openpilot/selfdrive/manager.py

#!/usr/bin/env python3
import datetime
import importlib
import os
import sys
import fcntl
import errno
import signal
import shutil
import subprocess
import textwrap
import time
import traceback

from multiprocessing import Process
from typing import Dict

from common.basedir import BASEDIR
from common.spinner import Spinner
from common.text_window import TextWindow
import selfdrive.crash as crash
from selfdrive.hardware import HARDWARE, EON, PC
from selfdrive.hardware.eon.apk import update_apks, pm_apply_packages, start_offroad
from selfdrive.swaglog import cloudlog, add_logentries_handler
from selfdrive.version import version, dirty

os.environ['BASEDIR'] = BASEDIR
sys.path.append(os.path.join(BASEDIR, "pyextra"))

TOTAL_SCONS_NODES = 1040
MAX_BUILD_PROGRESS = 70
WEBCAM = os.getenv("WEBCAM") is not None
PREBUILT = os.path.exists(os.path.join(BASEDIR, 'prebuilt'))


def unblock_stdout():
  # get a non-blocking stdout
  child_pid, child_pty = os.forkpty()
  if child_pid != 0:  # parent

    # child is in its own process group, manually pass kill signals
    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_GETFL) | os.O_NONBLOCK)

    while True:
      try:
        dat = os.read(child_pty, 4096)
      except OSError as e:
        if e.errno == errno.EIO:
          break
        continue

      if not dat:
        break

      try:
        sys.stdout.write(dat.decode('utf8'))
      except (OSError, IOError, UnicodeDecodeError):
        pass

    # os.wait() returns a tuple with the pid and a 16 bit value
    # whose low byte is the signal number and whose high byte is the exit satus
    exit_status = os.wait()[1] >> 8
    os._exit(exit_status)

if __name__ == "__main__":
  unblock_stdout()


# Start spinner
spinner = Spinner()
spinner.update_progress(0, 100)
if __name__ != "__main__":
  spinner.close()

def build():
  env = os.environ.copy()
  env['SCONS_PROGRESS'] = "1"
  env['SCONS_CACHE'] = "1"
  nproc = os.cpu_count()
  j_flag = "" if nproc is None else f"-j{nproc - 1}"

  for retry in [True, False]:
    scons = subprocess.Popen(["scons", j_flag], cwd=BASEDIR, env=env, stderr=subprocess.PIPE)

    compile_output = []

    # Read progress from stderr and update spinner
    while scons.poll() is None:
      try:
        line = scons.stderr.readline()
        if line is None:
          continue
        line = line.rstrip()

        prefix = b'progress: '
        if line.startswith(prefix):
          i = int(line[len(prefix):])
          spinner.update_progress(MAX_BUILD_PROGRESS * min(1., i / TOTAL_SCONS_NODES), 100.)
        elif len(line):
          compile_output.append(line)
          print(line.decode('utf8', 'replace'))
      except Exception:
        pass

    if scons.returncode != 0:
      # Read remaining output
      r = scons.stderr.read().split(b'\n')
      compile_output += r

      if retry:
        if not os.getenv("CI"):
          print("scons build failed, cleaning in")
          for i in range(3, -1, -1):
            print("....%d" % i)
            time.sleep(1)
          subprocess.check_call(["scons", "-c"], cwd=BASEDIR, env=env)
          shutil.rmtree("/tmp/scons_cache", ignore_errors=True)
          shutil.rmtree("/data/scons_cache", ignore_errors=True)
        else:
          print("scons build failed after retry")
          sys.exit(1)
      else:
        # Build failed log errors
        errors = [line.decode('utf8', 'replace') for line in compile_output
                  if any([err in line for err in [b'error: ', b'not found, needed by target']])]
        error_s = "\n".join(errors)
        add_logentries_handler(cloudlog)
        cloudlog.error("scons build failed\n" + error_s)

        # Show TextWindow
        spinner.close()
        error_s = "\n \n".join(["\n".join(textwrap.wrap(e, 65)) for e in errors])
        with TextWindow("openpilot failed to build\n \n" + error_s) as t:
          t.wait_for_exit()
        exit(1)
    else:
      break

if __name__ == "__main__" and not PREBUILT:
  build()

import cereal.messaging as messaging

from common.params import Params
from selfdrive.registration import register
from selfdrive.loggerd.config import ROOT
from selfdrive.launcher import launcher


# comment out anything you don't want to run
managed_processes = {
  "thermald": "selfdrive.thermald.thermald",
  "uploader": "selfdrive.loggerd.uploader",
  "deleter": "selfdrive.loggerd.deleter",
  "controlsd": "selfdrive.controls.controlsd",
  "plannerd": "selfdrive.controls.plannerd",
  "radard": "selfdrive.controls.radard",
  "dmonitoringd": "selfdrive.monitoring.dmonitoringd",
  "ubloxd": ("selfdrive/locationd", ["./ubloxd"]),
  "loggerd": ("selfdrive/loggerd", ["./loggerd"]),
  "logmessaged": "selfdrive.logmessaged",
  "locationd": "selfdrive.locationd.locationd",
  "tombstoned": "selfdrive.tombstoned",
  "logcatd": ("selfdrive/logcatd", ["./logcatd"]),
  "proclogd": ("selfdrive/proclogd", ["./proclogd"]),
  "boardd": ("selfdrive/boardd", ["./boardd"]),   # not used directly
  "pandad": "selfdrive.pandad",
  "ui": ("selfdrive/ui", ["./ui"]),
  "calibrationd": "selfdrive.locationd.calibrationd",
  "paramsd": "selfdrive.locationd.paramsd",
  "camerad": ("selfdrive/camerad", ["./camerad"]),
  "sensord": ("selfdrive/sensord", ["./sensord"]),
  "clocksd": ("selfdrive/clocksd", ["./clocksd"]),
  "gpsd": ("selfdrive/sensord", ["./gpsd"]),
  "updated": "selfdrive.updated",
  "dmonitoringmodeld": ("selfdrive/modeld", ["./dmonitoringmodeld"]),
  "modeld": ("selfdrive/modeld", ["./modeld"]),
  "rtshield": "selfdrive.rtshield",
}

daemon_processes = {
  "manage_athenad": ("selfdrive.athena.manage_athenad", "AthenadPid"),
}

running: Dict[str, Process] = {}
def get_running():
  return running

# due to qualcomm kernel bugs SIGKILLing camerad sometimes causes page table corruption
unkillable_processes = ['camerad']

# processes to end with SIGKILL instead of SIGTERM
kill_processes = []
if EON:
  kill_processes += [
    'sensord',
  ]

persistent_processes = [
  'pandad',
  'thermald',
  'logmessaged',
  'ui',
  'uploader',
  'deleter',
]

if not PC:
  persistent_processes += [
    'updated',
    'logcatd',
    'tombstoned',
  ]

if EON:
  persistent_processes += [
    'sensord',
  ]

car_started_processes = [
  'controlsd',
  'plannerd',
  'loggerd',
  'radard',
  'calibrationd',
  'paramsd',
  'camerad',
  'modeld',
  'proclogd',
  'locationd',
  'clocksd',
]

driver_view_processes = [
  'camerad',
  'dmonitoringd',
  'dmonitoringmodeld'
]

if not PC or WEBCAM:
  car_started_processes += [
    'ubloxd',
    'dmonitoringd',
    'dmonitoringmodeld',
  ]

if EON:
  car_started_processes += [
    'gpsd',
    'rtshield',
  ]
else:
   car_started_processes += [
    'sensord',
  ]

def register_managed_process(name, desc, car_started=False):
  global managed_processes, car_started_processes, persistent_processes
  managed_processes[name] = desc
  if car_started:
    car_started_processes.append(name)
  else:
    persistent_processes.append(name)

# ****************** process management functions ******************
def nativelauncher(pargs, cwd):
  # exec the process
  os.chdir(cwd)

  # because when extracted from pex zips permissions get lost -_-
  os.chmod(pargs[0], 0o700)

  os.execvp(pargs[0], pargs)

def start_managed_process(name):
  if name in running or name not in managed_processes:
    return
  proc = managed_processes[name]
  if isinstance(proc, str):
    cloudlog.info("starting python %s" % proc)
    running[name] = Process(name=name, target=launcher, args=(proc,))
  else:
    pdir, pargs = proc
    cwd = os.path.join(BASEDIR, pdir)
    cloudlog.info("starting process %s" % name)
    running[name] = Process(name=name, target=nativelauncher, args=(pargs, cwd))
  running[name].start()

def start_daemon_process(name):
  params = Params()
  proc, pid_param = daemon_processes[name]
  pid = params.get(pid_param, encoding='utf-8')

  if pid is not None:
    try:
      os.kill(int(pid), 0)
      with open(f'/proc/{pid}/cmdline') as f:
        if proc in f.read():
          # daemon is running
          return
    except (OSError, FileNotFoundError):
      # process is dead
      pass

  cloudlog.info("starting daemon %s" % name)
  proc = subprocess.Popen(['python', '-m', proc],  # pylint: disable=subprocess-popen-preexec-fn
                          stdin=open('/dev/null', 'r'),
                          stdout=open('/dev/null', 'w'),
                          stderr=open('/dev/null', 'w'),
                          preexec_fn=os.setpgrp)

  params.put(pid_param, str(proc.pid))

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], "SConscript")) and build:
    # build this process
    cloudlog.info("building %s" % (proc,))
    try:
      subprocess.check_call(["scons", "u", "-j4", "."], cwd=os.path.join(BASEDIR, proc[0]))
    except subprocess.CalledProcessError:
      # clean and retry if the build failed
      cloudlog.warning("building %s failed, cleaning and retrying" % (proc, ))
      subprocess.check_call(["scons", "-u", "-c", "."], 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):
  # Process().join(timeout) will hang due to a python 3 bug: https://bugs.python.org/issue28382
  # We have to poll the exitcode instead
  t = time.time()
  while time.time() - t < timeout and process.exitcode is None:
    time.sleep(0.001)


def kill_managed_process(name, retry=True):
  if name not in running or name not in managed_processes:
    return
  cloudlog.info(f"killing {name}")

  if running[name].exitcode is None:
    sig = signal.SIGKILL if name in kill_processes else signal.SIGINT
    os.kill(running[name].pid, sig)

    join_process(running[name], 5)

    if running[name].exitcode is None:
      if not retry:
        raise Exception(f"{name} failed to die")

      if name in unkillable_processes:
        cloudlog.critical("unkillable process %s failed to exit! rebooting in 15 if it doesn't die" % name)
        join_process(running[name], 15)
        if running[name].exitcode is None:
          cloudlog.critical("unkillable process %s failed to die!" % name)
          os.system("date >> /data/unkillable_reboot")
          os.sync()
          HARDWARE.reboot()
          raise RuntimeError
      else:
        cloudlog.info("killing %s with SIGKILL" % name)
        os.kill(running[name].pid, signal.SIGKILL)
        running[name].join()

  ret = running[name].exitcode
  cloudlog.info(f"{name} is dead with {ret}")
  del running[name]
  return ret


def cleanup_all_processes(signal, frame):
  cloudlog.info("caught ctrl-c %s %s" % (signal, frame))

  if EON:
    pm_apply_packages('disable')

  for name in list(running.keys()):
    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 or \
     running[name].exitcode is not None:
    return

  cloudlog.info(f"sending signal {sig} to {name}")
  os.kill(running[name].pid, sig)


# ****************** run loop ******************

def manager_init():
  # 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(spinner)
  if reg_res:
    dongle_id = reg_res
  else:
    raise Exception("server registration failed")
  os.environ['DONGLE_ID'] = dongle_id

  if not dirty:
    os.environ['CLEAN'] = '1'

  cloudlog.bind_global(dongle_id=dongle_id, version=version, dirty=dirty, is_eon=True)
  crash.bind_user(id=dongle_id)
  crash.bind_extra(version=version, dirty=dirty, is_eon=True)

  os.umask(0)
  try:
    os.mkdir(ROOT, 0o777)
  except OSError:
    pass

  # 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():

  cloudlog.info("manager start")
  cloudlog.info({"environ": os.environ})

  # save boot log
  subprocess.call(["./loggerd", "--bootlog"], cwd=os.path.join(BASEDIR, "selfdrive/loggerd"))

  # start daemon processes
  for p in daemon_processes:
    start_daemon_process(p)

  # start persistent processes
  for p in persistent_processes:
    start_managed_process(p)

  # start offroad
  if EON:
    pm_apply_packages('enable')
    start_offroad()

  if os.getenv("NOBOARD") is not None:
    del managed_processes["pandad"]

  if os.getenv("BLOCK") is not None:
    for k in os.getenv("BLOCK").split(","):
      del managed_processes[k]

  started_prev = False
  logger_dead = False
  params = Params()
  thermal_sock = messaging.sub_sock('thermal')

  while 1:
    msg = messaging.recv_sock(thermal_sock, wait=True)

    if msg.thermal.freeSpace < 0.05:
      logger_dead = True

    if msg.thermal.started:
      for p in car_started_processes:
        if p == "loggerd" and logger_dead:
          kill_managed_process(p)
        else:
          start_managed_process(p)
    else:
      logger_dead = False
      driver_view = params.get("IsDriverViewEnabled") == b"1"

      # TODO: refactor how manager manages processes
      for p in reversed(car_started_processes):
        if p not in driver_view_processes or not driver_view:
          kill_managed_process(p)

      for p in driver_view_processes:
        if driver_view:
          start_managed_process(p)
        else:
          kill_managed_process(p)

      # trigger an update after going offroad
      if started_prev:
        os.sync()
        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))

    # Exit main loop when uninstall is needed
    if params.get("DoUninstall", encoding='utf8') == "1":
      break

def manager_prepare():
  # build all processes
  os.chdir(os.path.dirname(os.path.abspath(__file__)))

  total = 100.0 - (0 if PREBUILT else MAX_BUILD_PROGRESS)

  for i, p in enumerate(managed_processes):
    perc = (100.0 - total) + total * (i + 1) / len(managed_processes)
    spinner.update_progress(perc, 100.)
    prepare_managed_process(p)

def main():
  params = Params()
  params.manager_start()

  default_params = [
    ("CommunityFeaturesToggle", "0"),
    ("CompletedTrainingVersion", "0"),
    ("IsRHD", "0"),
    ("IsMetric", "0"),
    ("RecordFront", "0"),
    ("HasAcceptedTerms", "0"),
    ("HasCompletedSetup", "0"),
    ("IsUploadRawEnabled", "1"),
    ("IsLdwEnabled", "1"),
    ("LastUpdateTime", datetime.datetime.utcnow().isoformat().encode('utf8')),
    ("OpenpilotEnabledToggle", "1"),
    ("LaneChangeEnabled", "1"),
    ("IsDriverViewEnabled", "0"),
  ]

  # set unset params
  for k, v in default_params:
    if params.get(k) is None:
      params.put(k, v)

  # is this dashcam?
  if os.getenv("PASSIVE") is not None:
    params.put("Passive", str(int(os.getenv("PASSIVE"))))

  if params.get("Passive") is None:
    raise Exception("Passive must be set to continue")

  if EON:
    update_apks()
  manager_init()
  manager_prepare()
  spinner.close()

  if os.getenv("PREPAREONLY") is not None:
    return

  # SystemExit on sigterm
  signal.signal(signal.SIGTERM, lambda signum, frame: sys.exit(1))

  try:
    manager_thread()
  except Exception:
    traceback.print_exc()
    crash.capture_exception()
  finally:
    cleanup_all_processes(None, None)

  if params.get("DoUninstall", encoding='utf8') == "1":
    cloudlog.warning("uninstalling")
    HARDWARE.uninstall()


if __name__ == "__main__":
  try:
    main()
  except Exception:
    add_logentries_handler(cloudlog)
    cloudlog.exception("Manager failed to start")

    # Show last 3 lines of traceback
    error = traceback.format_exc(-3)
    error = "Manager failed to start\n\n" + error
    spinner.close()
    with TextWindow(error) as t:
      t.wait_for_exit()

    raise

  # manual exit because we are forked
  sys.exit(0)