Files
OmniSocketGo/scripts/boot/common.sh

662 lines
18 KiB
Bash

#!/usr/bin/env bash
set -euo pipefail
BOOT_SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
DEV_SCRIPT_DIR="$(cd "${BOOT_SCRIPT_DIR}/../dev" && pwd)"
source_with_nounset_off() {
set +u
# shellcheck disable=SC1090
source "$1"
set -u
}
blitz_host_from_addr() {
local value="${1:-}"
if [[ -z "${value}" ]]; then
return 1
fi
if [[ "${value}" == \[*\]:* ]]; then
value="${value#\[}"
printf '%s\n' "${value%%]:*}"
return 0
fi
printf '%s\n' "${value%%:*}"
}
blitz_load_boot_env() {
local env_file
local default_time_server
local dev_run_root
local dev_runtime_dir
if [[ "${BLITZ_BOOT_ENV_LOADED:-0}" == "1" ]]; then
return 0
fi
export BLITZ_BOOT_LOADING_ENV="1"
# shellcheck disable=SC1091
source "${DEV_SCRIPT_DIR}/load-env.sh"
unset BLITZ_BOOT_LOADING_ENV
for env_file in \
"${BOOT_SCRIPT_DIR}/robot-boot.env" \
"${BOOT_SCRIPT_DIR}/robot-boot.env.local"
do
if [[ -f "${env_file}" ]]; then
set -a
# shellcheck disable=SC1090
source "${env_file}"
set +a
fi
done
if declare -F normalize_loaded_env_vars >/dev/null 2>&1; then
normalize_loaded_env_vars
fi
dev_run_root="${OMNISOCKETGO_ROOT}/logs"
dev_runtime_dir="${dev_run_root}/runtime"
if [[ -z "${BLITZ_RUN_ROOT:-}" || "${BLITZ_RUN_ROOT}" == "${dev_run_root}" ]]; then
export BLITZ_RUN_ROOT="/var/log/blitz-robot"
fi
if [[ -z "${BLITZ_RUNTIME_DIR:-}" || "${BLITZ_RUNTIME_DIR}" == "${dev_runtime_dir}" ]]; then
export BLITZ_RUNTIME_DIR="/run/blitz-robot"
fi
if [[ -z "${BLITZ_RUN_CONTEXT_FILE:-}" || "${BLITZ_RUN_CONTEXT_FILE}" == "${dev_runtime_dir}/run-context.env" ]]; then
export BLITZ_RUN_CONTEXT_FILE="${BLITZ_RUNTIME_DIR}/run-context.env"
fi
if [[ -z "${BLITZ_RUN_ID_FILE:-}" || "${BLITZ_RUN_ID_FILE}" == "${dev_runtime_dir}/run-id" ]]; then
export BLITZ_RUN_ID_FILE="${BLITZ_RUNTIME_DIR}/run-id"
fi
if [[ -z "${BLITZ_CURRENT_RUN_LINK:-}" || "${BLITZ_CURRENT_RUN_LINK}" == "${dev_run_root}/current" ]]; then
export BLITZ_CURRENT_RUN_LINK="${BLITZ_RUN_ROOT}/current"
fi
default_time_server="$(blitz_host_from_addr "${ROBOT_SIDE_OMNISOCKET_SERVER_ADDR:-}" || true)"
export BLITZ_BOOT_DELAY_SEC="${BLITZ_BOOT_DELAY_SEC:-30}"
export BLITZ_RUN_ROOT="${BLITZ_RUN_ROOT:-/var/log/blitz-robot}"
export BLITZ_LOG_FILE="${BLITZ_LOG_FILE:-/var/log/blitz-robot/startup.log}"
export BLITZ_RUNTIME_DIR="${BLITZ_RUNTIME_DIR:-/run/blitz-robot}"
export BLITZ_RUN_CONTEXT_FILE="${BLITZ_RUN_CONTEXT_FILE:-${BLITZ_RUNTIME_DIR}/run-context.env}"
export BLITZ_RUN_ID_FILE="${BLITZ_RUN_ID_FILE:-${BLITZ_RUNTIME_DIR}/run-id}"
export BLITZ_CURRENT_RUN_LINK="${BLITZ_CURRENT_RUN_LINK:-${BLITZ_RUN_ROOT}/current}"
export BLITZ_5G_DIAL_DIR="${BLITZ_5G_DIAL_DIR:-${BOOT_SCRIPT_DIR}}"
export BLITZ_5G_SERIAL_PORT="${BLITZ_5G_SERIAL_PORT:-/dev/ttyUSB7}"
export BLITZ_5G_INTERFACE="${BLITZ_5G_INTERFACE:-}"
export BLITZ_5G_MODEM_SUBNET="${BLITZ_5G_MODEM_SUBNET:-192.168.224.0/22}"
export BLITZ_5G_GATEWAY="${BLITZ_5G_GATEWAY:-192.168.225.1}"
export BLITZ_5G_SKIP_DHCP="${BLITZ_5G_SKIP_DHCP:-0}"
export BLITZ_5G_REMOVE_DEFAULT_ROUTE="${BLITZ_5G_REMOVE_DEFAULT_ROUTE:-1}"
export BLITZ_5G_ROUTE_TARGETS="${BLITZ_5G_ROUTE_TARGETS:-106.55.173.235}"
export BLITZ_5G_INFO_JSON="${BLITZ_5G_INFO_JSON:-${BLITZ_5G_DIAL_DIR}/modem_network_info.json}"
export BLITZ_5G_DISABLE_INTERFACES="${BLITZ_5G_DISABLE_INTERFACES:-}"
export BLITZ_5G_SERIAL_WAIT_SEC="${BLITZ_5G_SERIAL_WAIT_SEC:-60}"
export BLITZ_5G_ROUTE_WAIT_SEC="${BLITZ_5G_ROUTE_WAIT_SEC:-30}"
export BLITZ_TIME_SERVER_IP="${BLITZ_TIME_SERVER_IP:-${default_time_server}}"
export BLITZ_ROS_USER="${BLITZ_ROS_USER:-nvidia}"
export BLITZ_ROS_SOCKET_WAIT_SEC="${BLITZ_ROS_SOCKET_WAIT_SEC:-20}"
export BLITZ_WATCHDOG_INTERVAL_SEC="${BLITZ_WATCHDOG_INTERVAL_SEC:-5}"
export BLITZ_HEALTH_STALE_SEC="${BLITZ_HEALTH_STALE_SEC:-15}"
export BLITZ_OMNID_THREAD_HEARTBEAT_TIMEOUT_SEC="${BLITZ_OMNID_THREAD_HEARTBEAT_TIMEOUT_SEC:-15}"
export BLITZ_KCP_STATS_INTERVAL_MS="${BLITZ_KCP_STATS_INTERVAL_MS:-1000}"
export BLITZ_CONTROL_LATENCY_LOG_ENABLED="${BLITZ_CONTROL_LATENCY_LOG_ENABLED:-1}"
export BLITZ_CONTROL_LATENCY_LOG_SAMPLE_MOD="${BLITZ_CONTROL_LATENCY_LOG_SAMPLE_MOD:-100}"
export BLITZ_5G_LINK_LOG_INTERVAL_SEC="${BLITZ_5G_LINK_LOG_INTERVAL_SEC:-5}"
export BLITZ_JSONL_FLUSH_INTERVAL_MS="${BLITZ_JSONL_FLUSH_INTERVAL_MS:-1000}"
export BLITZ_JSONL_FLUSH_BYTES="${BLITZ_JSONL_FLUSH_BYTES:-262144}"
export BLITZ_JSONL_ROTATE_BYTES="${BLITZ_JSONL_ROTATE_BYTES:-134217728}"
export BLITZ_JSONL_ROTATE_FILES="${BLITZ_JSONL_ROTATE_FILES:-8}"
export BLITZ_INCIDENT_COMMAND_TIMEOUT_SEC="${BLITZ_INCIDENT_COMMAND_TIMEOUT_SEC:-5}"
export BLITZ_INCIDENT_TOTAL_TIMEOUT_SEC="${BLITZ_INCIDENT_TOTAL_TIMEOUT_SEC:-30}"
export BLITZ_NETWORK_FAIL_THRESHOLD="${BLITZ_NETWORK_FAIL_THRESHOLD:-3}"
export BLITZ_NETWORK_RECOVERY_COOLDOWN_SEC="${BLITZ_NETWORK_RECOVERY_COOLDOWN_SEC:-30}"
export BLITZ_GPS_MONITOR_ENABLED="${BLITZ_GPS_MONITOR_ENABLED:-1}"
export BLITZ_GPS_DEVICE_GLOB="${BLITZ_GPS_DEVICE_GLOB:-/dev/ttyCH341USB*}"
export BLITZ_GPS_CHECK_INTERVAL_SEC="${BLITZ_GPS_CHECK_INTERVAL_SEC:-10}"
export BLITZ_GPS_RESTART_UNITS="${BLITZ_GPS_RESTART_UNITS:-gpsd.socket gpsd.service}"
export BLITZ_WATCHDOG_ALLOW_FAULT_INJECTION="${BLITZ_WATCHDOG_ALLOW_FAULT_INJECTION:-0}"
export BLITZ_BOOT_ENV_LOADED="1"
}
blitz_timestamp() {
date '+%Y-%m-%d %H:%M:%S%z'
}
blitz_sanitize_detail() {
local detail="${1:-}"
detail="${detail//$'\n'/ ; }"
detail="${detail//$'\r'/ }"
printf '%s' "${detail}"
}
blitz_log() {
local step="${1:-unknown-step}"
local action="${2:-unknown-action}"
local result="${3:-info}"
local details="${4:-}"
local exit_code="${5:-0}"
printf '%s | %s | %s | %s | %s | %s\n' \
"$(blitz_timestamp)" \
"${step}" \
"${action}" \
"${result}" \
"$(blitz_sanitize_detail "${details}")" \
"${exit_code}"
}
blitz_join_cmd() {
local cmd=()
local arg
for arg in "$@"; do
cmd+=("$(printf '%q' "${arg}")")
done
printf '%s' "${cmd[*]}"
}
blitz_require_command() {
local command_name="$1"
local step="${2:-precheck}"
if command -v "${command_name}" >/dev/null 2>&1; then
blitz_log "${step}" "require-command" "success" "command=${command_name}" 0
return 0
fi
blitz_log "${step}" "require-command" "failure" "missing command: ${command_name}" 127
return 127
}
blitz_require_file() {
local path="$1"
local step="${2:-precheck}"
if [[ -f "${path}" ]]; then
blitz_log "${step}" "require-file" "success" "path=${path}" 0
return 0
fi
blitz_log "${step}" "require-file" "failure" "missing file: ${path}" 1
return 1
}
blitz_require_executable() {
local path="$1"
local step="${2:-precheck}"
if [[ -x "${path}" ]]; then
blitz_log "${step}" "require-executable" "success" "path=${path}" 0
return 0
fi
blitz_log "${step}" "require-executable" "failure" "missing executable: ${path}" 1
return 1
}
blitz_require_root() {
local step="${1:-precheck}"
if [[ "${EUID}" -eq 0 ]]; then
blitz_log "${step}" "require-root" "success" "uid=${EUID}" 0
return 0
fi
blitz_log "${step}" "require-root" "failure" "root privileges are required" 1
return 1
}
blitz_run() {
local step="$1"
local action="$2"
local rc
shift 2
blitz_log "${step}" "${action}" "start" "$(blitz_join_cmd "$@")" 0
if "$@"; then
blitz_log "${step}" "${action}" "success" "$(blitz_join_cmd "$@")" 0
return 0
else
rc=$?
fi
blitz_log "${step}" "${action}" "failure" "$(blitz_join_cmd "$@")" "${rc}"
return "${rc}"
}
blitz_route_ready() {
local target_ip="$1"
local expected_interface="${2:-}"
local route_output
route_output="$(ip route get "${target_ip}" 2>&1 || true)"
if [[ -z "${route_output}" ]]; then
return 1
fi
if [[ "${route_output}" == *"unreachable"* || "${route_output}" == *"prohibit"* ]]; then
return 1
fi
if [[ -n "${expected_interface}" && "${route_output}" != *" dev ${expected_interface} "* && "${route_output}" != *" dev ${expected_interface}" ]]; then
return 1
fi
printf '%s\n' "${route_output}"
return 0
}
blitz_interface_exists() {
local interface_name="${1:-}"
if [[ -z "${interface_name}" ]]; then
return 1
fi
ip link show dev "${interface_name}" >/dev/null 2>&1
}
blitz_read_5g_info_interface() {
local info_json="$1"
if [[ -z "${info_json}" || ! -f "${info_json}" ]]; then
return 1
fi
python3 - "${info_json}" <<'PY'
import json
import sys
path = sys.argv[1]
try:
with open(path, "r", encoding="utf-8") as handle:
payload = json.load(handle)
except Exception:
raise SystemExit(1)
interface = str(payload.get("interface") or "").strip()
if not interface:
raise SystemExit(1)
print(interface)
PY
}
blitz_detect_5g_interface_from_subnet() {
local modem_subnet="${1:-${BLITZ_5G_MODEM_SUBNET:-}}"
if [[ -z "${modem_subnet}" ]]; then
return 1
fi
python3 - "${modem_subnet}" <<'PY'
import ipaddress
import json
import subprocess
import sys
subnet = ipaddress.ip_network(sys.argv[1], strict=False)
skip = {"lo", "docker0", "l4tbr0"}
def priority(name: str) -> tuple[int, str]:
if name.startswith("enx"):
return (0, name)
if name.startswith("wwan"):
return (1, name)
if name.startswith("usb"):
return (2, name)
if name.startswith("eth"):
return (3, name)
return (9, name)
try:
output = subprocess.check_output(["ip", "-j", "-4", "addr", "show"], text=True)
payload = json.loads(output)
except Exception:
raise SystemExit(1)
candidates = []
for item in payload:
ifname = str(item.get("ifname") or "").strip()
if not ifname or ifname in skip:
continue
for addr in item.get("addr_info") or []:
if addr.get("family") != "inet":
continue
local = addr.get("local")
prefixlen = addr.get("prefixlen")
if not local or prefixlen is None:
continue
try:
iface = ipaddress.ip_interface(f"{local}/{prefixlen}")
except ValueError:
continue
if iface.ip in subnet:
candidates.append((priority(ifname), ifname))
break
if not candidates:
raise SystemExit(1)
candidates.sort(key=lambda item: item[0])
print(candidates[0][1])
PY
}
blitz_refresh_5g_info_json() {
local interface_name="$1"
local info_json="${2:-${BLITZ_5G_INFO_JSON:-}}"
if [[ -z "${interface_name}" || -z "${info_json}" ]]; then
return 1
fi
python3 - "${interface_name}" "${info_json}" <<'PY'
import json
import os
import subprocess
import sys
interface_name = sys.argv[1]
path = sys.argv[2]
try:
output = subprocess.check_output(["ip", "-j", "addr", "show", "dev", interface_name], text=True)
payload = json.loads(output)
except Exception:
raise SystemExit(1)
if not payload:
raise SystemExit(1)
item = payload[0]
ipv4 = []
ipv6 = []
for addr in item.get("addr_info") or []:
local = addr.get("local")
prefixlen = addr.get("prefixlen")
family = addr.get("family")
if not local or prefixlen is None:
continue
entry = f"{local}/{prefixlen}"
if family == "inet":
ipv4.append(entry)
elif family == "inet6":
ipv6.append(entry)
data = {
"interface": interface_name,
"ipv4": ipv4,
"ipv6": ipv6,
}
parent = os.path.dirname(path)
if parent:
os.makedirs(parent, exist_ok=True)
temp_path = f"{path}.tmp.{os.getpid()}"
with open(temp_path, "w", encoding="utf-8") as handle:
json.dump(data, handle, ensure_ascii=False, indent=2)
os.replace(temp_path, path)
PY
}
blitz_resolve_5g_interface() {
local explicit_interface="${BLITZ_5G_INTERFACE:-}"
local info_json="${BLITZ_5G_INFO_JSON:-}"
local recorded_interface=""
local detected_interface=""
if [[ -n "${explicit_interface}" ]]; then
if blitz_interface_exists "${explicit_interface}"; then
printf '%s\n' "${explicit_interface}"
return 0
fi
return 1
fi
recorded_interface="$(blitz_read_5g_info_interface "${info_json}" || true)"
if [[ -n "${recorded_interface}" ]] && blitz_interface_exists "${recorded_interface}"; then
printf '%s\n' "${recorded_interface}"
return 0
fi
detected_interface="$(blitz_detect_5g_interface_from_subnet || true)"
if [[ -n "${detected_interface}" ]]; then
if [[ "${detected_interface}" != "${recorded_interface}" ]]; then
blitz_refresh_5g_info_json "${detected_interface}" "${info_json}" >/dev/null 2>&1 || true
fi
printf '%s\n' "${detected_interface}"
return 0
fi
return 1
}
blitz_prepare_runtime_dir() {
local runtime_dir
blitz_load_boot_env
runtime_dir="${BLITZ_RUNTIME_DIR}"
mkdir -p "${runtime_dir}"
if [[ "${EUID}" -eq 0 ]]; then
chown "root:${BLITZ_ROS_USER}" "${runtime_dir}"
chmod 0775 "${runtime_dir}"
else
chmod 0775 "${runtime_dir}" 2>/dev/null || true
fi
blitz_log "runtime-dir" "prepare" "success" "path=${runtime_dir}" 0
}
blitz_prepare_run_root() {
local run_root
local run_dir
local incidents_dir
blitz_load_boot_env
run_root="${BLITZ_RUN_ROOT}"
run_dir="${run_root}/runs"
incidents_dir="${run_root}/incidents"
mkdir -p "${run_dir}" "${incidents_dir}"
if [[ "${EUID}" -eq 0 ]]; then
chown -R "root:${BLITZ_ROS_USER}" "${run_root}" 2>/dev/null || true
chmod 0775 "${run_root}" "${run_dir}" "${incidents_dir}" 2>/dev/null || true
fi
}
blitz_load_run_context_env() {
local context_file="${1:-${BLITZ_RUN_CONTEXT_FILE:-}}"
if [[ -z "${context_file}" || ! -f "${context_file}" ]]; then
return 1
fi
set -a
# shellcheck disable=SC1090
source "${context_file}"
set +a
return 0
}
blitz_read_run_id() {
local run_id_file="${BLITZ_RUN_ID_FILE:-}"
if [[ -z "${run_id_file}" || ! -f "${run_id_file}" ]]; then
return 1
fi
tr -d '\r\n' < "${run_id_file}"
}
blitz_utc_compact_timestamp() {
date -u '+%Y%m%dT%H%M%SZ'
}
blitz_new_run_id() {
printf '%s\n' "$(blitz_utc_compact_timestamp)"
}
blitz_new_incident_id() {
local prefix="${1:-incident}"
printf '%s-%s-%d\n' "${prefix}" "$(blitz_utc_compact_timestamp)" "$$"
}
blitz_new_instance_id() {
printf '%s-%d\n' "$(blitz_utc_compact_timestamp)" "$$"
}
blitz_git_commit() {
git -C "${OMNISOCKETGO_ROOT}" rev-parse HEAD 2>/dev/null || true
}
blitz_git_dirty_flag() {
if git -C "${OMNISOCKETGO_ROOT}" diff --quiet --ignore-submodules=dirty >/dev/null 2>&1; then
printf '0\n'
return 0
fi
printf '1\n'
}
blitz_write_run_context() {
local run_id="$1"
local run_dir="$2"
local boot_id="$3"
local context_file="${BLITZ_RUN_CONTEXT_FILE}"
local id_file="${BLITZ_RUN_ID_FILE}"
local temp_context
local temp_info
local commit_hash
local dirty_flag
local started_at
commit_hash="$(blitz_git_commit)"
dirty_flag="$(blitz_git_dirty_flag)"
started_at="$(date -u '+%Y-%m-%dT%H:%M:%SZ')"
temp_context="${context_file}.tmp.$$"
temp_info="${run_dir}/run-info.json.tmp.$$"
mkdir -p "${run_dir}"
printf '%s\n' "${run_id}" > "${id_file}"
cat > "${temp_context}" <<EOF
BLITZ_RUN_ID=${run_id}
BLITZ_RUN_DIR=${run_dir}
BLITZ_BOOT_ID=${boot_id}
BLITZ_RUN_ROOT=${BLITZ_RUN_ROOT}
EOF
mv -f "${temp_context}" "${context_file}"
python3 - "${temp_info}" "${run_id}" "${run_dir}" "${boot_id}" "${started_at}" "${commit_hash}" "${dirty_flag}" "${HOSTNAME:-$(hostname)}" <<'PY'
import json
import os
import sys
path, run_id, run_dir, boot_id, started_at, commit_hash, dirty_flag, hostname = sys.argv[1:9]
payload = {
"run_id": run_id,
"run_dir": run_dir,
"boot_id": boot_id,
"started_at": started_at,
"hostname": hostname,
"git_commit": commit_hash,
"git_dirty": dirty_flag == "1",
"env": {
key: os.environ.get(key, "")
for key in sorted(os.environ)
if key.startswith(("BLITZ_", "OMNI_", "ROBOT_RECEIVER_"))
},
}
with open(path, "w", encoding="utf-8") as handle:
json.dump(payload, handle, ensure_ascii=False, indent=2, sort_keys=True)
PY
mv -f "${temp_info}" "${run_dir}/run-info.json"
ln -sfn "${run_dir}" "${BLITZ_CURRENT_RUN_LINK}"
}
blitz_init_run_context() {
local run_id
local boot_id
local run_dir
blitz_load_boot_env
blitz_prepare_runtime_dir
blitz_prepare_run_root
run_id="$(blitz_new_run_id)"
boot_id="$(cat /proc/sys/kernel/random/boot_id 2>/dev/null || blitz_new_run_id)"
run_dir="${BLITZ_RUN_ROOT}/runs/${run_id}"
export BLITZ_RUN_ID="${run_id}"
export BLITZ_RUN_DIR="${run_dir}"
export BLITZ_BOOT_ID="${boot_id}"
blitz_write_run_context "${run_id}" "${run_dir}" "${boot_id}"
blitz_log "run-context" "init" "success" "run_id=${run_id} run_dir=${run_dir}" 0
}
blitz_require_run_context() {
blitz_load_boot_env
if blitz_load_run_context_env; then
return 0
fi
blitz_log "run-context" "load" "failure" "missing ${BLITZ_RUN_CONTEXT_FILE}" 1
return 1
}
blitz_ensure_instance_id() {
if [[ -n "${BLITZ_INSTANCE_ID:-}" ]]; then
return 0
fi
export BLITZ_INSTANCE_ID="$(blitz_new_instance_id)"
}
blitz_jsonl_rotate_if_needed() {
local path="$1"
local max_bytes="${2:-${BLITZ_JSONL_ROTATE_BYTES:-0}}"
local max_files="${3:-${BLITZ_JSONL_ROTATE_FILES:-0}}"
local size=0
local index
if [[ -z "${path}" || ! -f "${path}" ]]; then
return 0
fi
if (( max_bytes <= 0 || max_files <= 0 )); then
return 0
fi
size="$(stat -c %s "${path}" 2>/dev/null || echo 0)"
if (( size < max_bytes )); then
return 0
fi
for (( index=max_files; index>=1; index-- )); do
if [[ "${index}" -eq "${max_files}" ]]; then
rm -f "${path}.${index}"
fi
if [[ -f "${path}.${index}" ]]; then
mv -f "${path}.${index}" "${path}.$(( index + 1 ))"
fi
done
mv -f "${path}" "${path}.1"
}
blitz_jsonl_append_line() {
local path="$1"
local line="$2"
mkdir -p "$(dirname "${path}")"
blitz_jsonl_rotate_if_needed "${path}"
printf '%s\n' "${line}" >> "${path}"
}
blitz_launch_incident_capture() {
local launch_script="${BOOT_SCRIPT_DIR}/blitz-incident-capture-launch.sh"
if [[ ! -f "${launch_script}" ]]; then
return 1
fi
/bin/bash "${launch_script}" "$@" >/dev/null 2>&1 || return 1
}