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
This commit is contained in:
2026-09-23 22:14:15 -04:00
co-authored by Claude Sonnet 5
parent 0f643be719
commit c7dff14807
2 changed files with 163 additions and 27 deletions
+53 -9
View File
@@ -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):
"""-> ("netdev", device) | ("heartbeat", None) | ("none", None)."""
def state_stale(state, now, max_age):
"""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}")
stop.wait(float(cfg["poll_interval_s"]))
log_once("poll", f"poll failed: {e}")
else:
clear_log("poll")
stop.wait(interval(cfg))
log("van-leds stopping, LEDs left as they are")
return 0