|
| 1 | +sources: |
| 2 | + bq-export: |
| 3 | + kind: "bigquery" |
| 4 | + project: "aceti-462716" |
| 5 | + |
| 6 | +tools: |
| 7 | + system_chaos_summary: |
| 8 | + kind: bigquery-sql |
| 9 | + source: bq-export |
| 10 | + description: Aggregated view of all chaos logs in the system. |
| 11 | + parameters: [] |
| 12 | + statement: | |
| 13 | + SELECT |
| 14 | + severity, |
| 15 | + COUNT(*) AS count, |
| 16 | + COUNT(DISTINCT jsonPayload.agent_id) AS affected_agents, |
| 17 | + COUNT(DISTINCT jsonPayload.experiment_id) AS affected_experiments, |
| 18 | + STRING_AGG(DISTINCT jsonPayload.region) AS involved_regions |
| 19 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 20 | + GROUP BY severity |
| 21 | + ORDER BY count DESC; |
| 22 | +
|
| 23 | + frequent_failure_patterns: |
| 24 | + kind: bigquery-sql |
| 25 | + source: bq-export |
| 26 | + description: Top recurring failure messages in chaos logs. |
| 27 | + statement: | |
| 28 | + SELECT |
| 29 | + jsonPayload.message AS failure_message, |
| 30 | + COUNT(*) AS occurrences, |
| 31 | + STRING_AGG(DISTINCT jsonPayload.agent_id) AS involved_agents |
| 32 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 33 | + WHERE severity IN ('CRITICAL', 'ERROR') |
| 34 | + GROUP BY jsonPayload.message |
| 35 | + ORDER BY occurrences DESC |
| 36 | + LIMIT 10; |
| 37 | +
|
| 38 | + list_experiments: |
| 39 | + kind: bigquery-sql |
| 40 | + source: bq-export |
| 41 | + description: List all chaos experiments. |
| 42 | + statement: | |
| 43 | + SELECT |
| 44 | + jsonPayload.experiment_id AS id, |
| 45 | + MAX(jsonPayload.timestamp) AS last_seen, |
| 46 | + COUNT(*) AS total_logs, |
| 47 | + SUM(CASE WHEN severity IN ('CRITICAL', 'ERROR') THEN 1 ELSE 0 END) AS error_logs |
| 48 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 49 | + GROUP BY jsonPayload.experiment_id |
| 50 | + ORDER BY last_seen DESC; |
| 51 | +
|
| 52 | + get_experiment_by_id: |
| 53 | + kind: bigquery-sql |
| 54 | + source: bq-export |
| 55 | + description: Get all logs for a specific experiment. |
| 56 | + parameters: |
| 57 | + - name: experiment_id |
| 58 | + type: string |
| 59 | + description: The experiment ID (e.g., "exp0001"). |
| 60 | + statement: | |
| 61 | + SELECT * |
| 62 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 63 | + WHERE jsonPayload.experiment_id = ? |
| 64 | + ORDER BY TIMESTAMP(jsonPayload.timestamp) DESC; |
| 65 | +
|
| 66 | + list_anomalies: |
| 67 | + kind: bigquery-sql |
| 68 | + source: bq-export |
| 69 | + description: List critical or unusual chaos log entries (anomalies). |
| 70 | + statement: | |
| 71 | + SELECT |
| 72 | + jsonPayload.experiment_id, |
| 73 | + jsonPayload.message AS anomaly_type, |
| 74 | + severity, |
| 75 | + jsonPayload.timestamp AS detected_at |
| 76 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 77 | + WHERE severity IN ('CRITICAL', 'EMERGENCY', 'ALERT') |
| 78 | + ORDER BY TIMESTAMP(jsonPayload.timestamp) DESC; |
| 79 | +
|
| 80 | + recent_failed_experiments: |
| 81 | + kind: bigquery-sql |
| 82 | + source: bq-export |
| 83 | + description: List experiments with at least one CRITICAL error in the past 7 days. |
| 84 | + statement: | |
| 85 | + SELECT |
| 86 | + jsonPayload.experiment_id AS id, |
| 87 | + MAX(jsonPayload.timestamp) AS last_failure_time, |
| 88 | + COUNT(*) AS failure_count |
| 89 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 90 | + WHERE severity = 'CRITICAL' |
| 91 | + AND TIMESTAMP(jsonPayload.timestamp) >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 7 DAY) |
| 92 | + GROUP BY jsonPayload.experiment_id |
| 93 | + ORDER BY last_failure_time DESC; |
| 94 | +
|
| 95 | + recent_errors: |
| 96 | + kind: bigquery-sql |
| 97 | + source: bq-export |
| 98 | + description: Last 5 ERROR and CRITICAL logs. |
| 99 | + parameters: [] |
| 100 | + statement: | |
| 101 | + SELECT |
| 102 | + jsonPayload.experiment_id, |
| 103 | + jsonPayload.message AS failure_type, |
| 104 | + severity, |
| 105 | + jsonPayload.region, |
| 106 | + jsonPayload.timestamp |
| 107 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 108 | + WHERE severity IN ('ERROR', 'CRITICAL') |
| 109 | + ORDER BY TIMESTAMP(jsonPayload.timestamp) DESC |
| 110 | + LIMIT 5; |
| 111 | +
|
| 112 | + most_frequent_error_types: |
| 113 | + kind: bigquery-sql |
| 114 | + source: bq-export |
| 115 | + description: Most frequent error messages by count. |
| 116 | + parameters: [] |
| 117 | + statement: | |
| 118 | + SELECT |
| 119 | + jsonPayload.message AS failure_type, |
| 120 | + COUNT(*) AS count, |
| 121 | + STRING_AGG(DISTINCT jsonPayload.region) AS regions |
| 122 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 123 | + WHERE severity IN ('ERROR', 'CRITICAL') |
| 124 | + GROUP BY jsonPayload.message |
| 125 | + ORDER BY count DESC |
| 126 | + LIMIT 5; |
| 127 | +
|
| 128 | + errors_logs_grouped_by_severity: |
| 129 | + kind: bigquery-sql |
| 130 | + source: bq-export |
| 131 | + description: Errors grouped by severity. |
| 132 | + parameters: [] |
| 133 | + statement: | |
| 134 | + SELECT severity, COUNT(*) AS count |
| 135 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 136 | + WHERE severity IN ('ERROR', 'CRITICAL') |
| 137 | + GROUP BY severity; |
| 138 | +
|
| 139 | + critical_error_logs_grouped_by_region: |
| 140 | + kind: bigquery-sql |
| 141 | + source: bq-export |
| 142 | + description: Count of errors by region and severity. |
| 143 | + parameters: [] |
| 144 | + statement: | |
| 145 | + SELECT jsonPayload.region, severity, COUNT(*) AS count |
| 146 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 147 | + WHERE severity IN ('ERROR', 'CRITICAL') |
| 148 | + GROUP BY jsonPayload.region, severity; |
| 149 | +
|
| 150 | + total_error_logs: |
| 151 | + kind: bigquery-sql |
| 152 | + source: bq-export |
| 153 | + description: Total ERROR and CRITICAL logs. |
| 154 | + parameters: [] |
| 155 | + statement: | |
| 156 | + SELECT COUNT(*) AS total_error_logs |
| 157 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 158 | + WHERE severity IN ('ERROR', 'CRITICAL'); |
| 159 | +
|
| 160 | + agent_failure_rate: |
| 161 | + kind: bigquery-sql |
| 162 | + source: bq-export |
| 163 | + description: Failure rate per agent. |
| 164 | + statement: | |
| 165 | + SELECT |
| 166 | + jsonPayload.agent_id, |
| 167 | + COUNT(*) AS total_logs, |
| 168 | + SUM(CASE WHEN severity IN ('CRITICAL', 'ERROR') THEN 1 ELSE 0 END) AS error_count, |
| 169 | + ROUND(SAFE_DIVIDE(SUM(CASE WHEN severity IN ('CRITICAL', 'ERROR') THEN 1 ELSE 0 END), COUNT(*)), 2) AS failure_rate |
| 170 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 171 | + GROUP BY jsonPayload.agent_id |
| 172 | + ORDER BY failure_rate DESC; |
| 173 | +
|
| 174 | + user_impact_summary: |
| 175 | + kind: bigquery-sql |
| 176 | + source: bq-export |
| 177 | + description: Impact on users from error logs. |
| 178 | + statement: | |
| 179 | + SELECT |
| 180 | + jsonPayload.details.user_id, |
| 181 | + COUNT(*) AS total_events, |
| 182 | + COUNTIF(severity = 'CRITICAL') AS critical_events, |
| 183 | + COUNTIF(severity = 'ERROR') AS error_events |
| 184 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 185 | + WHERE severity IN ('CRITICAL', 'ERROR') |
| 186 | + GROUP BY jsonPayload.details.user_id |
| 187 | + ORDER BY critical_events DESC; |
| 188 | +
|
| 189 | + http_error_summary: |
| 190 | + kind: bigquery-sql |
| 191 | + source: bq-export |
| 192 | + description: Common HTTP failure patterns. |
| 193 | + statement: | |
| 194 | + SELECT |
| 195 | + jsonPayload.httprequest.requestmethod AS method, |
| 196 | + jsonPayload.httprequest.requesturl AS url, |
| 197 | + jsonPayload.status_code, |
| 198 | + COUNT(*) AS failures |
| 199 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 200 | + WHERE severity = 'ERROR' |
| 201 | + GROUP BY method, url, jsonPayload.status_code |
| 202 | + ORDER BY failures DESC |
| 203 | + LIMIT 10; |
| 204 | +
|
| 205 | + incidents_by_agent_and_experiment: |
| 206 | + kind: bigquery-sql |
| 207 | + source: bq-export |
| 208 | + description: Count of logs by agent and experiment, grouped by severity and region. |
| 209 | + statement: | |
| 210 | + SELECT |
| 211 | + jsonPayload.agent_id, |
| 212 | + jsonPayload.experiment_id, |
| 213 | + severity, |
| 214 | + jsonPayload.region, |
| 215 | + COUNT(*) AS log_count |
| 216 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 217 | + GROUP BY agent_id, experiment_id, severity, region |
| 218 | + ORDER BY log_count DESC |
| 219 | +
|
| 220 | + error_trends_by_agent: |
| 221 | + kind: bigquery-sql |
| 222 | + source: bq-export |
| 223 | + description: Error and critical log trends by agent over time. |
| 224 | + statement: | |
| 225 | + SELECT |
| 226 | + jsonPayload.agent_id, |
| 227 | + severity, |
| 228 | + FORMAT_TIMESTAMP('%Y-%m-%d %H:00:00', TIMESTAMP(jsonPayload.timestamp)) AS hour, |
| 229 | + COUNT(*) AS count |
| 230 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 231 | + WHERE severity IN ('ERROR', 'CRITICAL') |
| 232 | + GROUP BY agent_id, severity, hour |
| 233 | + ORDER BY hour DESC, count DESC |
| 234 | +
|
| 235 | + top_regions_by_error: |
| 236 | + kind: bigquery-sql |
| 237 | + source: bq-export |
| 238 | + description: Top regions by error and critical log count. |
| 239 | + statement: | |
| 240 | + SELECT |
| 241 | + jsonPayload.region, |
| 242 | + severity, |
| 243 | + COUNT(*) AS error_count |
| 244 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 245 | + WHERE severity IN ('ERROR', 'CRITICAL') |
| 246 | + GROUP BY region, severity |
| 247 | + ORDER BY error_count DESC |
| 248 | +
|
| 249 | + user_impact_by_experiment: |
| 250 | + kind: bigquery-sql |
| 251 | + source: bq-export |
| 252 | + description: User impact by experiment. |
| 253 | + statement: | |
| 254 | + SELECT |
| 255 | + jsonPayload.experiment_id, |
| 256 | + COUNT(DISTINCT jsonPayload.details.user_id) AS affected_users, |
| 257 | + COUNT(*) AS total_events |
| 258 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 259 | + WHERE jsonPayload.details.user_id IS NOT NULL |
| 260 | + GROUP BY experiment_id |
| 261 | + ORDER BY affected_users DESC |
| 262 | +
|
| 263 | + most_frequent_action_messages: |
| 264 | + kind: bigquery-sql |
| 265 | + source: bq-export |
| 266 | + description: Most frequent log messages (potential actions) by agent and experiment. |
| 267 | + statement: | |
| 268 | + SELECT |
| 269 | + jsonPayload.agent_id AS agent_id, |
| 270 | + jsonPayload.experiment_id AS experiment_id, |
| 271 | + jsonPayload.message AS action_message, |
| 272 | + COUNT(*) AS count |
| 273 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 274 | + WHERE jsonPayload.message IS NOT NULL |
| 275 | + GROUP BY agent_id, experiment_id, action_message |
| 276 | + ORDER BY count DESC |
| 277 | + LIMIT 20; |
| 278 | +
|
| 279 | + error_actions_by_agent: |
| 280 | + kind: bigquery-sql |
| 281 | + source: bq-export |
| 282 | + description: Count of ERROR and CRITICAL log messages by agent. |
| 283 | + statement: | |
| 284 | + SELECT |
| 285 | + jsonPayload.agent_id AS agent_id, |
| 286 | + COUNT(*) AS error_count |
| 287 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 288 | + WHERE severity IN ('ERROR', 'CRITICAL') |
| 289 | + GROUP BY agent_id |
| 290 | + ORDER BY error_count DESC; |
| 291 | +
|
| 292 | + recent_actions_by_experiment: |
| 293 | + kind: bigquery-sql |
| 294 | + source: bq-export |
| 295 | + description: Recent log messages for a specific experiment. |
| 296 | + parameters: |
| 297 | + - name: experiment_id |
| 298 | + type: string |
| 299 | + description: The experiment ID (e.g., "exp0001"). |
| 300 | + statement: | |
| 301 | + SELECT |
| 302 | + jsonPayload.agent_id, |
| 303 | + jsonPayload.message, |
| 304 | + severity, |
| 305 | + jsonPayload.timestamp |
| 306 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 307 | + WHERE jsonPayload.experiment_id = @experiment_id |
| 308 | + ORDER BY jsonPayload.timestamp DESC |
| 309 | + LIMIT 20; |
| 310 | +
|
| 311 | + user_impact_by_action: |
| 312 | + kind: bigquery-sql |
| 313 | + source: bq-export |
| 314 | + description: User impact summary by log message. |
| 315 | + statement: | |
| 316 | + SELECT |
| 317 | + jsonPayload.message AS action_message, |
| 318 | + COUNT(DISTINCT jsonPayload.details.user_id) AS affected_users, |
| 319 | + COUNT(*) AS total_events |
| 320 | + FROM `aceti-462716.bqexport.chaospilot_fake_logs_20250620` |
| 321 | + WHERE jsonPayload.details.user_id IS NOT NULL |
| 322 | + GROUP BY action_message |
| 323 | + ORDER BY affected_users DESC |
| 324 | + LIMIT 10; |
| 325 | +
|
| 326 | +toolsets: |
| 327 | + detector_toolset: |
| 328 | + - recent_errors |
| 329 | + - most_frequent_error_types |
| 330 | + - errors_logs_grouped_by_severity |
| 331 | + - critical_error_logs_grouped_by_region |
| 332 | + - total_error_logs |
| 333 | + - list_experiments |
| 334 | + - get_experiment_by_id |
| 335 | + - list_anomalies |
| 336 | + - recent_failed_experiments |
| 337 | + - system_chaos_summary |
| 338 | + - frequent_failure_patterns |
| 339 | + - agent_failure_rate |
| 340 | + - user_impact_summary |
| 341 | + - http_error_summary |
| 342 | + planner_toolset: |
| 343 | + - incidents_by_agent_and_experiment |
| 344 | + - error_trends_by_agent |
| 345 | + - top_regions_by_error |
| 346 | + - user_impact_by_experiment |
| 347 | + action_recommender_toolset: |
| 348 | + - most_frequent_action_messages |
| 349 | + - error_actions_by_agent |
| 350 | + - recent_actions_by_experiment |
| 351 | + - user_impact_by_action |
0 commit comments