diff options
Diffstat (limited to 'makima/src/server/handlers/mesh_daemon.rs')
| -rw-r--r-- | makima/src/server/handlers/mesh_daemon.rs | 126 |
1 files changed, 102 insertions, 24 deletions
diff --git a/makima/src/server/handlers/mesh_daemon.rs b/makima/src/server/handlers/mesh_daemon.rs index 887183a..34e2cc3 100644 --- a/makima/src/server/handlers/mesh_daemon.rs +++ b/makima/src/server/handlers/mesh_daemon.rs @@ -933,7 +933,7 @@ async fn handle_daemon_connection(socket: WebSocket, state: SharedState, auth_re "Supervisor heartbeat received" ); - // Store heartbeat in database + // Store heartbeat in database and update supervisor state (Task 3.3) if let Some(ref pool) = state.db_pool { let pool = pool.clone(); let pending_ids = pending_task_ids.clone(); @@ -960,6 +960,22 @@ async fn handle_daemon_connection(socket: WebSocket, state: SharedState, auth_re ); } + // Update supervisor_states table (lightweight heartbeat state update - Task 3.3) + if let Err(e) = repository::update_supervisor_heartbeat_state( + &pool, + contract_id, + &state_str, + activity.as_deref(), + progress as i32, + &pending_ids, + ).await { + tracing::debug!( + contract_id = %contract_id, + error = %e, + "Failed to update supervisor state from heartbeat (may not exist yet)" + ); + } + // Also update the daemon heartbeat if let Ok(Some(task)) = repository::get_task(&pool, task_id).await { if let Some(daemon_id) = task.daemon_id { @@ -1035,55 +1051,117 @@ async fn handle_daemon_connection(socket: WebSocket, state: SharedState, auth_re updated_by: "daemon".into(), }); - // Initialize supervisor_state when supervisor task starts running + // Initialize or restore supervisor_state when supervisor task starts running (Task 3.4) if updated_task.is_supervisor && new_status_owned == "running" { if let Some(contract_id) = updated_task.contract_id { - // Get contract to get its phase - match repository::get_contract_for_owner( - &pool, - contract_id, - updated_task.owner_id, - ).await { - Ok(Some(contract)) => { - match repository::upsert_supervisor_state( + // Check if supervisor state already exists (restoration scenario) + match repository::get_supervisor_state(&pool, contract_id).await { + Ok(Some(existing_state)) => { + // State exists - this is a restoration + tracing::info!( + task_id = %task_id, + contract_id = %contract_id, + existing_state = %existing_state.state, + restoration_count = existing_state.restoration_count, + "Supervisor starting with existing state - restoration in progress" + ); + + // Mark as restored (increments restoration_count) + match repository::mark_supervisor_restored( &pool, contract_id, - task_id, - serde_json::json!([]), // Empty conversation - &[], // No pending tasks - &contract.phase, + "daemon_restart", ).await { - Ok(_) => { + Ok(restored_state) => { tracing::info!( task_id = %task_id, contract_id = %contract_id, - phase = %contract.phase, - "Initialized supervisor state for running supervisor" + restoration_count = restored_state.restoration_count, + "Supervisor restoration marked" ); + + // Check for pending questions to re-deliver + if let Ok(questions) = serde_json::from_value::<Vec<crate::db::models::PendingQuestion>>( + restored_state.pending_questions.clone() + ) { + if !questions.is_empty() { + tracing::info!( + contract_id = %contract_id, + question_count = questions.len(), + "Pending questions found for re-delivery" + ); + // Questions will be re-delivered by the supervisor when it restores + } + } } Err(e) => { tracing::warn!( task_id = %task_id, contract_id = %contract_id, error = %e, - "Failed to initialize supervisor state" + "Failed to mark supervisor as restored" ); } } } Ok(None) => { - tracing::warn!( - task_id = %task_id, - contract_id = %contract_id, - "Contract not found when initializing supervisor state" - ); + // No existing state - fresh start + // Get contract to get its phase + match repository::get_contract_for_owner( + &pool, + contract_id, + updated_task.owner_id, + ).await { + Ok(Some(contract)) => { + match repository::upsert_supervisor_state( + &pool, + contract_id, + task_id, + serde_json::json!([]), // Empty conversation + &[], // No pending tasks + &contract.phase, + ).await { + Ok(_) => { + tracing::info!( + task_id = %task_id, + contract_id = %contract_id, + phase = %contract.phase, + "Initialized fresh supervisor state" + ); + } + Err(e) => { + tracing::warn!( + task_id = %task_id, + contract_id = %contract_id, + error = %e, + "Failed to initialize supervisor state" + ); + } + } + } + Ok(None) => { + tracing::warn!( + task_id = %task_id, + contract_id = %contract_id, + "Contract not found when initializing supervisor state" + ); + } + Err(e) => { + tracing::warn!( + task_id = %task_id, + contract_id = %contract_id, + error = %e, + "Failed to get contract for supervisor state" + ); + } + } } Err(e) => { tracing::warn!( task_id = %task_id, contract_id = %contract_id, error = %e, - "Failed to get contract for supervisor state" + "Failed to check existing supervisor state" ); } } |