@adhdev/daemon-core 0.9.82-rc.5 → 0.9.82-rc.50

package/src/mesh/mesh-events.ts

@@ -1,57 +1,106 @@
+import { appendFileSync, existsSync, readFileSync, unlinkSync } from 'fs';
+import { join } from 'path';
 import type { DaemonComponents } from '../boot/daemon-lifecycle.js';
 import { loadConfig } from '../config/config.js';
 import { getMesh, getMeshByRepo } from '../config/mesh-config.js';
 import { detectCLI } from '../detection/cli-detector.js';
 import { LOG } from '../logging/logger.js';
-import { appendLedgerEntry, buildTaskCompletionEvidence, getSessionRecoveryContext, isIntentionalCleanupStopEntry, readLedgerEntries } from './mesh-ledger.js';
+import { appendLedgerEntry, buildTaskCompletionEvidence, getLedgerDir, getSessionRecoveryContext, isIntentionalCleanupStopEntry, readLedgerEntries } from './mesh-ledger.js';
 import type { MeshLedgerKind, SessionRecoveryContext } from './mesh-ledger.js';
 import { claimNextTask, updateSessionTaskStatus, enqueueTask, updateTaskStatus, getQueue, recordTaskAutoLaunch } from './mesh-work-queue.js';
 
 // ---------------------------------------------------------------------------
 // Remote Node Idle Session Tracking
 // ---------------------------------------------------------------------------
-// Tracks remote sessions that emitted 'agent:ready' so triggerMeshQueue 
-// can assign tasks to them.
+// Tracks remote sessions that emitted 'agent:ready' so triggerMeshQueue
+// can assign tasks to them. Each entry carries an expiresAt timestamp;
+// entries are swept on insertion to prevent unbounded growth.
 // ---------------------------------------------------------------------------
 interface RemoteIdleSession {
     nodeId: string;
     sessionId: string;
     providerType: string;
+    expiresAt: number;
 }
+const REMOTE_IDLE_SESSION_TTL_MS = 5 * 60 * 1000; // 5 minutes
 const remoteIdleSessions = new Map<string, RemoteIdleSession>(); // key: `${nodeId}:${sessionId}`
 
+function sweepExpiredRemoteIdleSessions(): void {
+    const now = Date.now();
+    for (const [key, session] of remoteIdleSessions) {
+        if (session.expiresAt <= now) remoteIdleSessions.delete(key);
+    }
+}
+
 // ---------------------------------------------------------------------------
-// MCP coordinator pending-event queue
+// MCP coordinator pending-event queue — FILE-BASED PERSISTENCE
 // ---------------------------------------------------------------------------
 // When a mesh event fires but no CLI coordinator session is registered (e.g.
-// the coordinator is Claude Code running via MCP), we buffer the event here.
-// The MCP server drains this queue on every mesh_status / mesh_send_task poll.
+// the coordinator is Claude Code running via MCP), we persist the event to a
+// per-mesh JSONL file so it survives daemon restarts. The 50-entry hard cap
+// is removed; the file is drained atomically on each get_pending_mesh_events
+// call and limited to 100 KB to prevent runaway growth.
+//
+// File: <ledgerDir>/<meshId>.pending-events.jsonl
 // ---------------------------------------------------------------------------
 
 export interface PendingMeshCoordinatorEvent {
     event: string;
     meshId: string;
     nodeLabel: string;
+    nodeId?: string;
+    workspace?: string;
     metadataEvent: Record<string, unknown>;
     queuedAt: number;
 }
 
-const MAX_PENDING_EVENTS = 50;
-const pendingMeshCoordinatorEvents: PendingMeshCoordinatorEvent[] = [];
+function getPendingEventsPath(meshId: string): string {
+    const safe = meshId.replace(/[^a-zA-Z0-9_-]/g, '_');
+    return join(getLedgerDir(), `${safe}.pending-events.jsonl`);
+}
+
+export function queuePendingMeshCoordinatorEvent(event: PendingMeshCoordinatorEvent): boolean {
+    try {
+        appendFileSync(getPendingEventsPath(event.meshId), JSON.stringify(event) + '\n', 'utf-8');
+        return true;
+    } catch (e: any) {
+        LOG.warn('MeshEvents', `Failed to persist pending coordinator event: ${e?.message || e}`);
+        return false;
+    }
+}
 
-/** Drain and return all pending coordinator events, clearing the queue. */
-export function drainPendingMeshCoordinatorEvents(): PendingMeshCoordinatorEvent[] {
-    return pendingMeshCoordinatorEvents.splice(0);
+/** Drain and return all pending coordinator events for meshId, removing them from disk. */
+export function drainPendingMeshCoordinatorEvents(meshId?: string): PendingMeshCoordinatorEvent[] {
+    if (!meshId) return [];
+    const path = getPendingEventsPath(meshId);
+    if (!existsSync(path)) return [];
+    try {
+        const raw = readFileSync(path, 'utf-8');
+        try { unlinkSync(path); } catch { /* concurrent drain already removed it */ }
+        return raw.split('\n').filter(Boolean).flatMap(line => {
+            try { return [JSON.parse(line) as PendingMeshCoordinatorEvent]; } catch { return []; }
+        });
+    } catch { return []; }
 }
 
 /** Peek at pending coordinator events without draining (non-destructive). */
-export function getPendingMeshCoordinatorEvents(): readonly PendingMeshCoordinatorEvent[] {
-    return pendingMeshCoordinatorEvents.slice();
+export function getPendingMeshCoordinatorEvents(meshId?: string): readonly PendingMeshCoordinatorEvent[] {
+    if (!meshId) return [];
+    const path = getPendingEventsPath(meshId);
+    if (!existsSync(path)) return [];
+    try {
+        const raw = readFileSync(path, 'utf-8');
+        return raw.split('\n').filter(Boolean).flatMap(line => {
+            try { return [JSON.parse(line) as PendingMeshCoordinatorEvent]; } catch { return []; }
+        });
+    } catch { return []; }
 }
 
-/** Explicitly clear all pending coordinator events. */
-export function clearPendingMeshCoordinatorEvents(): void {
-    pendingMeshCoordinatorEvents.splice(0);
+/** Explicitly clear all pending coordinator events for a mesh. */
+export function clearPendingMeshCoordinatorEvents(meshId?: string): void {
+    if (!meshId) return;
+    const path = getPendingEventsPath(meshId);
+    if (existsSync(path)) try { unlinkSync(path); } catch { /* already removed */ }
 }
 
 function readNonEmptyString(value: unknown): string {
@@ -140,6 +189,62 @@ function shouldSuppressIntentionalCleanupStop(args: {
     return hasRecentIntentionalCleanupStop(args.meshId, args.sessionId, args.nodeId);
 }
 
+const RECENT_COMPLETION_FINGERPRINT_TTL_MS = 10 * 60 * 1000;
+const recentCompletionFingerprints = new Map<string, number>();
+
+function readEventTimestamp(value: unknown): number | null {
+    if (typeof value === 'number' && Number.isFinite(value)) return value;
+    if (typeof value === 'string' && value.trim()) {
+        const numeric = Number(value);
+        if (Number.isFinite(numeric)) return numeric;
+        const parsed = Date.parse(value);
+        if (Number.isFinite(parsed)) return parsed;
+    }
+    return null;
+}
+
+function buildMeshCompletionFingerprint(args: {
+    meshId: string;
+    event: string;
+    sessionId: string;
+    providerType?: string;
+    providerSessionId?: string;
+    timestamp?: number | null;
+    finalSummary?: string;
+}): string {
+    const timestampPart = Number.isFinite(args.timestamp)
+        ? String(args.timestamp)
+        : readNonEmptyString(args.finalSummary).slice(0, 200);
+    return [
+        args.meshId,
+        args.event,
+        args.sessionId,
+        args.providerType || '',
+        args.providerSessionId || '',
+        timestampPart,
+    ].join('::');
+}
+
+function isDuplicateMeshCompletionEvent(args: {
+    meshId: string;
+    event: string;
+    sessionId: string;
+    providerType?: string;
+    providerSessionId?: string;
+    timestamp?: number | null;
+    finalSummary?: string;
+}): boolean {
+    const fingerprint = buildMeshCompletionFingerprint(args);
+    if (!fingerprint) return false;
+    const now = Date.now();
+    for (const [key, seenAt] of recentCompletionFingerprints.entries()) {
+        if (now - seenAt > RECENT_COMPLETION_FINGERPRINT_TTL_MS) recentCompletionFingerprints.delete(key);
+    }
+    if (recentCompletionFingerprints.has(fingerprint)) return true;
+    recentCompletionFingerprints.set(fingerprint, now);
+    return false;
+}
+
 
 export function tryAssignQueueTask(
     components: DaemonComponents,
@@ -170,7 +275,16 @@ export function tryAssignQueueTask(
                 message: task.message,
             }).catch((e: any) => {
                 LOG.error('MeshQueue', `Failed to dispatch task via P2P to remote node ${nodeId}: ${e?.message}`);
-                updateTaskStatus(meshId, task.id, 'failed');
+                // Revert to pending so the task can be retried rather than permanently failing
+                updateTaskStatus(meshId, task.id, 'pending');
+                try {
+                    appendLedgerEntry(meshId, {
+                        kind: 'dispatch_failed' as any,
+                        nodeId,
+                        sessionId,
+                        payload: { taskId: task.id, error: e?.message, retryable: true },
+                    });
+                } catch { /* ledger write is best-effort */ }
             });
             return true;
         }
@@ -593,6 +707,23 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
         return { success: true, forwarded: 0, suppressed: true, intentionalCleanupStop: true };
     }
 
+    const eventTimestamp = readEventTimestamp(args.metadataEvent.timestamp);
+    if (args.event === 'agent:generating_completed' && eventSessionId) {
+        const duplicateCompletion = isDuplicateMeshCompletionEvent({
+            meshId: args.meshId,
+            event: args.event,
+            sessionId: eventSessionId,
+            providerType: readNonEmptyString(args.metadataEvent.providerType) || undefined,
+            providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
+            timestamp: eventTimestamp,
+            finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
+        });
+        if (duplicateCompletion) {
+            LOG.info('MeshEvents', `Suppressed duplicate completion for mesh ${args.meshId} session ${eventSessionId}`);
+            return { success: true, forwarded: 0, suppressed: true, duplicateCompletion: true };
+        }
+    }
+
     // ── Task Queue & Ledger ──
     let completedTaskForLedger: { id?: string } | null = null;
     if (args.event === 'agent:generating_completed') {
@@ -601,13 +732,16 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
         const providerType = readNonEmptyString(args.metadataEvent.providerType);
         
         if (sessionId) {
-            const completedTask = updateSessionTaskStatus(args.meshId, sessionId, 'completed');
+            const completedTask = updateSessionTaskStatus(args.meshId, sessionId, 'completed', {
+                occurredAt: eventTimestamp !== null ? new Date(eventTimestamp).toISOString() : undefined,
+            });
             completedTaskForLedger = completedTask ? { id: completedTask.id } : null;
             if (nodeId && providerType) {
-                // Short delay to allow completion event to propagate before pulling next
-                setTimeout(() => {
+                // Queue state is already updated above; setImmediate avoids the
+                // 500 ms artificial delay while still deferring past this call frame.
+                setImmediate(() => {
                     tryAssignQueueTask(components, args.meshId, nodeId, sessionId, providerType);
-                }, 500);
+                });
             }
         }
     } else if (args.event === 'agent:ready') {
@@ -648,13 +782,15 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
         }
         
         if (sessionId && nodeId && providerType) {
-            remoteIdleSessions.set(`${nodeId}:${sessionId}`, { nodeId, sessionId, providerType });
-            setTimeout(() => {
+            sweepExpiredRemoteIdleSessions();
+            remoteIdleSessions.set(`${nodeId}:${sessionId}`, {
+                nodeId, sessionId, providerType,
+                expiresAt: Date.now() + REMOTE_IDLE_SESSION_TTL_MS,
+            });
+            setImmediate(() => {
                 const assigned = tryAssignQueueTask(components, args.meshId, nodeId, sessionId, providerType);
-                if (assigned) {
-                    remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
-                }
-            }, 500);
+                if (assigned) remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
+            });
         }
     } else if (args.event === 'agent:generating_started') {
         const sessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId);
@@ -781,17 +917,18 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
 
     if (coordinatorInstances.length === 0) {
         // No CLI coordinator session found — buffer for MCP-based coordinators.
-        if (pendingMeshCoordinatorEvents.length < MAX_PENDING_EVENTS) {
-            pendingMeshCoordinatorEvents.push({
+        if (queuePendingMeshCoordinatorEvent({
                 event: args.event,
                 meshId: args.meshId,
                 nodeLabel: args.nodeLabel,
+                nodeId: args.nodeId || undefined,
+                workspace: readNonEmptyString(args.metadataEvent.workspace),
                 metadataEvent: {
                     ...args.metadataEvent,
                     ...(recoveryContext ? { recoveryContext } : {}),
                 },
                 queuedAt: Date.now(),
-            });
+            })) {
             LOG.info('MeshEvents', `Queued ${args.event} for MCP coordinator (mesh ${args.meshId})`);
         }
         return { success: true, forwarded: 0 };
@@ -834,6 +971,7 @@ export function handleMeshForwardEvent(components: DaemonComponents, payload: Re
             providerType: readNonEmptyString(payload.providerType),
             providerSessionId: readNonEmptyString(payload.providerSessionId),
             finalSummary: readNonEmptyString(payload.finalSummary) || readNonEmptyString(payload.summary),
+            ...(payload.timestamp !== undefined ? { timestamp: payload.timestamp } : {}),
             intentional: payload.intentional === true,
             intentionalStop: payload.intentionalStop === true,
             operatorCleanup: payload.operatorCleanup === true,

package/src/mesh/mesh-host-ownership.ts

@@ -0,0 +1,73 @@
+import type { RepoMeshDaemonRole, RepoMeshHostMetadata, RepoMeshHostStatus } from '../repo-mesh-types.js';
+
+function readObject(value: unknown): Record<string, unknown> | null {
+    return value && typeof value === 'object' && !Array.isArray(value) ? value as Record<string, unknown> : null;
+}
+
+function readString(value: unknown): string | undefined {
+    return typeof value === 'string' && value.trim() ? value.trim() : undefined;
+}
+
+export function normalizeMeshDaemonRole(value: unknown): RepoMeshDaemonRole | undefined {
+    return value === 'host' || value === 'member' ? value : undefined;
+}
+
+export function resolveMeshHostStatus(mesh: unknown): RepoMeshHostStatus {
+    const meshRecord = readObject(mesh);
+    const raw = readObject(meshRecord?.meshHost);
+    const role = normalizeMeshDaemonRole(raw?.role) ?? 'host';
+    const pairing = readObject(raw?.pairing);
+    const normalized: RepoMeshHostStatus = {
+        role,
+        canOwnCoordinator: role === 'host',
+        canOwnQueue: role === 'host',
+        defaulted: !raw,
+    };
+    const hostDaemonId = readString(raw?.hostDaemonId);
+    const hostNodeId = readString(raw?.hostNodeId);
+    const hostAddress = readString(raw?.hostAddress);
+    if (hostDaemonId) normalized.hostDaemonId = hostDaemonId;
+    if (hostNodeId) normalized.hostNodeId = hostNodeId;
+    if (hostAddress) normalized.hostAddress = hostAddress;
+    if (pairing) {
+        const status = pairing.status === 'pairing' || pairing.status === 'paired' || pairing.status === 'rejected' || pairing.status === 'revoked'
+            ? pairing.status
+            : 'not_configured';
+        normalized.pairing = {
+            status,
+            ...(readString(pairing.tokenId) ? { tokenId: readString(pairing.tokenId) } : {}),
+            ...(readString(pairing.joinedAt) ? { joinedAt: readString(pairing.joinedAt) } : {}),
+            ...(readString(pairing.lastPairedAt) ? { lastPairedAt: readString(pairing.lastPairedAt) } : {}),
+            ...(readString(pairing.lastRejectedAt) ? { lastRejectedAt: readString(pairing.lastRejectedAt) } : {}),
+            ...(readString(pairing.expiresAt) ? { expiresAt: readString(pairing.expiresAt) } : {}),
+        };
+    }
+    return normalized;
+}
+
+export function isMeshHostOwner(mesh: unknown): boolean {
+    return resolveMeshHostStatus(mesh).role === 'host';
+}
+
+export function buildMeshHostRequiredFailure(mesh: unknown, operation: string): Record<string, unknown> {
+    const meshHost = resolveMeshHostStatus(mesh);
+    return {
+        success: false,
+        code: 'mesh_host_required',
+        error: `Mesh Host daemon required for ${operation}; member daemons must pair with the host and cannot own coordinator/queue mutations.`,
+        meshHost,
+    };
+}
+
+export function requireMeshHostQueueOwner(opts?: { ownerRole?: RepoMeshDaemonRole }): void {
+    if (opts?.ownerRole === 'member') {
+        throw new Error('Mesh Host daemon required to mutate mesh queue; member daemons must use the host-owned queue.');
+    }
+}
+
+export function createDefaultMeshHostMetadata(): RepoMeshHostMetadata {
+    return {
+        role: 'host',
+        pairing: { status: 'not_configured' },
+    };
+}

package/src/mesh/mesh-ledger.ts

@@ -31,6 +31,7 @@ export type MeshLedgerKind =
     | 'session_stopped'
     | 'checkpoint_created'
     | 'node_cloned'
+    | 'node_joined'
     | 'node_removed'
     | 'coordinator_started'
     | 'recovery_attempted'