kachow

launch_chffrplus.sh

@@ -6,6 +6,7 @@ source "$DIR/launch_env.sh"
 
 export SP_BOOT_TIMING_LOG="${SP_BOOT_TIMING_LOG:-/tmp/starpilot_boot_timing.log}"
 : > "$SP_BOOT_TIMING_LOG" 2>/dev/null || true
+PREBUILT_COMPAT_CACHE="${SP_PREBUILT_COMPAT_CACHE:-/cache/starpilot_prebuilt_runtime_compatible_v1}"
 SP_LAUNCH_LAST_SECONDS=$SECONDS
 
 function sp_boot_timing_line {
@@ -20,6 +21,54 @@ function sp_launch_timing {
   SP_LAUNCH_LAST_SECONDS=$now
 }
 
+function launch_param_migrations_needed {
+  local marker_dir="/data/params/.starpilot_param_migrations/${OPENPILOT_PREFIX:-d}"
+  [ ! -f "$marker_dir/.starpilot_launch_param_migrations_v2" ] && return 0
+  [ ! -f "$marker_dir/.starpilot_branch_defaults_migrations_v1" ] && return 0
+  [ ! -f "$marker_dir/.starpilot_acceleration_profile_default_v1" ] && return 0
+  return 1
+}
+
+function prebuilt_runtime_artifacts {
+  printf '%s\n' \
+    "$DIR/common/params_pyx.so" \
+    "$DIR/msgq/ipc_pyx.so" \
+    "$DIR/msgq/visionipc/visionipc_pyx.so" \
+    "$DIR/common/transformations/transformations.so" \
+    "$DIR/selfdrive/modeld/models/driving_tinygrad.pkl" \
+    "$DIR/selfdrive/modeld/models/dmonitoring_model_metadata.pkl" \
+    "$DIR/selfdrive/modeld/models/dmonitoring_model_tinygrad.pkl" \
+    "$DIR/selfdrive/modeld/models/dm_warp_1928x1208_tinygrad.pkl" \
+    "$DIR/selfdrive/modeld/models/dm_warp_1344x760_tinygrad.pkl" \
+    "$DIR/selfdrive/pandad/pandad_api_impl.so" \
+    "$DIR/selfdrive/controls/lib/lateral_mpc_lib/c_generated_code/acados_ocp_solver_pyx.so" \
+    "$DIR/selfdrive/controls/lib/lateral_mpc_lib/c_generated_code/libacados_ocp_solver_lat.so" \
+    "$DIR/selfdrive/controls/lib/longitudinal_mpc_lib/c_generated_code/acados_ocp_solver_pyx.so" \
+    "$DIR/selfdrive/controls/lib/longitudinal_mpc_lib/c_generated_code/libacados_ocp_solver_long.so" \
+    "$DIR/opendbc_repo/opendbc/dbc/gm_global_a_powertrain_generated.dbc"
+}
+
+function prebuilt_compat_cache_valid {
+  [ -f "$PREBUILT_COMPAT_CACHE" ] || return 1
+  [ -f "$DIR/prebuilt" ] || return 1
+  [ "$DIR/prebuilt" -nt "$PREBUILT_COMPAT_CACHE" ] && return 1
+
+  while IFS= read -r artifact; do
+    [ -e "$artifact" ] || return 1
+    [ "$artifact" -nt "$PREBUILT_COMPAT_CACHE" ] && return 1
+  done < <(prebuilt_runtime_artifacts)
+
+  return 0
+}
+
+function prebuilt_artifacts_present {
+  while IFS= read -r artifact; do
+    [ -e "$artifact" ] || return 1
+  done < <(prebuilt_runtime_artifacts)
+
+  return 0
+}
+
 function agnos_init {
   sp_launch_timing "agnos_init_start"
 
@@ -126,11 +175,15 @@ function launch {
   # start manager
   cd system/manager
 
-  sp_launch_timing "launch_param_migrations_start"
-  if ! python3 ./launch_param_migrations.py; then
-    echo "Launch param migrations failed; continuing boot."
+  if launch_param_migrations_needed; then
+    sp_launch_timing "launch_param_migrations_start"
+    if ! python3 ./launch_param_migrations.py; then
+      echo "Launch param migrations failed; continuing boot."
+    fi
+    sp_launch_timing "launch_param_migrations_done"
+  else
+    sp_launch_timing "launch_param_migrations_skipped"
   fi
-  sp_launch_timing "launch_param_migrations_done"
 
   # Bootstrap runtime (e.g. /usr/comma after reset/uninstall) must go straight
   # to manager/setup flow. Do not run StarPilot prebuilt checks/builds here.
@@ -213,9 +266,21 @@ PY
   fi
 
   sp_launch_timing "prebuilt_decision_done"
-  if [ "$USE_PREBUILT" = "1" ] && [ -f $DIR/prebuilt ] && ! prebuilt_runtime_compatible; then
-    echo "Prebuilt runtime artifacts are incompatible on this device; rebuilding locally."
-    USE_PREBUILT=0
+  if [ "$USE_PREBUILT" = "1" ] && [ -f $DIR/prebuilt ]; then
+    if prebuilt_compat_cache_valid; then
+      sp_launch_timing "prebuilt_compat_cache_hit"
+    elif [ "${SP_STRICT_PREBUILT_CHECK:-0}" != "1" ] && prebuilt_artifacts_present; then
+      sp_launch_timing "prebuilt_compat_fast_path"
+      mkdir -p "$(dirname "$PREBUILT_COMPAT_CACHE")" 2>/dev/null || true
+      touch "$PREBUILT_COMPAT_CACHE" 2>/dev/null || true
+    elif prebuilt_runtime_compatible; then
+      mkdir -p "$(dirname "$PREBUILT_COMPAT_CACHE")" 2>/dev/null || true
+      touch "$PREBUILT_COMPAT_CACHE" 2>/dev/null || true
+    else
+      echo "Prebuilt runtime artifacts are incompatible on this device; rebuilding locally."
+      rm -f "$PREBUILT_COMPAT_CACHE" 2>/dev/null || true
+      USE_PREBUILT=0
+    fi
   fi
   sp_launch_timing "prebuilt_compat_done"
 

system/manager/helpers.py

@@ -51,17 +51,20 @@ def write_onroad_params(started, params):
 
 
 def save_bootlog():
-  # copy current params
-  tmp = tempfile.mkdtemp()
-  params_dirname = pathlib.Path(Params().get_param_path()).name
-  params_dir = os.path.join(tmp, params_dirname)
-  shutil.copytree(Params().get_param_path(), params_dir, dirs_exist_ok=True)
-
-  def fn(tmpdir):
+  def fn():
+    tmpdir = tempfile.mkdtemp()
     env = os.environ.copy()
     env['PARAMS_COPY_PATH'] = tmpdir
-    subprocess.call("./bootlog", cwd=os.path.join(BASEDIR, "system/loggerd"), env=env)
-    shutil.rmtree(tmpdir)
-  t = threading.Thread(target=fn, args=(tmp, ))
+
+    try:
+      params = Params()
+      params_dirname = pathlib.Path(params.get_param_path()).name
+      params_dir = os.path.join(tmpdir, params_dirname)
+      shutil.copytree(params.get_param_path(), params_dir, dirs_exist_ok=True)
+      subprocess.call("./bootlog", cwd=os.path.join(BASEDIR, "system/loggerd"), env=env)
+    finally:
+      shutil.rmtree(tmpdir, ignore_errors=True)
+
+  t = threading.Thread(target=fn)
   t.daemon = True
   t.start()

system/manager/manager.py

@@ -6,6 +6,7 @@ import shutil
 from pathlib import Path
 import signal
 import sys
+import threading
 import time
 import traceback
 
@@ -23,7 +24,6 @@ def _append_boot_timing_line(line: str) -> None:
 
 from cereal import car, log
 import cereal.messaging as messaging
-import openpilot.system.sentry as sentry
 from openpilot.common.utils import atomic_write
 from openpilot.common.params import Params, ParamKeyFlag, ParamKeyType
 from openpilot.common.text_window import TextWindow
@@ -83,6 +83,37 @@ def _log_boot_timing(scope: str, label: str, start: float, previous: float | Non
   return now
 
 
+def _sync_params_cache_async(cache_params_path: str, values: list[tuple[bytes | str, object]]) -> None:
+  try:
+    params_cache = Params(cache_params_path, return_defaults=True)
+    for key, value in values:
+      if params_cache.get(key) != value:
+        params_cache.put(key, value)
+  except Exception:
+    cloudlog.exception("failed to sync params cache")
+
+
+def _init_sentry_async() -> None:
+  def fn() -> None:
+    try:
+      import openpilot.system.sentry as sentry
+
+      sentry.init(sentry.SentryProject.SELFDRIVE)
+    except Exception:
+      cloudlog.exception("failed to initialize sentry")
+
+  threading.Thread(target=fn, daemon=True).start()
+
+
+def _capture_manager_exception() -> None:
+  try:
+    import openpilot.system.sentry as sentry
+
+    sentry.capture_exception()
+  except Exception:
+    cloudlog.exception("failed to capture manager exception")
+
+
 def _to_text(value):
   if value is None:
     return None
@@ -810,6 +841,7 @@ def manager_init() -> None:
   last_timing = _log_boot_timing("manager_init", "starpilot_migrations", manager_init_start, last_timing)
 
   # set unset params to their default value
+  params_cache_updates = []
   for k in params.all_keys():
     current_value = params.get(k)
     if current_value is None:
@@ -817,7 +849,9 @@ def manager_init() -> None:
       if cached_value is not None:
         params.put(k, cached_value)
     else:
-      params_cache.put(k, current_value)
+      params_cache_updates.append((k, current_value))
+  if params_cache_updates:
+    threading.Thread(target=_sync_params_cache_async, args=(cache_params_path, params_cache_updates), daemon=True).start()
   last_timing = _log_boot_timing("manager_init", "params_defaults_cache_sync", manager_init_start, last_timing)
 
   # Create folders needed for msgq
@@ -862,7 +896,6 @@ def manager_init() -> None:
     os.environ['CLEAN'] = '1'
 
   # init logging
-  sentry.init(sentry.SentryProject.SELFDRIVE)
   cloudlog.bind_global(dongle_id=dongle_id,
                        version=build_metadata.openpilot.version,
                        origin=build_metadata.openpilot.git_normalized_origin,
@@ -870,12 +903,18 @@ def manager_init() -> None:
                        commit=build_metadata.openpilot.git_commit,
                        dirty=build_metadata.openpilot.is_dirty,
                        device=HARDWARE.get_device_type())
+  _init_sentry_async()
   last_timing = _log_boot_timing("manager_init", "logging_ready", manager_init_start, last_timing)
 
-  # preimport all processes
-  for p in managed_processes.values():
-    p.prepare()
-  last_timing = _log_boot_timing("manager_init", "preimport_processes", manager_init_start, last_timing)
+  # Preimporting every process serializes a lot of import work before manager can
+  # start the always-on processes. Keep the old behavior available for debugging,
+  # but optimize normal boot by letting children import their modules in parallel.
+  if os.getenv("SP_MANAGER_PREIMPORT", "0").lower() in ("1", "true", "yes", "on"):
+    for p in managed_processes.values():
+      p.prepare()
+    last_timing = _log_boot_timing("manager_init", "preimport_processes", manager_init_start, last_timing)
+  else:
+    last_timing = _log_boot_timing("manager_init", "preimport_processes_skipped", manager_init_start, last_timing)
 
   # StarPilot variables
   install_starpilot(build_metadata, params)
@@ -1024,7 +1063,7 @@ def main() -> None:
     manager_thread()
   except Exception:
     traceback.print_exc()
-    sentry.capture_exception()
+    _capture_manager_exception()
   finally:
     manager_cleanup()
 

system/manager/process.py

@@ -16,7 +16,6 @@ from setproctitle import setproctitle
 
 from cereal import car, log
 import cereal.messaging as messaging
-import openpilot.system.sentry as sentry
 from openpilot.common.basedir import BASEDIR
 from openpilot.common.params import Params
 from openpilot.common.swaglog import cloudlog
@@ -429,7 +428,6 @@ def launcher(proc: str, name: str, nice: int | None = None) -> None:
 
     # add daemon name tag to logs
     cloudlog.bind(daemon=name)
-    sentry.set_tag("daemon", name)
 
     # exec the process
     mod.main()
@@ -438,7 +436,13 @@ def launcher(proc: str, name: str, nice: int | None = None) -> None:
   except Exception:
     # can't install the crash handler because sys.excepthook doesn't play nice
     # with threads, so catch it here.
-    sentry.capture_exception()
+    try:
+      import openpilot.system.sentry as sentry
+
+      sentry.set_tag("daemon", name)
+      sentry.capture_exception()
+    except Exception:
+      cloudlog.exception(f"failed to capture exception for child {proc}")
     raise
 
 

system/manager/test/test_launch_param_migrations.py

@@ -50,7 +50,9 @@ class FileBackedFakeParams:
 
 
 def marker_path(tmp_path: Path, marker_name: str) -> Path:
-  return tmp_path / MARKER_DIRNAME / "params" / marker_name
+  path = tmp_path / MARKER_DIRNAME / "params" / marker_name
+  path.parent.mkdir(parents=True, exist_ok=True)
+  return path
 
 
 def test_apply_launch_param_migrations_sets_branch_defaults_once(tmp_path):