van-leds fix round 1: realistic sysfs fake, error paths, stale failover state, log-once, interval and empty-iface guards

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KqhjP7nud5gdBg25zCiRk9

leds/van-leds

@@ -23,6 +23,8 @@ import os
 import signal
 import sys
 import threading
+import time
+from datetime import datetime
 
 DEFAULTS = {
     "led_power": "", "led_lan": "", "led_wan": "", "led_wlan": "",
@@ -31,6 +33,7 @@ DEFAULTS = {
     "sysfs_leds": "/sys/class/leds",
     "sysfs_net": "/sys/class/net",
     "failover_state": "/run/van-failover/state.json",
+    "state_max_age_s": 60,      # older failover state is ignored (wired-uplink fallback)
 }
 LED_KEYS = (("power", "led_power"), ("lan", "led_lan"), ("wan", "led_wan"), ("wlan", "led_wlan"))
 
@@ -41,6 +44,27 @@ def log(msg):
     print(msg, flush=True)
 
 
+_last = {}
+
+
+def log_once(key, msg):
+    """Log msg unless it is what this key last logged (a persistent failure is logged once)."""
+    if _last.get(key) != msg:
+        _last[key] = msg
+        log(msg)
+
+
+def clear_log(key):
+    _last.pop(key, None)
+
+
+def interval(cfg):
+    try:
+        return max(1.0, float(cfg["poll_interval_s"]))
+    except (TypeError, ValueError):
+        return 5.0
+
+
 def load_config(path):
     cfg = dict(DEFAULTS)
     with open(path) as f:
@@ -65,8 +89,19 @@ def write_attr(path, value):
         f.write(str(value))
 
 
-def desired_wan(state, uplink_carrier, uplink=None):
+def state_stale(state, now, max_age):
-    """-> ("netdev", device) | ("heartbeat", None) | ("none", None)."""
+    """True when state["updated"] parses and is older than max_age seconds (unparseable: not stale)."""
+    try:
+        t = datetime.strptime(state["updated"], "%Y-%m-%dT%H:%M:%S%z").timestamp()
+    except (TypeError, KeyError, ValueError):
+        return False
+    return now - t > max_age
+
+
+def desired_wan(state, uplink_carrier, uplink=None, now=None, max_age=60):
+    """-> ("netdev", device) | ("heartbeat", None) | ("none", None). `now`: epoch seconds (default: clock)."""
+    if isinstance(state, dict) and state_stale(state, time.time() if now is None else now, max_age):
+        state = None
     valid = isinstance(state, dict) and isinstance(state.get("wans"), list) and "active_device" in state
     if not valid:  # no failover info: judge by the wired uplink alone
         if not uplink_carrier:
@@ -113,9 +148,10 @@ def carrier(cfg, iface):
     return bool(iface) and read_attr(os.path.join(cfg["sysfs_net"], iface, "carrier")) == "1"
 
 
-def wanted(cfg):
+def wanted(cfg, now=None):
     """{role: (trigger, device)} for every role (before skipping absent LEDs)."""
-    wan = desired_wan(load_state(cfg["failover_state"]), carrier(cfg, cfg["uplink_iface"]), cfg["uplink_iface"] or None)
+    wan = desired_wan(load_state(cfg["failover_state"]), carrier(cfg, cfg["uplink_iface"]),
+                      cfg["uplink_iface"] or None, now, cfg["state_max_age_s"])
     return {
         "power": ("default-on", None),
         "lan": ("netdev", cfg["lan_iface"]),
@@ -133,6 +169,11 @@ def usable_leds(cfg, announce=False):
             if announce:
                 log(f"LED {role}: no name configured, skipped")
             continue
+        iface_key = {"lan": "lan_iface", "wlan": "wlan_iface"}.get(role)
+        if iface_key and not cfg[iface_key]:
+            if announce:
+                log(f"LED {role}: {iface_key} is empty, skipped")
+            continue
         d = os.path.join(cfg["sysfs_leds"], name)
         if not os.path.isdir(d):
             if announce:
@@ -142,17 +183,18 @@ def usable_leds(cfg, announce=False):
     return out
 
 
-def poll_once(cfg):
+def poll_once(cfg, now=None):
     """One evaluation of every usable LED. Returns the files written."""
-    want = wanted(cfg)
+    want = wanted(cfg, now)
     writes = []
     for role, d in usable_leds(cfg).items():
         trig, dev = want[role]
         try:
             w = apply_led(d, trig, dev)
         except OSError as e:
-            log(f"LED {role}: cannot set {trig}: {e}")
+            log_once(role, f"LED {role}: cannot set {trig}: {e}")
             continue
+        clear_log(role)
         if w:
             log(f"LED {role}: {trig}" + (f" {dev}" if dev else ""))
         writes += w
@@ -170,8 +212,10 @@ def main(argv):
         try:
             poll_once(cfg)
         except Exception as e:  # never die on a transient sysfs/state oddity
-            log(f"poll failed: {e}")
+            log_once("poll", f"poll failed: {e}")
-        stop.wait(float(cfg["poll_interval_s"]))
+        else:
+            clear_log("poll")
+        stop.wait(interval(cfg))
     log("van-leds stopping, LEDs left as they are")
     return 0
 

tests/check-templates.sh

@@ -847,42 +847,98 @@ for label, bs in (("missing", None), ("invalid", "garbage"), ("no wans", {"x": 1
     check(f"22b: {label} state + carrier -> netdev on the uplink", d(bs, True, "up0") == ("netdev", "up0"))
     check(f"22b: {label} state + no carrier -> none", d(bs, False, "up0") == ("none", None))
 
-# fake sysfs
+check("22b: healthy None -> heartbeat", d(st("eth1", [{"device": "eth1", "healthy": None}]), True) == ("heartbeat", None))
+check("22b: active device absent from wans -> heartbeat", d(st("ghost", [{"device": "eth1", "healthy": True}]), True) == ("heartbeat", None))
+# stale state (fixed clock): older than 60 s is treated as missing
+import time as _t, datetime as _dt
+NOW = 1_800_000_000
+def stamp(age): return _dt.datetime.fromtimestamp(NOW - age, _dt.timezone.utc).strftime("%Y-%m-%dT%H:%M:%S%z")
+fresh, old = dict(H, updated=stamp(10)), dict(H, updated=stamp(120))
+check("22b: fresh state is used", d(fresh, False, "up0", NOW) == ("netdev", "eth1"))
+check("22b: stale state + carrier -> netdev on the uplink", d(old, True, "up0", NOW) == ("netdev", "up0"))
+check("22b: stale state + no carrier -> none", d(old, False, "up0", NOW) == ("none", None))
+check("22b: unparseable updated is valid", d(dict(H, updated="garbage"), False, "up0", NOW) == ("netdev", "eth1"))
+check("22b: missing updated is valid", d({k: v for k, v in H.items() if k != "updated"}, False, "up0", NOW) == ("netdev", "eth1"))
+check("22b: max_age is a parameter", d(old, False, "up0", NOW, 300) == ("netdev", "eth1"))
+
+# fake sysfs with the kernel's formats and write rules
+TRIGS = ["none", "default-on", "heartbeat", "netdev", "activity"]
+NDATTR = ("device_name", "link", "tx", "rx")
 leds, net = os.path.join(tmp, "leds"), os.path.join(tmp, "net")
+def wr(led, f, v): open(os.path.join(leds, led, f), "w").write(v + "\n")
+def rd(led, f): return open(os.path.join(leds, led, f)).read().strip()
+def settrig(led, t): wr(led, "trigger", " ".join(f"[{x}]" if x == t else x for x in TRIGS))
 for n in ("red:power", "green:lan", "green:wan", "green:wlan"):
     os.makedirs(os.path.join(leds, n))
-    for f, v in (("trigger", "none"), ("brightness", "0"), ("device_name", ""), ("link", "0"), ("tx", "0"), ("rx", "0")):
+    settrig(n, "none"); wr(n, "brightness", "0")
-        open(os.path.join(leds, n, f), "w").write(v + "\n")
+    for f in NDATTR: wr(n, f, "")
 for i, c in (("lan0", "1"), ("wl0", "1"), ("up0", "1"), ("eth1", "1")):
     os.makedirs(os.path.join(net, i))
     open(os.path.join(net, i, "carrier"), "w").write(c + "\n")
+log_w = []            # every write: (led, attr, value)
+netdev_loaded = [True]
+def kernel_write(path, value):
+    led, attr = os.path.basename(os.path.dirname(path)), os.path.basename(path)
+    value = str(value)
+    log_w.append((led, attr, value))
+    if attr == "trigger":
+        if value not in TRIGS or (value == "netdev" and not netdev_loaded[0]):
+            raise OSError(22, "Invalid argument")
+        if value != rd(led, "trigger").split("[")[1].split("]")[0]:
+            for f in NDATTR: wr(led, f, "")      # leaving netdev drops its attributes
+        settrig(led, value)
+    elif attr in NDATTR:
+        if "[netdev]" not in rd(led, "trigger"):
+            raise OSError(2, "No such file or directory")
+        wr(led, attr, value)
+    else:
+        wr(led, attr, value)
+orig_write = m.write_attr
+m.write_attr = kernel_write
 sf = os.path.join(tmp, "state.json")
 json.dump(H, open(sf, "w"))
 cfg = dict(m.DEFAULTS, led_power="red:power", led_lan="green:lan", led_wan="green:wan", led_wlan="green:wlan",
            lan_iface="lan0", wlan_iface="wl0", uplink_iface="up0", sysfs_leds=leds, sysfs_net=net, failover_state=sf)
-def rd(led, f): return open(os.path.join(leds, led, f)).read().strip()
 def poll(c=cfg):
+    del log_w[:]
     with contextlib.redirect_stdout(io.StringIO()):
-        return m.poll_once(c)
+        m.poll_once(c)
+    return list(log_w)
+def trig(led): return rd(led, "trigger").split("[")[1].split("]")[0]
 w = poll()
-check("22c: power default-on", rd("red:power", "trigger") == "default-on")
+check("22c: power default-on", trig("red:power") == "default-on")
-check("22c: lan netdev on lan0, link/tx/rx 1", (rd("green:lan", "trigger"), rd("green:lan", "device_name"), rd("green:lan", "link"), rd("green:lan", "tx"), rd("green:lan", "rx")) == ("netdev", "lan0", "1", "1", "1"))
+check("22c: lan netdev on lan0, link/tx/rx 1", (trig("green:lan"), rd("green:lan", "device_name"), rd("green:lan", "link"), rd("green:lan", "tx"), rd("green:lan", "rx")) == ("netdev", "lan0", "1", "1", "1"))
-check("22c: wlan netdev on wl0", (rd("green:wlan", "trigger"), rd("green:wlan", "device_name")) == ("netdev", "wl0"))
+check("22c: wlan netdev on wl0", (trig("green:wlan"), rd("green:wlan", "device_name")) == ("netdev", "wl0"))
-check("22c: wan netdev on the active device", (rd("green:wan", "trigger"), rd("green:wan", "device_name")) == ("netdev", "eth1"))
+check("22c: wan netdev on the active device", (trig("green:wan"), rd("green:wan", "device_name")) == ("netdev", "eth1"))
-check("22c: first run writes", len(w) > 0)
+check("22c: trigger is written before its netdev attributes", w.index(("green:lan", "trigger", "netdev")) < w.index(("green:lan", "device_name", "lan0")))
-w = poll()
+check("22c: second run makes ZERO writes on the kernel-format trigger list", poll() == [])
-check("22c: second run makes no writes", w == [])
 json.dump(U, open(sf, "w")); w = poll()
-check("22c: unhealthy state -> wan heartbeat", rd("green:wan", "trigger") == "heartbeat" and len(w) == 1)
+check("22c: netdev -> heartbeat: one trigger write, no attribute writes", trig("green:wan") == "heartbeat" and w == [("green:wan", "trigger", "heartbeat")])
+check("22c: no stale netdev attributes after leaving netdev", rd("green:wan", "device_name") == "" and rd("green:wan", "link") == "")
+check("22c: heartbeat is stable (zero writes)", poll() == [])
 json.dump(N, open(sf, "w")); open(os.path.join(net, "up0", "carrier"), "w").write("0\n"); poll()
-check("22c: nothing -> wan none, brightness 0", rd("green:wan", "trigger") == "none" and rd("green:wan", "brightness") == "0")
+check("22c: -> none: trigger none, brightness 0", trig("green:wan") == "none" and rd("green:wan", "brightness") == "0")
+wr("green:wan", "brightness", "1"); w = poll()
+check("22c: stray brightness under none is reset", w == [("green:wan", "brightness", "0")])
+json.dump(H, open(sf, "w")); w = poll()
+check("22c: none -> netdev: trigger first, then device_name/link/tx/rx", w == [("green:wan", "trigger", "netdev"), ("green:wan", "device_name", "eth1"), ("green:wan", "link", "1"), ("green:wan", "tx", "1"), ("green:wan", "rx", "1")])
 open(sf, "w").write("{broken"); open(os.path.join(net, "up0", "carrier"), "w").write("1\n"); poll()
-check("22c: invalid state falls back to wired uplink", (rd("green:wan", "trigger"), rd("green:wan", "device_name")) == ("netdev", "up0"))
+check("22c: invalid state falls back to wired uplink", (trig("green:wan"), rd("green:wan", "device_name")) == ("netdev", "up0"))
-open(os.path.join(leds, "green:lan", "trigger"), "w").write("none\n"); w = poll()
+settrig("green:lan", "none"); w = poll()
-check("22c: a reset trigger is re-asserted", rd("green:lan", "trigger") == "netdev" and any(x.endswith("trigger") for x in w))
+check("22c: a reset trigger is re-asserted", trig("green:lan") == "netdev" and rd("green:lan", "device_name") == "lan0")
+json.dump(dict(H, updated=stamp(0)), open(sf, "w"))
+def poll_at(now):
+    del log_w[:]
+    with contextlib.redirect_stdout(io.StringIO()):
+        m.poll_once(cfg, now)
+json.dump(dict(H, updated=stamp(300)), open(sf, "w")); poll_at(NOW)
+check("22c: stale state file -> wired-uplink fallback", rd("green:wan", "device_name") == "up0")
+json.dump(dict(H, updated=stamp(1)), open(sf, "w")); poll_at(NOW)
+check("22c: fresh state file retargets the WAN LED", rd("green:wan", "device_name") == "eth1")
+json.dump(H, open(sf, "w"))
 c2 = dict(cfg, led_lan="green:nope", led_wlan="")
 try:
-    w = poll(c2); ok_ = True
+    poll(c2); ok_ = True
 except Exception as e:
     ok_ = False
 check("22c: missing / empty LED names are skipped without raising", ok_)
@@ -890,7 +946,43 @@ buf = io.StringIO()
 with contextlib.redirect_stdout(buf):
     m.usable_leds(c2, announce=True)
 check("22c: skipped LEDs are logged once at start", buf.getvalue().count("skipped") == 2)
+c3 = dict(cfg, lan_iface="")
+buf = io.StringIO()
+with contextlib.redirect_stdout(buf):
+    r3 = m.usable_leds(c3, announce=True)
+check("22c: named LED with empty iface is skipped and logged", "lan" not in r3 and "lan_iface is empty" in buf.getvalue())
+settrig("green:lan", "none")
+poll(c3)
+check("22c: empty iface: no netdev trigger written to that LED", trig("green:lan") == "none")
 check("22c: apply_led writes only on difference", m.apply_led(os.path.join(leds, "red:power"), "default-on") == [])
+check("22c: poll_interval_s: clamped to >=1, non-numeric -> 5", (m.interval({"poll_interval_s": 0}), m.interval({"poll_interval_s": "x"}), m.interval({"poll_interval_s": 7})) == (1.0, 5.0, 7.0))
+# error paths: netdev module not loaded
+for n in ("green:lan", "green:wlan", "green:wan"): settrig(n, "none")
+netdev_loaded[0] = False
+buf = io.StringIO()
+try:
+    with contextlib.redirect_stdout(buf):
+        m.poll_once(cfg); m.poll_once(cfg)
+    ok_ = True
+except Exception:
+    ok_ = False
+check("22c: netdev not loaded: poll_once does not raise", ok_)
+check("22c: netdev not loaded: other LEDs still update", trig("red:power") == "default-on")
+check("22c: persistent failure logged once per LED, not every poll", buf.getvalue().count("cannot set netdev") == 3)
+netdev_loaded[0] = True; w = poll()
+check("22c: next poll retries and succeeds", trig("green:lan") == "netdev" and trig("green:wan") == "netdev")
+# error path: a real unwritable trigger file (skipped as root)
+m.write_attr = orig_write
+if os.geteuid() != 0:
+    settrig("green:lan", "none"); settrig("red:power", "none")
+    os.chmod(os.path.join(leds, "green:lan", "trigger"), 0o444)
+    try:
+        poll()
+        ok_ = True
+    except Exception:
+        ok_ = False
+    check("22c: unwritable trigger: no raise, other LEDs update", ok_ and open(os.path.join(leds, "red:power", "trigger")).read().strip() == "default-on")
+    os.chmod(os.path.join(leds, "green:lan", "trigger"), 0o644)
 print("STATUS", int(bad))
 PY
 )