Improve update reliability and responsiveness (#1986)

* handle orphaned git lock

* trigger update after going offroad

* git ping

launch_chffrplus.sh

@@ -14,12 +14,17 @@ if [ -z "$PASSIVE" ]; then
   export PASSIVE="1"
 fi
 
+DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null && pwd )"
+
 STAGING_ROOT="/data/safe_staging"
 
 function launch {
   # Wifi scan
   wpa_cli IFNAME=wlan0 SCAN
 
+  # Remove orphaned git lock if it exists on boot
+  [ -f "$DIR/.git/index.lock" ] && rm -f $DIR/.git/index.lock
+
   # Check to see if there's a valid overlay-based update available. Conditions
   # are as follows:
   #
@@ -75,7 +80,6 @@ function launch {
   [ -d "/proc/irq/733" ] && echo 3 > /proc/irq/733/smp_affinity_list # USB for LeEco
   [ -d "/proc/irq/736" ] && echo 3 > /proc/irq/736/smp_affinity_list # USB for OP3T
 
-  DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null && pwd )"
 
   # Remove old NEOS update file
   # TODO: move this code to the updater

selfdrive/manager.py

@@ -389,6 +389,14 @@ def cleanup_all_processes(signal, frame):
     kill_managed_process(name)
   cloudlog.info("everything is dead")
 
+
+def send_managed_process_signal(name, sig):
+  if name not in running or name not in managed_processes:
+    return
+  cloudlog.info(f"sending signal {sig} to {name}")
+  os.kill(running[name].pid, sig)
+
+
 # ****************** run loop ******************
 
 def manager_init(should_register=True):
@@ -457,6 +465,7 @@ def manager_thread():
     for k in os.getenv("BLOCK").split(","):
       del managed_processes[k]
 
+  started_prev = False
   logger_dead = False
 
   while 1:
@@ -496,6 +505,12 @@ def manager_thread():
         else:
           kill_managed_process(p)
 
+      # trigger an update after going offroad
+      if started_prev:
+        send_managed_process_signal("updated", signal.SIGHUP)
+
+    started_prev = msg.thermal.started
+
     # check the status of all processes, did any of them die?
     running_list = ["%s%s\u001b[0m" % ("\u001b[32m" if running[p].is_alive() else "\u001b[31m", p) for p in running]
     cloudlog.debug(' '.join(running_list))

selfdrive/updated.py

@@ -263,7 +263,7 @@ def main():
 
     # Check for internet every 30s
     time_wrong = datetime.datetime.utcnow().year < 2019
-    ping_failed = subprocess.call(["ping", "-W", "4", "-c", "1", "8.8.8.8"])
+    ping_failed = os.system("git ls-remote --tags --quiet") != 0
     if ping_failed or time_wrong:
       wait_helper.sleep(30)
       continue