.message),
                retryable: Some(error.retry == ProviderRetryClass::Retryable),
            }),
        }
    }

    /// Reconciles derived index cursors before resident service work starts.
    pub async fn reconcile_startup_indexes(
        &self,
        context: RequestContext,
    ) -> Result<ServiceRecoveryReport, ApiError> {
        let store = self.storage.get().await.map_err(storage_api_error)?;
        let graph_version = store
            .current_graph_version()
            .await
            .map_err(storage_api_error)?;
        let before = store.index_statuses().await.map_err(storage_api_error)?;
        let active_before = before
            .iter()
            .filter(|status| self.runtime.retrieval.refreshes_index(status.kind))
            .cloned()
            .collect::<Vec<_>>();
        let stale_index_kinds = active_before
            .iter()
            .filter(|status| status.is_stale_for(graph_version))
            .map(|status| status.kind)
            .collect::<Vec<_>>();
        let index_lag_max = active_before
            .iter()
            .map(|status| {
                graph_version
                    .get()
                    .saturating_sub(status.indexed_graph_version.get())
            })
            .max()
            .unwrap_or(0);
        let outcome = if stale_index_kinds.is_empty() {
            index_refresh_outcome(&store).await?
        } else {
            recover_index_kinds(
                &store,
                stale_index_kinds.clone(),
                graph_version,
                &self.runtime.retrieval,
            )
            .await?
        };
        let refreshed = outcome
            .indexes
            .iter()
            .filter(|status| {
                stale_index_kinds.contains(&status.kind) && !status.is_stale_for(graph_version)
            })
            .map(|status| status.kind)
            .collect::<Vec<_>>();
        let after = outcome.indexes;
        let active_after = after
            .iter()
            .filter(|status| self.runtime.retrieval.refreshes_index(status.kind))
            .cloned()
            .collect::<Vec<_>>();
        let metadata = metadata_for_indexes(&context, graph_version, &active_after);

        Ok(ServiceRecoveryReport {
            metadata,
            graph_version: graph_version.get(),
            stale_index_kinds,
            refreshed_index_kinds: refreshed,
            index_lag_max,
            task_queue_depth: outcome.diagnostics.queue_depth,
            dead_letter_count: outcome.diagnostics.dead_letter_count,
            heartbeat_state: "ready".to_owned(),
        })
    }

    pub(super) async fn store(&self) -> Result<Arc<dyn KnowledgeStore>, StorageError> {
        self.storage.get().await
    }

    /// Returns whether graph storage has already been opened by this service.
    pub fn storage_is_ready(&self) -> bool {
        self.storage.ready_store().is_some()
    }

    /// Runs one split-worker preview code-index attempt through durable task leases.
    pub async fn run_code_index_worker_preview(
        &self,
        request: CodeIndexWorkerRunRequest,
        context: RequestContext,
    ) -> Result<CodeIndexWorkerRunResponse, ApiError> {
        let store = self.storage.get().await.map_err(storage_api_error)?;
        let task = self
            .run_code_index_task_once(request.task_id, context.clone())
            .await?;
        let graph_version = store
            .current_graph_version()
            .await
            .map_err(storage_api_error)?;

        Ok(CodeIndexWorkerRunResponse {
            metadata: ApiMetadata::graph_only(&context, graph_version),
            worker_kind: "code_index".to_owned(),
            claimed: task.is_some(),
            task,
        })
    }

    /// Returns the persistent agent audit log path resolved by the path boundary.
    pub fn agent_audit_log_path(&self) -> PathBuf {
        self.runtime.paths.agent_audit_log_file()
    }
}

/// Durable audit event input accepted from resident agent adapters.
#[derive(Debug, Clone)]
pub struct AgentDurableAuditInput {
    pub operation: String,
    pub interface: String,
    pub request_id: String,
    pub trace_id: String,
    pub status: AuditStatus,
    pub actor: Option<String>,
    pub source_scope: Option<String>,
    pub graph_version: u64,
    pub detail_json: String,
    pub message: Option<String>,
}

pub(super) fn storage_api_error(error: StorageError) -> ApiError {
    match error {
        StorageError::CapacityExceeded(message) => ApiError::qos_rejected(message),
        other => ApiError::storage_unavailable(other.to_string()),
    }
}

pub(super) async fn file_index_diagnostics_or_default(
    store: &Arc<dyn KnowledgeStore>,
) -> Result<FileIndexDiagnostics, ApiError> {
    match store.file_index_diagnostics().await {
        Ok(diagnostics) => Ok(diagnostics),
        Err(StorageError::InvalidInput(message))
            if message == "file index storage is unavailable" =>
        {
            Ok(FileIndexDiagnostics::default())
        }
        Err(error) => Err(storage_api_error(error)),
    }
}

pub(super) fn graph_with_repository_code_totals(
    mut graph: GraphInspection,
    repository_totals: &CodeRepositoryTotals,
) -> GraphInspection {
    graph.code_file_count = graph
        .code_file_count
        .saturating_add(repository_totals.indexed_file_count);
    graph.code_symbol_count = graph
        .code_symbol_count
        .saturating_add(repository_totals.symbol_count);
    graph.code_reference_count = graph
        .code_reference_count
        .saturating_add(repository_totals.reference_count);
    graph.code_chunk_count = graph
        .code_chunk_count
        .saturating_add(repository_totals.chunk_count);
    graph.code_parse_status_counts = add_parse_status_counts(
        graph.code_parse_status_counts,
        repository_totals.parse_status_counts,
    );

    graph
}

fn canvas_selection(kind: GraphCanvasKind) -> GraphCanvasSelection {
    match kind {
        GraphCanvasKind::Knowledge => GraphCanvasSelection::Knowledge,
        GraphCanvasKind::Code => GraphCanvasSelection::Code,
        GraphCanvasKind::Mixed => GraphCanvasSelection::Mixed,
    }
}

fn add_parse_status_counts(
    left: CodeParseStatusCounts,
    right: CodeParseStatusCounts,
) -> CodeParseStatusCounts {
    CodeParseStatusCounts {
        parsed: left.parsed.saturating_add(right.parsed),
        partial: left.partial.saturating_add(right.partial),
        text_only: left.text_only.saturating_add(right.text_only),
        failed: left.failed.saturating_add(right.failed),
    }
}

fn duration_millis(duration: std::time::Duration) -> u64 {
    u64::try_from(duration.as_millis()).unwrap_or(u64::MAX)
}

fn service_definition_filename() -> &'static str {
    if cfg!(target_os = "windows") {
        WINDOWS_SERVICE_DEFINITION_FILE_NAME
    } else if cfg!(target_os = "macos") {
        MACOS_SERVICE_DEFINITION_FILE_NAME
    } else {
        LINUX_SERVICE_DEFINITION_FILE_NAME
    }
}

mod health;
mod lifecycle_plan;
mod retrieval;
mod service_status;
mod storage_diagnostics;
mod storage_provider;
mod watcher;

#[cfg(test)]
#[path = "graph_only_tests.rs"]
mod graph_only_tests;

#[cfg(test)]
#[path = "recovery_tests.rs"]
mod recovery_tests;

#[cfg(test)]
#[path = "refresh_tests.rs"]
mod refresh_tests;

#[cfg(test)]
#[path = "storage_tests.rs"]
mod storage_tests;

#[cfg(test)]
#[path = "operations_tests.rs"]
mod operations_tests;

#[cfg(test)]
#[path = "mod_tests.rs"]
mod mod_tests;