fix(reliability): converge runtime on uplink revert; bound health endpoint cost (I3,I4)

payload/user/remote_access/pager-webui/mk8_events.py

@@ -1,5 +1,6 @@
 """Mark VIII reliability event journal. JSONL on /mmc, rotated."""
 import json, os, threading, time
+from collections import deque
 
 MK8_DIR = '/mmc/mk8'
 EVENTS_PATH = os.path.join(MK8_DIR, 'events.log')
@@ -87,6 +88,43 @@ def counters():
     return counts
 
 
+def snapshot(event_limit=20, scan=2000):
+    """Newest-first events (up to event_limit) plus kind counters computed
+    over at most `scan` most-recent entries, in ONE parse pass. Bounded so a
+    large rotated journal cannot spike memory/CPU on every health poll."""
+    events = []
+    counts = {v: 0 for v in _COUNTER_KEYS.values()}
+    scanned = 0
+    paths = [EVENTS_PATH]
+    paths.extend(EVENTS_PATH + '.%d' % i for i in range(KEEP, 0, -1))
+    tail_len = max(scan, event_limit)
+    for path in paths:
+        if len(events) >= event_limit and scanned >= scan:
+            break
+        try:
+            with open(path) as f:
+                tail = deque((l for l in f if l.strip()), maxlen=tail_len)
+        except OSError:
+            continue
+        for line in reversed(tail):
+            try:
+                row = json.loads(line)
+            except ValueError:
+                continue
+            if not isinstance(row, dict):
+                continue
+            scanned += 1
+            kind = row.get('kind')
+            if kind in _COUNTER_KEYS:
+                counts[_COUNTER_KEYS[kind]] += 1
+            if len(events) < event_limit:
+                events.append(row)
+            if len(events) >= event_limit and scanned >= scan:
+                break
+    events.sort(key=lambda r: r.get('ts', 0), reverse=True)
+    return {'events': events[:event_limit], 'reliability': counts}
+
+
 def mark_boot(unexpected=False):
     log_event('unexpected_boot' if unexpected else 'boot', sev='warn'
               if unexpected else 'info',

payload/user/remote_access/pager-webui/mk8_rfplan.py

@@ -94,6 +94,11 @@ def set_role(role, ssid=None, psk=None):
                 break
         if not assoc:
             disable_uplink()
+            # UCI alone does not converge runtime: without a reload wlan1up
+            # keeps scanning/authenticating and pins phy1 until some unrelated
+            # future reload, while current_role() already reports idle.
+            # Converge now like the idle branch, then reapply the hop policy.
+            device_run(['wifi', 'reload'], timeout=60)
             _resume_hop()
             return {'ok': False, 'error': 'association failed; reverted'}
         return {'ok': True, 'role': 'uplink', 'assoc': assoc}

payload/user/remote_access/pager-webui/server.py

@@ -4808,8 +4808,9 @@ def h_health(ctx):
                        for k in ('pass', 'fixed', 'warn', 'fail')},
             'steps': ENV_CHECK_STATE['report'],
         }
-    h['reliability'] = mk8_events.counters()
-    h['events'] = mk8_events.read_events(limit=20)
+    snap = mk8_events.snapshot()
+    h['reliability'] = snap['reliability']
+    h['events'] = snap['events']
     h['guard'] = mk8_guard.guard_report()
     return 200, h