Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 1 | #!/usr/bin/env python |
| 2 | |
| 3 | # Copyright 2020-present Open Networking Foundation |
| 4 | # |
Hyunsun Moon | 200eba5 | 2021-04-05 21:31:54 -0700 | [diff] [blame] | 5 | # SPDX-License-Identifier: LicenseRef-ONF-Member-Only-1.0 |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 6 | |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 7 | import os |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 8 | import time |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 9 | import datetime |
| 10 | import pytz |
| 11 | import threading |
| 12 | from icalevents.icalevents import events |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 13 | from flask import Flask, jsonify, abort, request, Response |
| 14 | import prometheus_client as prom |
Andy Bavier | 2d60fc5 | 2021-05-04 16:13:39 -0700 | [diff] [blame] | 15 | import jsonschema |
Andy Bavier | 55dc587 | 2021-05-05 11:31:42 -0700 | [diff] [blame] | 16 | from logging.config import dictConfig |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 17 | |
Andy Bavier | 8a5c987 | 2020-10-21 13:17:53 -0700 | [diff] [blame] | 18 | # URL of maintenance calendar |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 19 | SECRET_ICAL_URL = os.environ.get("SECRET_ICAL_URL") |
Andy Bavier | 8a5c987 | 2020-10-21 13:17:53 -0700 | [diff] [blame] | 20 | |
| 21 | # Aether environment that the server is monitoring (e.g., "production") |
| 22 | # To schedule downtime, postfix the cluster name with the env: "ace-tucson-production" |
| 23 | AETHER_ENV = os.environ.get("AETHER_ENV", "production") |
| 24 | |
| 25 | # Move to "no result" status if we don't hear from agent for this many seconds |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 26 | NO_RESULT_THRESHOLD = 720 |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 27 | |
Andy Bavier | 55dc587 | 2021-05-05 11:31:42 -0700 | [diff] [blame] | 28 | dictConfig({ |
| 29 | 'version': 1, |
| 30 | 'formatters': {'default': { |
Andy Bavier | 41fef77 | 2021-05-12 13:48:42 -0700 | [diff] [blame] | 31 | 'format': '%(levelname)s %(message)s', |
Andy Bavier | 55dc587 | 2021-05-05 11:31:42 -0700 | [diff] [blame] | 32 | }}, |
| 33 | 'handlers': {'wsgi': { |
| 34 | 'class': 'logging.StreamHandler', |
| 35 | 'stream': 'ext://flask.logging.wsgi_errors_stream', |
| 36 | 'formatter': 'default' |
| 37 | }}, |
| 38 | 'root': { |
| 39 | 'level': 'INFO', |
| 40 | 'handlers': ['wsgi'] |
| 41 | } |
| 42 | }) |
| 43 | |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 44 | app = Flask(__name__) |
Andy Bavier | 2d60fc5 | 2021-05-04 16:13:39 -0700 | [diff] [blame] | 45 | |
| 46 | edgeSchema = { |
| 47 | "type": "object", |
| 48 | "properties": { |
| 49 | "name": {"type": "string"}, |
| 50 | "status": { |
| 51 | "type": "object", |
| 52 | "properties": { |
| 53 | "control_plane": {"type": "string"}, |
| 54 | "user_plane": {"type": "string"} |
| 55 | }, |
| 56 | "required": ["control_plane", "user_plane"] |
| 57 | }, |
| 58 | "speedtest": { |
| 59 | "type": "object", |
| 60 | "properties": { |
| 61 | "ping": { |
| 62 | "type": "object", |
| 63 | "properties": { |
| 64 | "dns": { |
| 65 | "type": "object", |
| 66 | "properties": { |
| 67 | "min": {"type": "number"}, |
| 68 | "avg": {"type": "number"}, |
| 69 | "max": {"type": "number"}, |
| 70 | "stddev": {"type": "number"} |
| 71 | }, |
| 72 | "required": ["min", "avg", "max", "stddev"] |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 73 | }, |
| 74 | "iperf_server": { |
| 75 | "type": "object", |
| 76 | "properties": { |
| 77 | "min": {"type": "number"}, |
| 78 | "avg": {"type": "number"}, |
| 79 | "max": {"type": "number"}, |
| 80 | "stddev": {"type": "number"} |
| 81 | }, |
| 82 | "required": ["min", "avg", "max", "stddev"] |
Andy Bavier | 2d60fc5 | 2021-05-04 16:13:39 -0700 | [diff] [blame] | 83 | } |
| 84 | } |
| 85 | }, |
| 86 | "iperf": { |
| 87 | "type": "object", |
| 88 | "properties": { |
| 89 | "cluster": { |
| 90 | "type": "object", |
| 91 | "properties": { |
| 92 | "downlink": {"type": "number"}, |
| 93 | "uplink": {"type": "number"} |
| 94 | }, |
| 95 | "required": ["downlink", "uplink"] |
| 96 | } |
| 97 | } |
| 98 | } |
| 99 | } |
| 100 | }, |
| 101 | "signal_quality": { |
| 102 | "type": "object", |
| 103 | "properties": { |
| 104 | "rsrq": {"type": "number"}, |
| 105 | "rsrp": {"type": "number"} |
| 106 | }, |
| 107 | "required": ["rsrq", "rsrp"] |
| 108 | } |
| 109 | }, |
| 110 | "required": ["name", "status"] |
| 111 | } |
| 112 | |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 113 | edges = [ |
| 114 | { |
Andy Bavier | 8a5c987 | 2020-10-21 13:17:53 -0700 | [diff] [blame] | 115 | 'name': 'ace-example', |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 116 | 'status': { |
| 117 | 'control_plane': 'connected', |
| 118 | 'user_plane': 'connected' |
| 119 | }, |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 120 | 'speedtest': { |
| 121 | 'ping': { |
| 122 | 'dns': { |
Jeremy Ronquillo | a944fbc | 2021-03-30 10:57:45 -0700 | [diff] [blame] | 123 | 'min': 0.0, |
| 124 | 'avg': 0.0, |
| 125 | 'max': 0.0, |
| 126 | 'stddev': 0.0 |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 127 | }, |
| 128 | 'iperf_server': { |
| 129 | 'min': 0.0, |
| 130 | 'avg': 0.0, |
| 131 | 'max': 0.0, |
| 132 | 'stddev': 0.0 |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 133 | } |
Jeremy Ronquillo | c743462 | 2021-04-08 21:06:00 -0700 | [diff] [blame] | 134 | }, |
| 135 | 'iperf': { |
| 136 | 'cluster': { |
| 137 | 'downlink': 0.0, |
| 138 | 'uplink': 0.0 |
| 139 | } |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 140 | } |
| 141 | }, |
Hyunsun Moon | 200eba5 | 2021-04-05 21:31:54 -0700 | [diff] [blame] | 142 | 'signal_quality': { |
| 143 | 'rsrq': 0, |
| 144 | 'rsrp': 0 |
| 145 | }, |
| 146 | 'last_update': time.time() |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 147 | } |
| 148 | ] |
| 149 | |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 150 | status_codes = { |
| 151 | "no result": -2, |
| 152 | "error": -1, |
| 153 | "disconnected": 0, |
| 154 | "connecting": 1, |
| 155 | "connected": 2 |
| 156 | } |
| 157 | |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 158 | room_mapping = { |
Andy Bavier | 0423cbd | 2020-10-23 10:50:29 -0700 | [diff] [blame] | 159 | "ace-menlo-pixel-production": "(Compute)-MP-1-Aether Production", |
| 160 | "ace-menlo-staging": "(Compute)-MP-1-Aether Staging" |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 161 | } |
| 162 | |
Andy Bavier | 5b4e28f | 2021-03-09 15:48:20 -0700 | [diff] [blame] | 163 | # Legacy test status metrics, reporting a status code between -2 and 2 |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 164 | cp_status = prom.Gauge("aetheredge_status_control_plane", "Control plane status code", ["name"]) |
| 165 | up_status = prom.Gauge("aetheredge_status_user_plane", "User plane status code", ["name"]) |
Andy Bavier | 5b4e28f | 2021-03-09 15:48:20 -0700 | [diff] [blame] | 166 | |
| 167 | # Simplified binary test result metrics |
Andy Bavier | 3c7b78d | 2021-03-11 14:16:43 -0700 | [diff] [blame] | 168 | e2e_tests_ok = prom.Gauge("aetheredge_e2e_tests_ok", "Last connect and ping test both passed", ["name"]) |
Andy Bavier | a0c40aa | 2021-03-10 12:09:12 -0700 | [diff] [blame] | 169 | connect_test_ok = prom.Gauge("aetheredge_connect_test_ok", "Last connect test passed", ["name"]) |
Andy Bavier | a0c40aa | 2021-03-10 12:09:12 -0700 | [diff] [blame] | 170 | ping_test_ok = prom.Gauge("aetheredge_ping_test_ok", "Last ping test passed", ["name"]) |
Andy Bavier | 3c7b78d | 2021-03-11 14:16:43 -0700 | [diff] [blame] | 171 | e2e_tests_down = prom.Gauge("aetheredge_e2e_tests_down", "E2E tests not reporting", ["name"]) |
Andy Bavier | 5b4e28f | 2021-03-09 15:48:20 -0700 | [diff] [blame] | 172 | |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 173 | # Speedtest dns ping metrics |
| 174 | ping_dns_min = prom.Gauge("aetheredge_ping_dns_test_min","Last ping test to dns minimum value",["name"]) |
| 175 | ping_dns_avg = prom.Gauge("aetheredge_ping_dns_test_avg","Last ping test to dns average",["name"]) |
| 176 | ping_dns_max = prom.Gauge("aetheredge_ping_dns_test_max","Last ping test to dns maximum value",["name"]) |
| 177 | ping_dns_stddev = prom.Gauge("aetheredge_ping_dns_test_stddev","Last ping test to dns standard deviation",["name"]) |
| 178 | |
| 179 | # Speedtest iperf server ping metrics |
| 180 | ping_iperf_server_min = prom.Gauge("aetheredge_ping_iperf_server_test_min","Last ping test to iperf_server minimum value",["name"]) |
| 181 | ping_iperf_server_avg = prom.Gauge("aetheredge_ping_iperf_server_test_avg","Last ping test to iperf_server average",["name"]) |
| 182 | ping_iperf_server_max = prom.Gauge("aetheredge_ping_iperf_server_test_max","Last ping test to iperf_server maximum value",["name"]) |
| 183 | ping_iperf_server_stddev = prom.Gauge("aetheredge_ping_iperf_server_test_stddev","Last ping test to iperf_server standard deviation",["name"]) |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 184 | |
Jeremy Ronquillo | c743462 | 2021-04-08 21:06:00 -0700 | [diff] [blame] | 185 | # Speedtest iperf metrics |
| 186 | iperf_cluster_downlink = prom.Gauge("aetheredge_iperf_cluster_downlink_test","Last iperf test downlink result",["name"]) |
| 187 | iperf_cluster_uplink = prom.Gauge("aetheredge_iperf_cluster_uplink_test","Last iperf test downlink result",["name"]) |
| 188 | |
Hyunsun Moon | 200eba5 | 2021-04-05 21:31:54 -0700 | [diff] [blame] | 189 | # Signal quality metrics in CESQ format not dB |
| 190 | # RSRQ: >=53 excellent, 43 ~ 53 good, 33 ~ 43 mid, <=33 bad, 0 no signal |
| 191 | # RSRP: >=20 excellent, 10 ~ 20 good, 0 ~ 10 mid, 0 no signal |
| 192 | signal_quality_rsrq = prom.Gauge("aetheredge_signal_quality_rsrq", "Quality of the received signal", ["name"]) |
| 193 | signal_quality_rsrp = prom.Gauge("aetheredge_signal_quality_rsrp", "Power of the received signal", ["name"]) |
| 194 | |
Andy Bavier | 5b4e28f | 2021-03-09 15:48:20 -0700 | [diff] [blame] | 195 | # Other metrics |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 196 | last_update = prom.Gauge("aetheredge_last_update", "Last reported test result", ["name"]) |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 197 | maint_window = prom.Gauge("aetheredge_in_maintenance_window", "Currently in a maintenance window", ["name"]) |
| 198 | |
| 199 | def is_my_event(event, name): |
| 200 | for field in ["summary", "location", "description"]: |
Andy Bavier | 8a5c987 | 2020-10-21 13:17:53 -0700 | [diff] [blame] | 201 | fullname = name |
| 202 | if name.startswith("ace-"): |
| 203 | fullname = "%s-%s" % (name, AETHER_ENV) |
| 204 | if fullname in getattr(event, field, ""): |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 205 | return True |
Andy Bavier | 0423cbd | 2020-10-23 10:50:29 -0700 | [diff] [blame] | 206 | if fullname in room_mapping and room_mapping[fullname] in getattr(event, field, ""): |
| 207 | return True |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 208 | return False |
| 209 | |
Andy Bavier | c41cf0c | 2020-09-02 14:49:21 -0700 | [diff] [blame] | 210 | def is_naive_datetime(d): |
| 211 | return d.tzinfo is None or d.tzinfo.utcoffset(d) is None |
| 212 | |
| 213 | def process_all_day_events(es): |
| 214 | for event in es: |
| 215 | if event.all_day: |
| 216 | # All day events have naive datetimes, which breaks comparisons |
| 217 | pacific = pytz.timezone('US/Pacific') |
| 218 | if is_naive_datetime(event.start): |
| 219 | event.start = pacific.localize(event.start) |
| 220 | if is_naive_datetime(event.end): |
| 221 | event.end = pacific.localize(event.end) |
| 222 | |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 223 | def in_maintenance_window(events, name, now): |
| 224 | for event in events: |
| 225 | if event.start < now and event.end > now: |
| 226 | if is_my_event(event, name): |
| 227 | return True |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 228 | return False |
| 229 | |
| 230 | def pull_maintenance_events(): |
| 231 | while(True): |
| 232 | now = datetime.datetime.now(pytz.utc) |
| 233 | try: |
| 234 | es = events(SECRET_ICAL_URL, start = now) |
Andy Bavier | c41cf0c | 2020-09-02 14:49:21 -0700 | [diff] [blame] | 235 | process_all_day_events(es) |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 236 | except Exception as e: |
Andy Bavier | 55dc587 | 2021-05-05 11:31:42 -0700 | [diff] [blame] | 237 | app.logger.error(e) |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 238 | else: |
| 239 | for edge in edges: |
| 240 | if 'maintenance' not in edge: |
| 241 | edge['maintenance'] = {} |
| 242 | edge['maintenance']['in_window'] = in_maintenance_window(es, edge['name'], now) |
| 243 | edge['maintenance']['last_update'] = time.time() |
| 244 | time.sleep(60) |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 245 | |
| 246 | def time_out_stale_results(): |
| 247 | for edge in edges: |
| 248 | time_elapsed = time.time() - edge["last_update"] |
| 249 | if time_elapsed > NO_RESULT_THRESHOLD: |
| 250 | edge['status']['control_plane'] = "no result" |
| 251 | edge['status']['user_plane'] = "no result" |
Jeremy Ronquillo | a944fbc | 2021-03-30 10:57:45 -0700 | [diff] [blame] | 252 | edge['speedtest']['ping']['dns'] = {'min': 0.0, |
| 253 | 'avg': 0.0, |
| 254 | 'max': 0.0, |
| 255 | 'stddev': 0.0} |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 256 | edge['speedtest']['ping']['iperf_server'] = {'min': 0.0, |
| 257 | 'avg': 0.0, |
| 258 | 'max': 0.0, |
| 259 | 'stddev': 0.0} |
Jeremy Ronquillo | 56b0a1e | 2021-04-09 00:26:18 -0700 | [diff] [blame] | 260 | edge['speedtest']['iperf'] = {'cluster': { |
| 261 | 'downlink': 0.0, |
| 262 | 'uplink': 0.0 |
| 263 | } |
| 264 | } |
Hyunsun Moon | 200eba5 | 2021-04-05 21:31:54 -0700 | [diff] [blame] | 265 | edge.pop('signal_quality', None) |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 266 | |
Andy Bavier | e47157d | 2020-12-11 14:13:12 -0700 | [diff] [blame] | 267 | def remove_edge_from_metrics(name): |
| 268 | try: |
| 269 | cp_status.remove(name) |
| 270 | up_status.remove(name) |
| 271 | last_update.remove(name) |
Andy Bavier | 3c7b78d | 2021-03-11 14:16:43 -0700 | [diff] [blame] | 272 | e2e_tests_ok.remove(name) |
Andy Bavier | a0c40aa | 2021-03-10 12:09:12 -0700 | [diff] [blame] | 273 | connect_test_ok.remove(name) |
Andy Bavier | a0c40aa | 2021-03-10 12:09:12 -0700 | [diff] [blame] | 274 | ping_test_ok.remove(name) |
Andy Bavier | 3c7b78d | 2021-03-11 14:16:43 -0700 | [diff] [blame] | 275 | e2e_tests_down.remove(name) |
Andy Bavier | 5b4e28f | 2021-03-09 15:48:20 -0700 | [diff] [blame] | 276 | except: |
| 277 | pass |
| 278 | |
| 279 | try: |
Hyunsun Moon | 200eba5 | 2021-04-05 21:31:54 -0700 | [diff] [blame] | 280 | ping_dns_min.remove(name) |
| 281 | ping_dns_avg.remove(name) |
| 282 | ping_dns_max.remove(name) |
| 283 | ping_dns_stddev.remove(name) |
| 284 | except: |
| 285 | pass |
| 286 | |
| 287 | try: |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 288 | ping_iperf_server_min.remove(name) |
| 289 | ping_iperf_server_avg.remove(name) |
| 290 | ping_iperf_server_max.remove(name) |
| 291 | ping_iperf_server_stddev.remove(name) |
| 292 | except: |
| 293 | pass |
| 294 | |
| 295 | try: |
Jeremy Ronquillo | c743462 | 2021-04-08 21:06:00 -0700 | [diff] [blame] | 296 | iperf_cluster_downlink.remove(name) |
| 297 | iperf_cluster_uplink.remove(name) |
| 298 | except: |
| 299 | pass |
| 300 | |
| 301 | try: |
Hyunsun Moon | 200eba5 | 2021-04-05 21:31:54 -0700 | [diff] [blame] | 302 | signal_quality_rsrq.remove(name) |
| 303 | signal_quality_rsrp.remove(name) |
| 304 | except: |
| 305 | pass |
| 306 | |
| 307 | try: |
Andy Bavier | e47157d | 2020-12-11 14:13:12 -0700 | [diff] [blame] | 308 | maint_window.remove(name) |
| 309 | except: |
| 310 | pass |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 311 | |
| 312 | @app.route('/edges/metrics', methods=['GET']) |
| 313 | def get_prometheus_metrics(): |
| 314 | res = [] |
| 315 | time_out_stale_results() |
| 316 | for edge in edges: |
Andy Bavier | 8a5c987 | 2020-10-21 13:17:53 -0700 | [diff] [blame] | 317 | if edge['name'] == "ace-example": |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 318 | continue |
| 319 | |
Andy Bavier | 3c7b78d | 2021-03-11 14:16:43 -0700 | [diff] [blame] | 320 | connect_status = edge['status']['control_plane'] |
| 321 | ping_status = edge['status']['user_plane'] |
| 322 | |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 323 | # Add ping dns latency results if available |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 324 | if edge['speedtest']['ping']['dns']['avg']: |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 325 | ping_dns_min.labels(edge['name']).set(edge['speedtest']['ping']['dns']['min']) |
| 326 | ping_dns_avg.labels(edge['name']).set(edge['speedtest']['ping']['dns']['avg']) |
| 327 | ping_dns_max.labels(edge['name']).set(edge['speedtest']['ping']['dns']['max']) |
| 328 | ping_dns_stddev.labels(edge['name']).set(edge['speedtest']['ping']['dns']['stddev']) |
Jeremy Ronquillo | c743462 | 2021-04-08 21:06:00 -0700 | [diff] [blame] | 329 | |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 330 | # Add ping iperf_server latency results if available |
| 331 | if edge['speedtest']['ping']['iperf_server']['avg']: |
| 332 | ping_iperf_server_min.labels(edge['name']).set(edge['speedtest']['ping']['iperf_server']['min']) |
| 333 | ping_iperf_server_avg.labels(edge['name']).set(edge['speedtest']['ping']['iperf_server']['avg']) |
| 334 | ping_iperf_server_max.labels(edge['name']).set(edge['speedtest']['ping']['iperf_server']['max']) |
| 335 | ping_iperf_server_stddev.labels(edge['name']).set(edge['speedtest']['ping']['iperf_server']['stddev']) |
| 336 | |
| 337 | # Add iperf bandwidth results if available |
Jeremy Ronquillo | c743462 | 2021-04-08 21:06:00 -0700 | [diff] [blame] | 338 | if edge['speedtest']['iperf']['cluster']['downlink']: |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 339 | iperf_cluster_downlink.labels(edge['name']).set(edge['speedtest']['iperf']['cluster']['downlink']) |
| 340 | iperf_cluster_uplink.labels(edge['name']).set(edge['speedtest']['iperf']['cluster']['uplink']) |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 341 | |
Andy Bavier | 3c7b78d | 2021-03-11 14:16:43 -0700 | [diff] [blame] | 342 | cp_status.labels(edge['name']).set(status_codes[connect_status]) |
| 343 | up_status.labels(edge['name']).set(status_codes[ping_status]) |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 344 | |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 345 | last_update.labels(edge['name']).set(edge['last_update']) |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 346 | if 'maintenance' in edge: |
| 347 | maint_window.labels(edge['name']).set(int(edge['maintenance']['in_window'])) |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 348 | |
Andy Bavier | a0c40aa | 2021-03-10 12:09:12 -0700 | [diff] [blame] | 349 | connect_test_ok.labels(edge['name']).set(0) |
Andy Bavier | a0c40aa | 2021-03-10 12:09:12 -0700 | [diff] [blame] | 350 | ping_test_ok.labels(edge['name']).set(0) |
Andy Bavier | 3c7b78d | 2021-03-11 14:16:43 -0700 | [diff] [blame] | 351 | e2e_tests_ok.labels(edge['name']).set(0) |
| 352 | e2e_tests_down.labels(edge['name']).set(0) |
| 353 | |
| 354 | if connect_status in ["error", "no result"] or ping_status in ["error", "no result"]: |
| 355 | e2e_tests_down.labels(edge['name']).set(1) |
| 356 | else: |
| 357 | if connect_status == "connected": |
| 358 | connect_test_ok.labels(edge['name']).set(1) |
| 359 | if ping_status == "connected": |
| 360 | ping_test_ok.labels(edge['name']).set(1) |
| 361 | if connect_status == "connected" and ping_status == "connected": |
| 362 | e2e_tests_ok.labels(edge['name']).set(1) |
Andy Bavier | 5b4e28f | 2021-03-09 15:48:20 -0700 | [diff] [blame] | 363 | |
Hyunsun Moon | 200eba5 | 2021-04-05 21:31:54 -0700 | [diff] [blame] | 364 | if 'signal_quality' in edge.keys(): |
| 365 | signal_quality_rsrq.labels(edge['name']).set(edge['signal_quality']['rsrq']) |
| 366 | signal_quality_rsrp.labels(edge['name']).set(edge['signal_quality']['rsrp']) |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 367 | |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 368 | res.append(prom.generate_latest(cp_status)) |
| 369 | res.append(prom.generate_latest(up_status)) |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 370 | res.append(prom.generate_latest(ping_dns_min)) |
| 371 | res.append(prom.generate_latest(ping_dns_avg)) |
| 372 | res.append(prom.generate_latest(ping_dns_max)) |
| 373 | res.append(prom.generate_latest(ping_dns_stddev)) |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 374 | res.append(prom.generate_latest(ping_iperf_server_min)) |
| 375 | res.append(prom.generate_latest(ping_iperf_server_avg)) |
| 376 | res.append(prom.generate_latest(ping_iperf_server_max)) |
| 377 | res.append(prom.generate_latest(ping_iperf_server_stddev)) |
Jeremy Ronquillo | c743462 | 2021-04-08 21:06:00 -0700 | [diff] [blame] | 378 | res.append(prom.generate_latest(iperf_cluster_downlink)) |
| 379 | res.append(prom.generate_latest(iperf_cluster_uplink)) |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 380 | res.append(prom.generate_latest(last_update)) |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 381 | res.append(prom.generate_latest(maint_window)) |
Andy Bavier | a0c40aa | 2021-03-10 12:09:12 -0700 | [diff] [blame] | 382 | res.append(prom.generate_latest(connect_test_ok)) |
Andy Bavier | a0c40aa | 2021-03-10 12:09:12 -0700 | [diff] [blame] | 383 | res.append(prom.generate_latest(ping_test_ok)) |
Andy Bavier | 3c7b78d | 2021-03-11 14:16:43 -0700 | [diff] [blame] | 384 | res.append(prom.generate_latest(e2e_tests_ok)) |
| 385 | res.append(prom.generate_latest(e2e_tests_down)) |
Hyunsun Moon | 200eba5 | 2021-04-05 21:31:54 -0700 | [diff] [blame] | 386 | res.append(prom.generate_latest(signal_quality_rsrq)) |
| 387 | res.append(prom.generate_latest(signal_quality_rsrp)) |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 388 | |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 389 | return Response(res, mimetype="text/plain") |
| 390 | |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 391 | |
| 392 | @app.route('/edges/healthz', methods=['GET']) |
| 393 | def get_health(): |
| 394 | return {'message': 'healthy'} |
| 395 | |
| 396 | |
| 397 | @app.route('/edges', methods=['GET']) |
| 398 | def get_edges(): |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 399 | time_out_stale_results() |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 400 | return jsonify({'edges': edges}) |
| 401 | |
| 402 | |
| 403 | @app.route('/edges/<string:name>', methods=['GET']) |
| 404 | def get_edge(name): |
Andy Bavier | 4021a2f | 2020-07-29 12:39:47 -0700 | [diff] [blame] | 405 | time_out_stale_results() |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 406 | edge = [edge for edge in edges if edge['name'] == name] |
| 407 | if len(edge) == 0: |
| 408 | abort(404) |
| 409 | return jsonify({'edge': edge[0]}) |
| 410 | |
| 411 | |
| 412 | @app.route('/edges', methods=['POST']) |
Andy Bavier | f872e9a | 2021-03-22 12:06:25 -0700 | [diff] [blame] | 413 | @app.route('/testresults', methods=['POST']) |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 414 | def create_or_update_edge(): |
Andy Bavier | 2d60fc5 | 2021-05-04 16:13:39 -0700 | [diff] [blame] | 415 | try: |
| 416 | jsonschema.validate(instance=request.json, schema=edgeSchema) |
| 417 | except jsonschema.exceptions.ValidationError as err: |
Andy Bavier | 55dc587 | 2021-05-05 11:31:42 -0700 | [diff] [blame] | 418 | app.logger.warn(err) |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 419 | abort(400) |
| 420 | |
| 421 | req_edge = { |
| 422 | 'name': request.json['name'], |
| 423 | 'status': { |
| 424 | 'control_plane': request.json['status']['control_plane'], |
| 425 | 'user_plane': request.json['status']['user_plane'] |
| 426 | }, |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 427 | 'speedtest': { |
| 428 | 'ping': { |
| 429 | 'dns': { |
Jeremy Ronquillo | a944fbc | 2021-03-30 10:57:45 -0700 | [diff] [blame] | 430 | 'min': 0.0, |
| 431 | 'avg': 0.0, |
| 432 | 'max': 0.0, |
| 433 | 'stddev': 0.0 |
Jeremy Ronquillo | 6e352b7 | 2021-06-08 10:33:25 -0700 | [diff] [blame^] | 434 | }, |
| 435 | 'iperf_server': { |
| 436 | 'min': 0.0, |
| 437 | 'avg': 0.0, |
| 438 | 'max': 0.0, |
| 439 | 'stddev': 0.0 |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 440 | } |
Jeremy Ronquillo | c743462 | 2021-04-08 21:06:00 -0700 | [diff] [blame] | 441 | }, |
| 442 | 'iperf': { |
| 443 | 'cluster': { |
| 444 | 'downlink': 0.0, |
| 445 | 'uplink': 0.0 |
| 446 | } |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 447 | } |
| 448 | }, |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 449 | 'last_update': time.time() |
| 450 | } |
| 451 | |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 452 | if 'speedtest' in request.json: |
Jeremy Ronquillo | c743462 | 2021-04-08 21:06:00 -0700 | [diff] [blame] | 453 | if 'ping' in request.json['speedtest']: |
| 454 | req_edge['speedtest']['ping'] = request.json['speedtest']['ping'] |
| 455 | if 'iperf' in request.json['speedtest']: |
| 456 | req_edge['speedtest']['iperf'] = request.json['speedtest']['iperf'] |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 457 | |
Hyunsun Moon | 200eba5 | 2021-04-05 21:31:54 -0700 | [diff] [blame] | 458 | if 'signal_quality' in request.json: |
| 459 | req_edge['signal_quality'] = request.json['signal_quality'] |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 460 | |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 461 | edge = [edge for edge in edges if edge['name'] == req_edge['name']] |
| 462 | if len(edge) == 0: |
Andy Bavier | 55dc587 | 2021-05-05 11:31:42 -0700 | [diff] [blame] | 463 | app.logger.info("new edge request " + req_edge['name']) |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 464 | edges.append(req_edge) |
| 465 | else: |
| 466 | edge[0]['status']['control_plane'] = req_edge['status']['control_plane'] |
| 467 | edge[0]['status']['user_plane'] = req_edge['status']['user_plane'] |
Jeremy Ronquillo | f420025 | 2021-02-13 16:11:04 -0800 | [diff] [blame] | 468 | edge[0]['speedtest']['ping'] = req_edge['speedtest']['ping'] |
Jeremy Ronquillo | 56b0a1e | 2021-04-09 00:26:18 -0700 | [diff] [blame] | 469 | edge[0]['speedtest']['iperf'] = req_edge['speedtest']['iperf'] |
Hyunsun Moon | 200eba5 | 2021-04-05 21:31:54 -0700 | [diff] [blame] | 470 | if 'signal_quality' in req_edge.keys(): |
| 471 | edge[0]['signal_quality'] = req_edge['signal_quality'] |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 472 | edge[0]['last_update'] = req_edge['last_update'] |
| 473 | |
| 474 | return jsonify({'edge': req_edge}), 201 |
| 475 | |
| 476 | |
Hyunsun Moon | 5f237ec | 2020-09-29 14:45:52 -0700 | [diff] [blame] | 477 | @app.route('/edges/<string:name>', methods=['DELETE']) |
Andy Bavier | f872e9a | 2021-03-22 12:06:25 -0700 | [diff] [blame] | 478 | @app.route('/testresults/<string:name>', methods=['DELETE']) |
Hyunsun Moon | 5f237ec | 2020-09-29 14:45:52 -0700 | [diff] [blame] | 479 | def delete_edge(name): |
Andy Bavier | 55dc587 | 2021-05-05 11:31:42 -0700 | [diff] [blame] | 480 | app.logger.info("delete edge request " + name) |
Hyunsun Moon | 5f237ec | 2020-09-29 14:45:52 -0700 | [diff] [blame] | 481 | result = False |
| 482 | for i in range(len(edges)): |
| 483 | if edges[i]['name'] == name: |
| 484 | del edges[i] |
Andy Bavier | e47157d | 2020-12-11 14:13:12 -0700 | [diff] [blame] | 485 | remove_edge_from_metrics(name) |
Hyunsun Moon | 5f237ec | 2020-09-29 14:45:52 -0700 | [diff] [blame] | 486 | result = True |
| 487 | break |
| 488 | if not result: |
| 489 | abort(404) |
| 490 | return jsonify({'result': True}) |
| 491 | |
| 492 | |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 493 | if __name__ == '__main__': |
Andy Bavier | 8a5c987 | 2020-10-21 13:17:53 -0700 | [diff] [blame] | 494 | if SECRET_ICAL_URL and AETHER_ENV: |
Andy Bavier | 55dc587 | 2021-05-05 11:31:42 -0700 | [diff] [blame] | 495 | app.logger.info(" * Starting maintenance calendar polling thread (Aether env: %s)" % AETHER_ENV) |
Andy Bavier | 614af14 | 2020-08-07 14:49:56 -0700 | [diff] [blame] | 496 | t = threading.Thread(target=pull_maintenance_events) |
| 497 | t.start() |
Hyunsun Moon | f32ae9a | 2020-05-28 13:17:45 -0700 | [diff] [blame] | 498 | app.run(debug=True, host='0.0.0.0', port=80) |