Zack Williams | 9072664 | 2021-03-30 18:10:09 -0700 | [diff] [blame^] | 1 | #!/usr/bin/env python3 |
| 2 | |
| 3 | # Copyright 2020-present Open Networking Foundation |
| 4 | # |
| 5 | # Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 | # you may not use this file except in compliance with the License. |
| 7 | # You may obtain a copy of the License at |
| 8 | # |
| 9 | # http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | # |
| 11 | # Unless required by applicable law or agreed to in writing, software |
| 12 | # distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | # See the License for the specific language governing permissions and |
| 15 | # limitations under the License |
| 16 | |
| 17 | import sys |
| 18 | import os |
| 19 | import json |
| 20 | import logging |
| 21 | import enum |
| 22 | import requests |
| 23 | import time |
| 24 | import serial |
| 25 | import subprocess |
| 26 | from collections import namedtuple |
| 27 | |
| 28 | """ |
| 29 | "Simple" script that checks Aether network operational status periodically |
| 30 | by controlling the attached 4G/LTE modem with AT commands and |
| 31 | report the result to the central monitoring server. |
| 32 | """ |
| 33 | |
| 34 | |
| 35 | CONF = json.loads( |
| 36 | open(os.getenv("CONFIG_FILE", "./config.json")).read(), |
| 37 | object_hook=lambda d: namedtuple("X", d.keys())(*d.values()), |
| 38 | ) |
| 39 | |
| 40 | logging.basicConfig( |
| 41 | filename=CONF.log_file, |
| 42 | format="%(asctime)s [%(levelname)s] %(message)s", |
| 43 | level=logging.ERROR, |
| 44 | ) |
| 45 | |
| 46 | report = { |
| 47 | "name": CONF.edge_name, |
| 48 | "status": {"control_plane": None, "user_plane": None}, |
| 49 | "speedtest": { |
| 50 | "ping": {"dns": {"min": None, "avg": None, "max": None, "stddev": None}} |
| 51 | }, |
| 52 | } |
| 53 | |
| 54 | |
| 55 | class State(enum.Enum): |
| 56 | error = "-1" |
| 57 | disconnected = "0" |
| 58 | connected = "1" |
| 59 | |
| 60 | @classmethod |
| 61 | def has_value(cls, value): |
| 62 | return value in cls._value2member_map_ |
| 63 | |
| 64 | |
| 65 | class Modem: |
| 66 | log = logging.getLogger("aether_edge_monitoring.Modem") |
| 67 | |
| 68 | read_timeout = 0.1 |
| 69 | |
| 70 | def __init__(self, port, baudrate): |
| 71 | self.port = port |
| 72 | self.baudrate = baudrate |
| 73 | self._response = None |
| 74 | |
| 75 | def connect(self): |
| 76 | self.serial = serial.Serial(port=self.port, baudrate=self.baudrate, timeout=1) |
| 77 | |
| 78 | def _write(self, command): |
| 79 | if self.serial.inWaiting() > 0: |
| 80 | self.serial.flushInput() |
| 81 | |
| 82 | self._response = b"" |
| 83 | |
| 84 | self.serial.write(bytearray(command + "\r", "ascii")) |
| 85 | read = self.serial.inWaiting() |
| 86 | while True: |
| 87 | if read > 0: |
| 88 | self._response += self.serial.read(read) |
| 89 | else: |
| 90 | time.sleep(self.read_timeout) |
| 91 | read = self.serial.inWaiting() |
| 92 | if read == 0: |
| 93 | break |
| 94 | return self._response.decode("ascii").replace("\r\n", " ") |
| 95 | |
| 96 | def write(self, command, wait_resp=True): |
| 97 | response = self._write(command) |
| 98 | self.log.debug("%s: %s", command, response) |
| 99 | |
| 100 | if wait_resp and "ERROR" in response: |
| 101 | return False, None |
| 102 | return True, response |
| 103 | |
| 104 | def is_connected(self): |
| 105 | success, result = self.write("AT+CGATT?") |
| 106 | if not success or "CGATT:" not in result: |
| 107 | return State.error |
| 108 | state = result.split("CGATT:")[1].split(" ")[0] |
| 109 | return State(state) |
| 110 | |
| 111 | def close(self): |
| 112 | self.serial.close() |
| 113 | |
| 114 | |
| 115 | def get_control_plane_state(modem): |
| 116 | # Delete the existing session |
| 117 | # "echo" works more stable than serial for this action |
| 118 | # success, result = modem.write('AT+CFUN=0') |
| 119 | logging.debug("echo 'AT+CFUN=0' > " + CONF.modem.port) |
| 120 | success = os.system("echo 'AT+CFUN=0' > " + CONF.modem.port) |
| 121 | logging.debug("result: %s", success) |
| 122 | if success is not 0: |
| 123 | msg = "Write 'AT+CFUN=0' failed" |
| 124 | logging.error(msg) |
| 125 | return State.error, msg |
| 126 | |
| 127 | # Wait until the modem is fully disconnected |
| 128 | retry = 0 |
| 129 | state = None |
| 130 | while retry < 5: |
| 131 | state = modem.is_connected() |
| 132 | if state is State.disconnected: |
| 133 | break |
| 134 | time.sleep(1) |
| 135 | retry += 1 |
| 136 | |
| 137 | # Consider the modem is not responding if disconnection failed |
| 138 | if state is not State.disconnected: |
| 139 | msg = "Failed to disconnect." |
| 140 | logging.error(msg) |
| 141 | return State.error, msg |
| 142 | |
| 143 | time.sleep(2) |
| 144 | # Create a new session |
| 145 | # "echo" works more stable than serial for this action |
| 146 | # success, result = modem.write('AT+CGATT=1') |
| 147 | logging.debug("echo 'AT+CFUN=1' > " + CONF.modem.port) |
| 148 | success = os.system("echo 'AT+CFUN=1' > " + CONF.modem.port) |
| 149 | logging.debug("result: %s", success) |
| 150 | if success is not 0: |
| 151 | msg = "Write 'AT+CFUN=1' failed" |
| 152 | logging.error(msg) |
| 153 | return State.error, msg |
| 154 | |
| 155 | # Give 10 sec for the modem to be fully connected |
| 156 | retry = 0 |
| 157 | while retry < 30: |
| 158 | state = modem.is_connected() |
| 159 | if state is State.connected: |
| 160 | break |
| 161 | time.sleep(1) |
| 162 | retry += 1 |
| 163 | # CGATT sometimes returns None |
| 164 | if state is State.error: |
| 165 | state = State.disconnected |
| 166 | |
| 167 | time.sleep(2) |
| 168 | return state, None |
| 169 | |
| 170 | |
| 171 | def get_user_plane_state(modem): |
| 172 | resp = os.system("ping -c 3 " + CONF.ips.user_plane_ping_test + ">/dev/null 2>&1") |
| 173 | return State.connected if resp is 0 else State.disconnected, None |
| 174 | |
| 175 | |
| 176 | def run_ping_test(ip, count): |
| 177 | """ |
| 178 | Runs the ping test |
| 179 | Input: IP to ping, # times to ping |
| 180 | Returns: dict of the min/avg/max/stddev numbers from the ping command result |
| 181 | """ |
| 182 | result = {"min": 0.0, "avg": 0.0, "max": 0.0, "stddev": 0.0} |
| 183 | try: |
| 184 | pingResult = ( |
| 185 | subprocess.check_output( |
| 186 | "ping -c " + str(count) + " " + ip + " | tail -1 | awk '{print $4}'", |
| 187 | shell=True, |
| 188 | ) |
| 189 | .decode("UTF-8") |
| 190 | .split("/") |
| 191 | ) |
| 192 | result = { |
| 193 | "min": float(pingResult[0]), |
| 194 | "avg": float(pingResult[1]), |
| 195 | "max": float(pingResult[2]), |
| 196 | "stddev": float(pingResult[3]), |
| 197 | } |
| 198 | except Exception as e: |
| 199 | logging.error("Ping test failed for " + ip + ": %s", e) |
| 200 | return result |
| 201 | |
| 202 | |
| 203 | def get_ping_test(modem): |
| 204 | """ |
| 205 | Each ping test result saves the min/avg/max/stddev to dict. |
| 206 | 1) Performs ping test to Google Public DNS for 10 iterations. |
| 207 | 2) # TODO: Performs ping to device on network. |
| 208 | """ |
| 209 | speedtest_ping = {} |
| 210 | speedtest_ping["dns"] = run_ping_test(CONF.ips.speedtest_ping_dns, 10) |
| 211 | return speedtest_ping |
| 212 | |
| 213 | |
| 214 | def report_status(cp_state, up_state, speedtest_ping): |
| 215 | report["status"]["control_plane"] = cp_state.name |
| 216 | report["status"]["user_plane"] = up_state.name |
| 217 | report["speedtest"]["ping"] = speedtest_ping |
| 218 | |
| 219 | logging.info("Sending report %s", report) |
| 220 | try: |
| 221 | result = requests.post(CONF.report_url, json=report) |
| 222 | except requests.exceptions.ConnectionError: |
| 223 | logging.error("Failed to report for %s", e) |
| 224 | pass |
| 225 | try: |
| 226 | result.raise_for_status() |
| 227 | except requests.exceptions.HTTPError as e: |
| 228 | logging.error("Failed to report for %s", e) |
| 229 | pass |
| 230 | |
| 231 | |
| 232 | def main(): |
| 233 | modem = Modem(CONF.modem.port, CONF.modem.baud) |
| 234 | try: |
| 235 | modem.connect() |
| 236 | except serial.serialutil.SerialException as e: |
| 237 | logging.error("Failed to connect the modem for %s", e) |
| 238 | sys.exit(1) |
| 239 | |
| 240 | for ip in CONF.ips: |
| 241 | success = os.system( |
| 242 | "sudo ip route replace {}/32 via {}".format(ip, CONF.modem.ip_addr) |
| 243 | ) |
| 244 | if success is not 0: |
| 245 | logging.error("Failed to add test routing to " + ip) |
| 246 | sys.exit(1) |
| 247 | |
| 248 | while True: |
| 249 | cp_state, cp_msg = get_control_plane_state(modem) |
| 250 | up_state, up_msg = get_user_plane_state(modem) |
| 251 | speedtest_ping = get_ping_test(modem) |
| 252 | |
| 253 | if cp_state is State.error: |
| 254 | logging.error("Modem is in error state.") |
| 255 | sys.exit(1) |
| 256 | |
| 257 | report_status(cp_state, up_state, speedtest_ping) |
| 258 | time.sleep(CONF.report_interval) |
| 259 | |
| 260 | modem.close() |
| 261 | |
| 262 | |
| 263 | if __name__ == "__main__": |
| 264 | main() |