@adhdev/daemon-core 0.9.82-rc.504 → 0.9.82-rc.506

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@adhdev/daemon-core",
-  "version": "0.9.82-rc.504",
+  "version": "0.9.82-rc.506",
   "description": "ADHDev daemon core — CDP, IDE detection, providers, command execution",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
@@ -47,8 +47,8 @@
   "author": "vilmire",
   "license": "AGPL-3.0-or-later",
   "dependencies": {
-    "@adhdev/mesh-shared": "0.9.82-rc.504",
-    "@adhdev/session-host-core": "0.9.82-rc.504",
+    "@adhdev/mesh-shared": "0.9.82-rc.506",
+    "@adhdev/session-host-core": "0.9.82-rc.506",
     "@agentclientprotocol/sdk": "^0.16.1",
     "ajv": "^8.20.0",
     "ajv-formats": "^3.0.1",

package/src/mesh/mesh-event-forwarding.ts

@@ -16,7 +16,7 @@ import { enqueueUnresolvedDelegateForward, nudgeUnresolvedForwardRetry } from '.
 import { traceMeshEventStage, traceMeshEventDrop } from './mesh-event-trace.js';
 import { getLastDisplayMessage } from '../status/snapshot.js';
 import { resolveDelegatedWorkerAutoApprove } from '../repo-mesh-types.js';
-import { meshNodeIdMatches, daemonIdsEquivalent, expandDaemonIdForms, sessionIdsEquivalent, type MeshNodeIdentified } from '@adhdev/mesh-shared';
+import { meshNodeIdMatches, daemonIdsEquivalent, expandDaemonIdForms, sessionIdsEquivalent, withStatusProbeMarker, type MeshNodeIdentified } from '@adhdev/mesh-shared';
 import {
     findRecentTerminalLedgerEvidence,
     findTerminalLedgerEvidenceForTask,
@@ -760,7 +760,15 @@ function stopStaleMeshWorker(
             } catch { /* best-effort */ }
         }
         if (daemonId && components.dispatchMeshCommand) {
-            Promise.resolve(components.dispatchMeshCommand(daemonId, 'stop_cli', stopArgs))
+            // OFFLINE-NODE-BLOCKING: this stop is fire-and-forget. Without a short connect-wait,
+            // a stop_cli to a worker node whose daemon is powered off leaves a pending request
+            // hanging for the full 90s connect deadline before its .catch fires (a leaked
+            // pending per stale worker). Stamp the status-origin marker so the daemon-cloud relay
+            // grants the SHORT connect-wait budget — an offline node rejects in ~2s and the .catch
+            // logs it immediately. The marker only affects the connect wait and is stripped before
+            // stop_cli executes, so a live worker is stopped identically. (Best-effort by design:
+            // a failed stop only loses the belt-and-suspenders stop; the ack was already rejected.)
+            Promise.resolve(components.dispatchMeshCommand(daemonId, 'stop_cli', withStatusProbeMarker(stopArgs)))
                 .catch((e: any) => LOG.warn('MeshQueue', `Remote stop of stale worker ${sessionId} on daemon ${daemonId} failed: ${e?.message || e}`));
         } else {
             LOG.warn('MeshQueue', `Cannot stop stale worker ${sessionId}: no local adapter and no resolvable remote daemon id (node ${args.nodeId ?? '?'}). Ack already rejected — task will re-strand-and-fail if the worker completes.`);

package/src/mesh/mesh-node-identity.ts

@@ -14,7 +14,7 @@
 import type { ProviderLoader } from '../providers/provider-loader.js';
 import { detectCLI } from '../detection/cli-detector.js';
 import { getGitRepoStatus } from '../git/git-status.js';
-import { normalizeGitStatus as sharedNormalizeGitStatus, pickBestTransitGitStatus as sharedPickBestTransitGitStatus, summarizeGitShape as sharedSummarizeGitShape, normalizeMeshNodeId, daemonIdsEquivalent, meshWorkspacesEquivalent, sessionIdsEquivalent } from '@adhdev/mesh-shared';
+import { normalizeGitStatus as sharedNormalizeGitStatus, pickBestTransitGitStatus as sharedPickBestTransitGitStatus, summarizeGitShape as sharedSummarizeGitShape, normalizeMeshNodeId, daemonIdsEquivalent, meshWorkspacesEquivalent, sessionIdsEquivalent, withStatusProbeMarker } from '@adhdev/mesh-shared';
 import { LOG } from '../logging/logger.js';
 import { getSessionHostSurfaceKind } from '../session-host/runtime-surface.js';
 import { awaitWithWarmupDeadline, resolveWarmupDeadlineOpts } from '../mesh/mesh-warmup-deadline.js';
@@ -1468,7 +1468,16 @@ async function probeRemoteMeshGitStatus(args: {
     // cold-open handshake to the connect budget and only the warm round trip to
     // the response budget, so the first probe to a cold peer is no longer
     // false-timed-out before its channel has even opened.
-    const dispatch = args.dispatchMeshCommand(args.daemonId, 'git_status', { workspace: args.workspace, refreshUpstream: true });
+    // OFFLINE-NODE-STATUS-REFRESH: stamp the status-origin marker so the daemon-cloud
+    // dispatch wrapper grants this explicit_refresh / mesh_status git_status probe the
+    // SHORT connect-wait budget. Without it, an offline (powered-off) peer sinks the
+    // probe into the 90s connect deadline and blocks the whole status assembly. A
+    // user-driven / targeted git_status (no marker) is unaffected.
+    const dispatch = args.dispatchMeshCommand(
+        args.daemonId,
+        'git_status',
+        withStatusProbeMarker({ workspace: args.workspace, refreshUpstream: true }),
+    );
     // A missing connection getter no longer silently becomes `() => true`
     // ("always warm") — that charged a still-opening channel against the response
     // budget and re-introduced the cold-open false-timeout. resolveWarmupDeadlineOpts

package/src/mesh/mesh-queue-assignment.ts

@@ -15,7 +15,7 @@ import { traceMeshEventDrop } from './mesh-event-trace.js';
 import { awaitWithWarmupDeadline, resolveWarmupDeadlineOpts } from './mesh-warmup-deadline.js';
 import { resolveDelegatedWorkerAutoApprove, resolveProviderMaxParallel, resolveNodeSchedulingPriority, normalizeMeshSchedulingStrategy, resolveMaxParallelTasks, resolveMaxReadonlyParallelTasks } from '../repo-mesh-types.js';
 import type { RepoMeshSchedulingStrategy } from '../repo-mesh-types.js';
-import { normalizeMeshNodeId, meshNodeIdMatches, daemonIdsEquivalent, canonicalDaemonId, normalizeMeshWorkspaceForCompare, meshWorkspacesEquivalent, sessionIdsEquivalent, deriveSlotsFromLegacy, normalizeNodeCapabilitySlots, isMeshTaskDifficulty, type MeshNodeIdentified, type NodeCapabilitySlot, type MeshTaskDifficulty } from '@adhdev/mesh-shared';
+import { normalizeMeshNodeId, meshNodeIdMatches, daemonIdsEquivalent, canonicalDaemonId, normalizeMeshWorkspaceForCompare, meshWorkspacesEquivalent, sessionIdsEquivalent, deriveSlotsFromLegacy, normalizeNodeCapabilitySlots, isMeshTaskDifficulty, withStatusProbeMarker, type MeshNodeIdentified, type NodeCapabilitySlot, type MeshTaskDifficulty } from '@adhdev/mesh-shared';
 import { findTerminalLedgerEvidenceForTask, hasUnterminalDirectDispatchLedgerEntry } from './mesh-events-stale.js';
 import { readNonEmptyString } from './mesh-events-utils.js';
 import { readMeshNodeDaemonId } from './mesh-node-identity.js';
@@ -2100,7 +2100,15 @@ async function maybeAutoLaunchOneQueueSession(components: DaemonComponents, mesh
                     markAutoLaunch(meshId, task.id, { status: 'started', nodeId, providerType: resolved.providerType });
                     let launchResult: any;
                     try {
-                        launchResult = await components.dispatchMeshCommand!(launchTarget.daemonId!, 'launch_cli', {
+                        // OFFLINE-NODE-BLOCKING: no peer-connected pre-check before this remote
+                        // launch_cli meant an OFFLINE target node sank the dispatch into the 90s
+                        // connect deadline, stalling the 4s auto-launch loop for a full 90s. Stamp
+                        // the status-origin marker so the daemon-cloud relay grants the SHORT
+                        // connect-wait budget — an offline node throws in ~2s, the catch below sets
+                        // the 25s cooldown (autoLaunchCooldownUntil) that already gates retries, so
+                        // the loop moves on. The marker only affects the connect wait and is
+                        // stripped before launch_cli executes, so a live node spawns identically.
+                        launchResult = await components.dispatchMeshCommand!(launchTarget.daemonId!, 'launch_cli', withStatusProbeMarker({
                             cliType: resolved.providerType,
                             dir: node.workspace,
                             settings: remoteSettings,
@@ -2110,7 +2118,7 @@ async function maybeAutoLaunchOneQueueSession(components: DaemonComponents, mesh
                             ...(effectiveModel ? { initialModel: effectiveModel } : {}),
                             // BRAIN-ROUTING thinking axis: forward the effective thinking level (initialThinkingLevel).
                             ...(effectiveThinkingLevel ? { initialThinkingLevel: effectiveThinkingLevel } : {}),
-                        });
+                        }));
                     } catch (e: any) {
                         markAutoLaunch(meshId, task.id, { status: 'failed', reason: `remote_launch_dispatch_failed: ${e?.message || String(e)}`, nodeId, providerType: resolved.providerType });
                         autoLaunchCooldownUntil.set(launchKey, Date.now() + AUTO_LAUNCH_COOLDOWN_MS); sweepExpiredCooldowns();

package/src/mesh/mesh-reconcile-loop.ts

@@ -59,7 +59,7 @@ import {
 } from './mesh-unresolved-forward-outbox.js';
 import { readNonEmptyString, readMeshCompletionSummary, buildMeshSystemMessage } from './mesh-events-utils.js';
 import { traceMeshEventStage, traceMeshEventDrop } from './mesh-event-trace.js';
-import { expandDaemonIdForms, daemonIdsEquivalent, sessionIdsEquivalent, meshNodeIdMatches } from '@adhdev/mesh-shared';
+import { expandDaemonIdForms, daemonIdsEquivalent, sessionIdsEquivalent, meshNodeIdMatches, withStatusProbeMarker } from '@adhdev/mesh-shared';
 import { getQueue, reclaimStrandedAssignedTask, updateTaskStatus } from './mesh-work-queue.js';
 import { resolveSessionBusyVerdict } from './mesh-queue-assignment.js';
 import { readLedgerEntries } from './mesh-ledger.js';
@@ -1637,7 +1637,15 @@ async function retryUnresolvedDelegateForwards(components: DaemonComponents): Pr
         let result: any;
         try {
             traceMeshEventStage('forward_send', entryTraceCtx, `retry → ${entry.coordinatorDaemonId}`);
-            result = await dispatchMeshCommand(entry.coordinatorDaemonId, 'mesh_forward_event', pushPayload);
+            // OFFLINE-NODE-BLOCKING: stamp the status-origin marker so the daemon-cloud relay
+            // grants the SHORT connect-wait budget. Without it, a retry to a coordinator whose
+            // daemon is powered off sinks into the 90s connect deadline per entry, serializing
+            // the whole unresolved-forward outbox behind one dead coordinator. With it, the
+            // dispatch throws in ~2s and the entry is left queued for the next tick — the
+            // existing retry-backoff and age-expiry below are unchanged. The marker only
+            // affects the connect wait and is stripped before mesh_forward_event executes, so
+            // delivery semantics are identical.
+            result = await dispatchMeshCommand(entry.coordinatorDaemonId, 'mesh_forward_event', withStatusProbeMarker(pushPayload));
         } catch (e: any) {
             // Coordinator unreachable (transport threw) — keep the entry queued and try again
             // next tick. This is NOT a hard rejection, so it does not count toward the cap;