updated: move update alerts out of thermald (#23028)

* updated: move update alerts out of thermald

* cleanup last update time

* set on startup

selfdrive/thermald/thermald.py

@@ -22,15 +22,13 @@ from selfdrive.hardware import EON, TICI, PC, HARDWARE
 from selfdrive.loggerd.config import get_available_percent
 from selfdrive.swaglog import cloudlog
 from selfdrive.thermald.power_monitoring import PowerMonitoring
-from selfdrive.version import get_tested_branch, terms_version, training_version
+from selfdrive.version import terms_version, training_version
 
 ThermalStatus = log.DeviceState.ThermalStatus
 NetworkType = log.DeviceState.NetworkType
 NetworkStrength = log.DeviceState.NetworkStrength
 CURRENT_TAU = 15.   # 15s time constant
 TEMP_TAU = 5.   # 5s time constant
-DAYS_NO_CONNECTIVITY_MAX = 14     # do not allow to engage after this many days
-DAYS_NO_CONNECTIVITY_PROMPT = 10  # send an offroad prompt after this many days
 DISCONNECT_TIMEOUT = 5.  # wait 5 seconds before going offroad after disconnect so you get an alert
 
 ThermalBand = namedtuple("ThermalBand", ['min_temp', 'max_temp'])
@@ -324,47 +322,11 @@ def thermald_thread():
 
     # **** starting logic ****
 
-    # Check for last update time and display alerts if needed
+    # Ensure date/time are valid
     now = datetime.datetime.utcnow()
-
-    # show invalid date/time alert
     startup_conditions["time_valid"] = (now.year > 2020) or (now.year == 2020 and now.month >= 10)
     set_offroad_alert_if_changed("Offroad_InvalidTime", (not startup_conditions["time_valid"]))
 
-    # Show update prompt
-    try:
-      last_update = datetime.datetime.fromisoformat(params.get("LastUpdateTime", encoding='utf8'))
-    except (TypeError, ValueError):
-      last_update = now
-    dt = now - last_update
-
-    update_failed_count = params.get("UpdateFailedCount")
-    update_failed_count = 0 if update_failed_count is None else int(update_failed_count)
-    last_update_exception = params.get("LastUpdateException", encoding='utf8')
-
-    if update_failed_count > 15 and last_update_exception is not None:
-      if get_tested_branch():
-        extra_text = "Ensure the software is correctly installed"
-      else:
-        extra_text = last_update_exception
-
-      set_offroad_alert_if_changed("Offroad_ConnectivityNeeded", False)
-      set_offroad_alert_if_changed("Offroad_ConnectivityNeededPrompt", False)
-      set_offroad_alert_if_changed("Offroad_UpdateFailed", True, extra_text=extra_text)
-    elif dt.days > DAYS_NO_CONNECTIVITY_MAX and update_failed_count > 1:
-      set_offroad_alert_if_changed("Offroad_UpdateFailed", False)
-      set_offroad_alert_if_changed("Offroad_ConnectivityNeededPrompt", False)
-      set_offroad_alert_if_changed("Offroad_ConnectivityNeeded", True)
-    elif dt.days > DAYS_NO_CONNECTIVITY_PROMPT:
-      remaining = max(DAYS_NO_CONNECTIVITY_MAX - dt.days, 1)
-      set_offroad_alert_if_changed("Offroad_UpdateFailed", False)
-      set_offroad_alert_if_changed("Offroad_ConnectivityNeeded", False)
-      set_offroad_alert_if_changed("Offroad_ConnectivityNeededPrompt", True, extra_text=f"{remaining} day{'' if remaining == 1 else 's'}.")
-    else:
-      set_offroad_alert_if_changed("Offroad_UpdateFailed", False)
-      set_offroad_alert_if_changed("Offroad_ConnectivityNeeded", False)
-      set_offroad_alert_if_changed("Offroad_ConnectivityNeededPrompt", False)
-
     startup_conditions["up_to_date"] = params.get("Offroad_ConnectivityNeeded") is None or params.get_bool("DisableUpdates") or params.get_bool("SnoozeUpdate")
     startup_conditions["not_uninstalling"] = not params.get_bool("DoUninstall")
     startup_conditions["accepted_terms"] = params.get("HasAcceptedTerms") == terms_version

selfdrive/updated.py

@@ -40,6 +40,7 @@ from common.params import Params
 from selfdrive.hardware import EON, TICI, HARDWARE
 from selfdrive.swaglog import cloudlog
 from selfdrive.controls.lib.alertmanager import set_offroad_alert
+from selfdrive.version import get_tested_branch
 
 LOCK_FILE = os.getenv("UPDATER_LOCK_FILE", "/tmp/safe_staging_overlay.lock")
 STAGING_ROOT = os.getenv("UPDATER_STAGING_ROOT", "/data/safe_staging")
@@ -51,6 +52,8 @@ OVERLAY_METADATA = os.path.join(STAGING_ROOT, "metadata")
 OVERLAY_MERGED = os.path.join(STAGING_ROOT, "merged")
 FINALIZED = os.path.join(STAGING_ROOT, "finalized")
 
+DAYS_NO_CONNECTIVITY_MAX = 14     # do not allow to engage after this many days
+DAYS_NO_CONNECTIVITY_PROMPT = 10  # send an offroad prompt after this many days
 
 class WaitTimeHelper:
   def __init__(self, proc):
@@ -102,15 +105,24 @@ def set_params(new_version: bool, failed_count: int, exception: Optional[str]) -
   params = Params()
 
   params.put("UpdateFailedCount", str(failed_count))
+
+  last_update = datetime.datetime.utcnow()
   if failed_count == 0:
-    t = datetime.datetime.utcnow().isoformat()
+    t = last_update.isoformat()
     params.put("LastUpdateTime", t.encode('utf8'))
+  else:
+    try:
+      t = params.get("LastUpdateTime", encoding='utf8')
+      last_update = datetime.datetime.fromisoformat(t)
+    except (TypeError, ValueError):
+      pass
 
   if exception is None:
     params.delete("LastUpdateException")
   else:
     params.put("LastUpdateException", exception)
 
+  # Write out release notes for new versions
   if new_version:
     try:
       with open(os.path.join(FINALIZED, "RELEASES.md"), "rb") as f:
@@ -123,6 +135,24 @@ def set_params(new_version: bool, failed_count: int, exception: Optional[str]) -
       params.put("ReleaseNotes", "")
     params.put_bool("UpdateAvailable", True)
 
+  # Handle user prompt
+  for alert in ("Offroad_UpdateFailed", "Offroad_ConnectivityNeeded", "Offroad_ConnectivityNeededPrompt"):
+    set_offroad_alert(alert, False)
+
+  now = datetime.datetime.utcnow()
+  dt = now - last_update
+  if failed_count > 15 and exception is not None:
+    if get_tested_branch():
+      extra_text = "Ensure the software is correctly installed"
+    else:
+      extra_text = exception
+    set_offroad_alert("Offroad_UpdateFailed", True, extra_text=extra_text)
+  elif dt.days > DAYS_NO_CONNECTIVITY_MAX and failed_count > 1:
+    set_offroad_alert("Offroad_ConnectivityNeeded", True)
+  elif dt.days > DAYS_NO_CONNECTIVITY_PROMPT:
+    remaining = max(DAYS_NO_CONNECTIVITY_MAX - dt.days, 1)
+    set_offroad_alert("Offroad_ConnectivityNeededPrompt", True, extra_text=f"{remaining} day{'' if remaining == 1 else 's'}.")
+
 
 def setup_git_options(cwd: str) -> None:
   # We sync FS object atimes (which NEOS doesn't use) and mtimes, but ctimes
@@ -344,16 +374,14 @@ def main():
   params = Params()
 
   if params.get_bool("DisableUpdates"):
-    raise RuntimeError("updates are disabled by the DisableUpdates param")
+    cloudlog.warning("updates are disabled by the DisableUpdates param")
-
+    exit(0)
-  if EON and os.geteuid() != 0:
-    raise RuntimeError("updated must be launched as root!")
 
   ov_lock_fd = open(LOCK_FILE, 'w')
   try:
     fcntl.flock(ov_lock_fd, fcntl.LOCK_EX | fcntl.LOCK_NB)
   except IOError as e:
-    raise RuntimeError("couldn't get overlay lock; is another updated running?") from e
+    raise RuntimeError("couldn't get overlay lock; is another instance running?") from e
 
   # Set low io priority
   proc = psutil.Process()
@@ -368,10 +396,6 @@ def main():
     t = datetime.datetime.utcnow().isoformat()
     params.put("InstallDate", t.encode('utf8'))
 
-  # Wait for IsOffroad to be set before our first update attempt
-  wait_helper = WaitTimeHelper(proc)
-  wait_helper.sleep(30)
-
   overlay_init = Path(os.path.join(BASEDIR, ".overlay_init"))
   overlay_init.unlink(missing_ok=True)
 
@@ -379,6 +403,13 @@ def main():
   last_fetch_time = 0
   update_failed_count = 0
 
+  # Set initial params for offroad alerts
+  set_params(False, 0, None)
+
+  # Wait for IsOffroad to be set before our first update attempt
+  wait_helper = WaitTimeHelper(proc)
+  wait_helper.sleep(30)
+
   # Run the update loop
   #  * every 1m, do a lightweight internet/update check
   #  * every 10m, do a full git fetch