manager cleanup (#2634)

* start cleanup

* add uninstall to hw abstraction layer

* cleanup

* litte more

* don't build at import time

* fix tests

* don't build by default

* sync

Jenkinsfile

@@ -124,7 +124,7 @@ pipeline {
                 stage('Replay Tests') {
                   steps {
                     phone_steps("eon2", [
-                      ["camerad/modeld replay", "cd selfdrive/test/process_replay && ./camera_replay.py"],
+                      ["camerad/modeld replay", "QCOM_REPLAY=1 scons -j4 && cd selfdrive/test/process_replay && ./camera_replay.py"],
                     ])
                   }
                 }
@@ -132,7 +132,7 @@ pipeline {
                 stage('HW + Unit Tests') {
                   steps {
                     phone_steps("eon", [
-                      ["build cereal", "SCONS_CACHE=1 scons -j4 cereal/"],
+                      ["build", "SCONS_CACHE=1 scons -j4"],
                       ["test sounds", "nosetests -s selfdrive/test/test_sounds.py"],
                       ["test boardd loopback", "nosetests -s selfdrive/boardd/tests/test_boardd_loopback.py"],
                       ["test loggerd", "CI=1 python selfdrive/loggerd/tests/test_loggerd.py"],

selfdrive/hardware/base.py

@@ -16,6 +16,14 @@ class HardwareBase:
     except Exception:
       return default
 
+  @abstractmethod
+  def reboot(self, reason=None):
+    pass
+
+  @abstractmethod
+  def uninstall(self):
+    pass
+
   @abstractmethod
   def get_sound_card_online(self):
     pass
@@ -32,10 +40,6 @@ class HardwareBase:
   def get_subscriber_info(self):
     pass
 
-  @abstractmethod
-  def reboot(self, reason=None):
-    pass
-
   @abstractmethod
   def get_network_type(self):
     pass

selfdrive/hardware/eon/hardware.py

@@ -98,6 +98,12 @@ class Android(HardwareBase):
       "i32", "1"  # wait
     ])
 
+  def uninstall(self):
+    with open('/cache/recovery/command', 'w') as f:
+      f.write('--wipe_data\n')
+    # IPowerManager.reboot(confirm=false, reason="recovery", wait=true)
+    self.reboot(reason="recovery")
+
   def get_sim_info(self):
     # Used for athena
     # TODO: build using methods from this class

selfdrive/hardware/pc/hardware.py

@@ -11,6 +11,12 @@ class Pc(HardwareBase):
   def get_sound_card_online(self):
     return True
 
+  def reboot(self, reason=None):
+    print("REBOOT!")
+
+  def uninstall(self):
+    print("uninstall")
+
   def get_imei(self, slot):
     return "%015d" % random.randint(0, 1 << 32)
 
@@ -20,9 +26,6 @@ class Pc(HardwareBase):
   def get_subscriber_info(self):
     return ""
 
-  def reboot(self, reason=None):
-    print("REBOOT!")
-
   def get_network_type(self):
     return NetworkType.wifi
 

selfdrive/hardware/tici/hardware.py

@@ -35,12 +35,16 @@ class Tici(HardwareBase):
   def get_sound_card_online(self):
     return True
 
-  def get_serial(self):
-    return self.get_cmdline()['androidboot.serialno']
-
   def reboot(self, reason=None):
     subprocess.check_output(["sudo", "reboot"])
 
+  def uninstall(self):
+    # TODO: implement uninstall. reboot to factory reset?
+    pass
+
+  def get_serial(self):
+    return self.get_cmdline()['androidboot.serialno']
+
   def get_network_type(self):
     primary_connection = self.nm.Get(NM, 'PrimaryConnection', dbus_interface=DBUS_PROPS)
     primary_connection = self.bus.get_object(NM, primary_connection)

selfdrive/manager.py

@@ -1,37 +1,33 @@
 #!/usr/bin/env python3
+import datetime
+import importlib
 import os
-import time
 import sys
 import fcntl
 import errno
 import signal
 import shutil
 import subprocess
-import datetime
 import textwrap
-from typing import Dict, List
+import time
-from selfdrive.swaglog import cloudlog, add_logentries_handler
+import traceback
 
+from multiprocessing import Process
+from typing import Dict, List
 
 from common.basedir import BASEDIR
+from common.spinner import Spinner
+from common.text_window import TextWindow
 from selfdrive.hardware import HARDWARE, EON, PC
-WEBCAM = os.getenv("WEBCAM") is not None
+from selfdrive.swaglog import cloudlog, add_logentries_handler
-sys.path.append(os.path.join(BASEDIR, "pyextra"))
+
 os.environ['BASEDIR'] = BASEDIR
+sys.path.append(os.path.join(BASEDIR, "pyextra"))
 
 TOTAL_SCONS_NODES = 1040
-prebuilt = os.path.exists(os.path.join(BASEDIR, 'prebuilt'))
+WEBCAM = os.getenv("WEBCAM") is not None
-
+PREBUILT = os.path.exists(os.path.join(BASEDIR, 'prebuilt'))
-# Create folders needed for msgq
-try:
-  os.mkdir("/dev/shm")
-except FileExistsError:
-  pass
-except PermissionError:
-  print("WARNING: failed to make /dev/shm")
 
-if EON:
-  os.chmod("/dev/shm", 0o777)
 
 def unblock_stdout():
   # get a non-blocking stdout
@@ -42,8 +38,7 @@ def unblock_stdout():
     signal.signal(signal.SIGINT, lambda signum, frame: os.kill(child_pid, signal.SIGINT))
     signal.signal(signal.SIGTERM, lambda signum, frame: os.kill(child_pid, signal.SIGTERM))
 
-    fcntl.fcntl(sys.stdout, fcntl.F_SETFL,
+    fcntl.fcntl(sys.stdout, fcntl.F_SETFL, fcntl.fcntl(sys.stdout, fcntl.F_GETFL) | os.O_NONBLOCK)
-       fcntl.fcntl(sys.stdout, fcntl.F_GETFL) | os.O_NONBLOCK)
 
     while True:
       try:
@@ -70,12 +65,6 @@ def unblock_stdout():
 if __name__ == "__main__":
   unblock_stdout()
 
-from common.spinner import Spinner
-from common.text_window import TextWindow
-
-import importlib
-import traceback
-from multiprocessing import Process
 
 # Run scons
 spinner = Spinner()
@@ -83,7 +72,7 @@ spinner.update("0")
 if __name__ != "__main__":
   spinner.close()
 
-if not prebuilt:
+def build():
   for retry in [True, False]:
     # run scons
     env = os.environ.copy()
@@ -91,7 +80,7 @@ if not prebuilt:
     env['SCONS_CACHE'] = "1"
 
     nproc = os.cpu_count()
-    j_flag = "" if nproc is None else "-j%d" % (nproc - 1)
+    j_flag = "" if nproc is None else f"-j{nproc - 1}"
     scons = subprocess.Popen(["scons", j_flag], cwd=BASEDIR, env=env, stderr=subprocess.PIPE)
 
     compile_output = []
@@ -148,7 +137,9 @@ if not prebuilt:
     else:
       break
 
-import cereal
+if __name__ == "__main__" and not PREBUILT:
+  build()
+
 import cereal.messaging as messaging
 
 from common.params import Params
@@ -159,7 +150,6 @@ from selfdrive.loggerd.config import ROOT
 from selfdrive.launcher import launcher
 from selfdrive.hardware.eon.apk import update_apks, pm_apply_packages, start_offroad
 
-ThermalStatus = cereal.log.ThermalData.ThermalStatus
 
 # comment out anything you don't want to run
 managed_processes = {
@@ -245,13 +235,7 @@ driver_view_processes = [
   'dmonitoringmodeld'
 ]
 
-if WEBCAM:
+if not PC or WEBCAM:
-  car_started_processes += [
-    'dmonitoringd',
-    'dmonitoringmodeld',
-  ]
-
-if not PC:
   car_started_processes += [
     'ubloxd',
     'dmonitoringd',
@@ -267,7 +251,6 @@ if EON:
 
 def register_managed_process(name, desc, car_started=False):
   global managed_processes, car_started_processes, persistent_processes
-  print("registering %s" % name)
   managed_processes[name] = desc
   if car_started:
     car_started_processes.append(name)
@@ -323,22 +306,22 @@ def start_daemon_process(name):
 
   params.put(pid_param, str(proc.pid))
 
-def prepare_managed_process(p):
+def prepare_managed_process(p, build=False):
   proc = managed_processes[p]
   if isinstance(proc, str):
     # import this python
     cloudlog.info("preimporting %s" % proc)
     importlib.import_module(proc)
-  elif os.path.isfile(os.path.join(BASEDIR, proc[0], "Makefile")):
+  elif os.path.isfile(os.path.join(BASEDIR, proc[0], "SConscript")) and build:
     # build this process
     cloudlog.info("building %s" % (proc,))
     try:
-      subprocess.check_call(["make", "-j4"], cwd=os.path.join(BASEDIR, proc[0]))
+      subprocess.check_call(["scons", "u", "-j4", "."], cwd=os.path.join(BASEDIR, proc[0]))
     except subprocess.CalledProcessError:
-      # make clean if the build failed
+      # clean and retry if the build failed
-      cloudlog.warning("building %s failed, make clean" % (proc, ))
+      cloudlog.warning("building %s failed, cleaning and retrying" % (proc, ))
-      subprocess.check_call(["make", "clean"], cwd=os.path.join(BASEDIR, proc[0]))
+      subprocess.check_call(["scons", "-u", "-c", "."], cwd=os.path.join(BASEDIR, proc[0]))
-      subprocess.check_call(["make", "-j4"], cwd=os.path.join(BASEDIR, proc[0]))
+      subprocess.check_call(["scons", "-u", "-j4", "."], cwd=os.path.join(BASEDIR, proc[0]))
 
 
 def join_process(process, timeout):
@@ -404,21 +387,23 @@ def send_managed_process_signal(name, sig):
 
 # ****************** run loop ******************
 
-def manager_init(should_register=True):
+def manager_init():
-  if should_register:
+  # Create folders needed for msgq
+  try:
+    os.mkdir("/dev/shm")
+  except FileExistsError:
+    pass
+  except PermissionError:
+    print("WARNING: failed to make /dev/shm")
+
+  # set dongle id
   reg_res = register()
   if reg_res:
     dongle_id = reg_res
   else:
     raise Exception("server registration failed")
-  else:
-    dongle_id = "c"*16
-
-  # set dongle id
-  cloudlog.info("dongle id is " + dongle_id)
   os.environ['DONGLE_ID'] = dongle_id
 
-  cloudlog.info("dirty is %d" % dirty)
   if not dirty:
     os.environ['CLEAN'] = '1'
 
@@ -435,12 +420,11 @@ def manager_init(should_register=True):
   # ensure shared libraries are readable by apks
   if EON:
     os.chmod(BASEDIR, 0o755)
+    os.chmod("/dev/shm", 0o777)
     os.chmod(os.path.join(BASEDIR, "cereal"), 0o755)
     os.chmod(os.path.join(BASEDIR, "cereal", "libmessaging_shared.so"), 0o755)
 
 def manager_thread():
-  # now loop
-  thermal_sock = messaging.sub_sock('thermal')
 
   cloudlog.info("manager start")
   cloudlog.info({"environ": os.environ})
@@ -448,8 +432,6 @@ def manager_thread():
   # save boot log
   subprocess.call(["./loggerd", "--bootlog"], cwd=os.path.join(BASEDIR, "selfdrive/loggerd"))
 
-  params = Params()
-
   # start daemon processes
   for p in daemon_processes:
     start_daemon_process(p)
@@ -472,6 +454,8 @@ def manager_thread():
 
   started_prev = False
   logger_dead = False
+  params = Params()
+  thermal_sock = messaging.sub_sock('thermal')
 
   while 1:
     msg = messaging.recv_sock(thermal_sock, wait=True)
@@ -502,6 +486,7 @@ def manager_thread():
 
       # trigger an update after going offroad
       if started_prev:
+        os.sync()
         send_managed_process_signal("updated", signal.SIGHUP)
 
     started_prev = msg.thermal.started
@@ -514,25 +499,18 @@ def manager_thread():
     if params.get("DoUninstall", encoding='utf8') == "1":
       break
 
-def manager_prepare(spinner=None):
+def manager_prepare():
   # build all processes
   os.chdir(os.path.dirname(os.path.abspath(__file__)))
 
   # Spinner has to start from 70 here
-  total = 100.0 if prebuilt else 30.0
+  total = 100.0 if PREBUILT else 30.0
 
   for i, p in enumerate(managed_processes):
-    if spinner is not None:
+    perc = (100.0 - total) + total * (i + 1) / len(managed_processes)
-      spinner.update("%d" % ((100.0 - total) + total * (i + 1) / len(managed_processes),))
+    spinner.update(str(int(perc)))
     prepare_managed_process(p)
 
-def uninstall():
-  cloudlog.warning("uninstalling")
-  with open('/cache/recovery/command', 'w') as f:
-    f.write('--wipe_data\n')
-  # IPowerManager.reboot(confirm=false, reason="recovery", wait=true)
-  HARDWARE.reboot(reason="recovery")
-
 def main():
   params = Params()
   params.manager_start()
@@ -558,7 +536,7 @@ def main():
     if params.get(k) is None:
       params.put(k, v)
 
-  # is this chffrplus?
+  # is this dashcam?
   if os.getenv("PASSIVE") is not None:
     params.put("Passive", str(int(os.getenv("PASSIVE"))))
 
@@ -568,7 +546,7 @@ def main():
   if EON:
     update_apks()
   manager_init()
-  manager_prepare(spinner)
+  manager_prepare()
   spinner.close()
 
   if os.getenv("PREPAREONLY") is not None:
@@ -586,7 +564,8 @@ def main():
     cleanup_all_processes(None, None)
 
   if params.get("DoUninstall", encoding='utf8') == "1":
-    uninstall()
+    cloudlog.warning("uninstalling")
+    HARDWARE.uninstall()
 
 
 if __name__ == "__main__":
@@ -598,7 +577,7 @@ if __name__ == "__main__":
 
     # Show last 3 lines of traceback
     error = traceback.format_exc(-3)
-    error = "Manager failed to start\n \n" + error
+    error = "Manager failed to start\n\n" + error
     spinner.close()
     with TextWindow(error) as t:
       t.wait_for_exit()