Harden live server apply for alerts and peer roles

Flush alert config immediately, re-arm runtime without restart, and push
role updates to connected clients so ACL changes apply live.

server/handlers/peers.js

@@ -5,7 +5,7 @@
 import * as peerPolicy from '../core/peer-policy.js'
 import * as connectionInvites from '../core/connection-invites.js'
 import { peers } from '../core/peer-registry.js'
-import { Roles } from '../../shared/protocol.js'
+import { Roles, Pushes } from '../../shared/protocol.js'
 import * as validation from '../utils/validation.js'
 import { decodePeardockInvite } from '../../shared/crypto-auth.js'
 
@@ -263,7 +263,20 @@ export function registerPeerHandlers(session) {
     if (!peerId || !role) throw new Error('peerId and role required')
     const entry = peerPolicy.setPeerRole(peerId, role)
     const live = peers.get(peerId)
-    if (live) live.role = role
-    return { success: true, data: entry }
+    if (live) {
+      // Apply ACL live — no reconnect / server restart
+      live.role = role
+      try {
+        live.push(Pushes.session, {
+          type: 'roleUpdate',
+          role,
+          peerId,
+          updatedAt: new Date().toISOString(),
+        })
+      } catch {
+        // ignore push failures
+      }
+    }
+    return { success: true, data: entry, appliedLive: Boolean(live) }
   })
 }

server/services/alerts.js

@@ -334,15 +334,28 @@ function normalizeConfig(raw) {
   }
 }
 
+/**
+ * Persist config promptly. Coalesce bursts so rapid UI edits don't thrash disk,
+ * but never wait more than ~120ms so process death can't strand live config.
+ */
 function scheduleSave() {
   if (saveTimer) return
   saveTimer = setTimeout(() => {
     saveTimer = null
     saveDisk()
-  }, 400)
+  }, 120)
   if (typeof saveTimer.unref === 'function') saveTimer.unref()
 }
 
+/** Flush any pending debounced save immediately (e.g. before critical ops). */
+function flushSave() {
+  if (saveTimer) {
+    clearTimeout(saveTimer)
+    saveTimer = null
+  }
+  saveDisk()
+}
+
 function saveDisk() {
   try {
     const payload = {
@@ -1260,7 +1273,7 @@ export function updateAlertsConfig(partial, opts = {}) {
     config = normalizeConfig(merged)
   }
   config.updatedAt = new Date().toISOString()
-  saveDisk()
+  flushSave()
   // Live apply: poll interval, enable/disable — no process restart
   applyRuntimeConfig({
     kickPoll: config.enabled && (!wasEnabled || partial.pollIntervalMs != null),
@@ -1284,8 +1297,8 @@ export function upsertAlertChannel(input) {
   } else {
     config.channels.push(ch)
   }
-  scheduleSave()
-  // Channel list is read on each fire — already live; log for ops clarity
+  // In-memory config is used on next fire; flush disk promptly (no restart)
+  flushSave()
   logger.debug('alerts: channel upserted live', {
     id: ch.id,
     enabled: ch.enabled,
@@ -1303,7 +1316,7 @@ export function deleteAlertChannel(id) {
       rule.channelIds = rule.channelIds.filter((x) => x !== id)
     }
   }
-  scheduleSave()
+  flushSave()
   return before !== config.channels.length
 }
 
@@ -1312,7 +1325,7 @@ export function upsertAlertRule(input) {
   const idx = config.rules.findIndex((r) => r.id === rule.id)
   if (idx >= 0) config.rules[idx] = rule
   else config.rules.push(rule)
-  scheduleSave()
+  flushSave()
   logger.debug('alerts: rule upserted live', {
     id: rule.id,
     enabled: rule.enabled,
@@ -1324,7 +1337,7 @@ export function upsertAlertRule(input) {
 export function deleteAlertRule(id) {
   const before = config.rules.length
   config.rules = config.rules.filter((r) => r.id !== id)
-  scheduleSave()
+  flushSave()
   return before !== config.rules.length
 }
 

server/services/schedules.js

@@ -88,6 +88,7 @@ function arm(job) {
     job.timer = null
   }
   if (!job.enabled || job.intervalMs < 60_000) return
+  // Live: re-arm whenever upserted — no server restart required
   job.timer = setInterval(() => {
     runJob(job)
   }, job.intervalMs)