|
| 1 | +mappings: |
| 2 | + # Airflow StatsD metrics mappings (https://airflow.apache.org/docs/apache-airflow/stable/logging-monitoring/metrics.html) |
| 3 | + # === Counters === |
| 4 | + - match: "(.+)\\.(.+)_start$" |
| 5 | + match_metric_type: counter |
| 6 | + name: "af_agg_job_start" |
| 7 | + match_type: regex |
| 8 | + labels: |
| 9 | + airflow_id: "$1" |
| 10 | + job_name: "$2" |
| 11 | + - match: "(.+)\\.(.+)_end$" |
| 12 | + match_metric_type: counter |
| 13 | + name: "af_agg_job_end" |
| 14 | + match_type: regex |
| 15 | + labels: |
| 16 | + airflow_id: "$1" |
| 17 | + job_name: "$2" |
| 18 | + - match: "(.+)\\.operator_failures_(.+)$" |
| 19 | + match_metric_type: counter |
| 20 | + name: "af_agg_operator_failures" |
| 21 | + match_type: regex |
| 22 | + labels: |
| 23 | + airflow_id: "$1" |
| 24 | + operator_name: "$2" |
| 25 | + - match: "(.+)\\.operator_successes_(.+)$" |
| 26 | + match_metric_type: counter |
| 27 | + name: "af_agg_operator_successes" |
| 28 | + match_type: regex |
| 29 | + labels: |
| 30 | + airflow_id: "$1" |
| 31 | + operator_name: "$2" |
| 32 | + - match: "*.ti_failures" |
| 33 | + match_metric_type: counter |
| 34 | + name: "af_agg_ti_failures" |
| 35 | + labels: |
| 36 | + airflow_id: "$1" |
| 37 | + - match: "*.ti_successes" |
| 38 | + match_metric_type: counter |
| 39 | + name: "af_agg_ti_successes" |
| 40 | + labels: |
| 41 | + airflow_id: "$1" |
| 42 | + - match: "*.zombies_killed" |
| 43 | + match_metric_type: counter |
| 44 | + name: "af_agg_zombies_killed" |
| 45 | + labels: |
| 46 | + airflow_id: "$1" |
| 47 | + - match: "*.scheduler_heartbeat" |
| 48 | + match_metric_type: counter |
| 49 | + name: "af_agg_scheduler_heartbeat" |
| 50 | + labels: |
| 51 | + airflow_id: "$1" |
| 52 | + - match: "*.dag_processing.processes" |
| 53 | + match_metric_type: counter |
| 54 | + name: "af_agg_dag_processing_processes" |
| 55 | + labels: |
| 56 | + airflow_id: "$1" |
| 57 | + - match: "*.scheduler.tasks.killed_externally" |
| 58 | + match_metric_type: counter |
| 59 | + name: "af_agg_scheduler_tasks_killed_externally" |
| 60 | + labels: |
| 61 | + airflow_id: "$1" |
| 62 | + - match: "*.scheduler.tasks.running" |
| 63 | + match_metric_type: counter |
| 64 | + name: "af_agg_scheduler_tasks_running" |
| 65 | + labels: |
| 66 | + airflow_id: "$1" |
| 67 | + - match: "*.scheduler.tasks.starving" |
| 68 | + match_metric_type: counter |
| 69 | + name: "af_agg_scheduler_tasks_starving" |
| 70 | + labels: |
| 71 | + airflow_id: "$1" |
| 72 | + - match: "*.scheduler.orphaned_tasks.cleared" |
| 73 | + match_metric_type: counter |
| 74 | + name: "af_agg_scheduler_orphaned_tasks_cleared" |
| 75 | + labels: |
| 76 | + airflow_id: "$1" |
| 77 | + - match: "*.scheduler.orphaned_tasks.adopted" |
| 78 | + match_metric_type: counter |
| 79 | + name: "af_agg_scheduler_orphaned_tasks_adopted" |
| 80 | + labels: |
| 81 | + airflow_id: "$1" |
| 82 | + - match: "*.scheduler.critical_section_busy" |
| 83 | + match_metric_type: counter |
| 84 | + name: "af_agg_scheduler_critical_section_busy" |
| 85 | + labels: |
| 86 | + airflow_id: "$1" |
| 87 | + - match: "*.sla_email_notification_failure" |
| 88 | + match_metric_type: counter |
| 89 | + name: "af_agg_sla_email_notification_failure" |
| 90 | + labels: |
| 91 | + airflow_id: "$1" |
| 92 | + - match: "*.ti.start.*.*" |
| 93 | + match_metric_type: counter |
| 94 | + name: "af_agg_ti_start" |
| 95 | + labels: |
| 96 | + airflow_id: "$1" |
| 97 | + dag_id: "$2" |
| 98 | + task_id: "$3" |
| 99 | + - match: "*.ti.finish.*.*.*" |
| 100 | + match_metric_type: counter |
| 101 | + name: "af_agg_ti_finish" |
| 102 | + labels: |
| 103 | + airflow_id: "$1" |
| 104 | + dag_id: "$2" |
| 105 | + task_id: "$3" |
| 106 | + state: "$4" |
| 107 | + - match: "*.dag.callback_exceptions" |
| 108 | + match_metric_type: counter |
| 109 | + name: "af_agg_dag_callback_exceptions" |
| 110 | + labels: |
| 111 | + airflow_id: "$1" |
| 112 | + - match: "*.celery.task_timeout_error" |
| 113 | + match_metric_type: counter |
| 114 | + name: "af_agg_celery_task_timeout_error" |
| 115 | + labels: |
| 116 | + airflow_id: "$1" |
| 117 | + |
| 118 | + # === Gauges === |
| 119 | + - match: "*.dagbag_size" |
| 120 | + match_metric_type: gauge |
| 121 | + name: "af_agg_dagbag_size" |
| 122 | + labels: |
| 123 | + airflow_id: "$1" |
| 124 | + - match: "*.dag_processing.import_errors" |
| 125 | + match_metric_type: gauge |
| 126 | + name: "af_agg_dag_processing_import_errors" |
| 127 | + labels: |
| 128 | + airflow_id: "$1" |
| 129 | + - match: "*.dag_processing.total_parse_time" |
| 130 | + match_metric_type: gauge |
| 131 | + name: "af_agg_dag_processing_total_parse_time" |
| 132 | + labels: |
| 133 | + airflow_id: "$1" |
| 134 | + - match: "*.dag_processing.last_runtime.*" |
| 135 | + match_metric_type: gauge |
| 136 | + name: "af_agg_dag_processing_last_runtime" |
| 137 | + labels: |
| 138 | + airflow_id: "$1" |
| 139 | + dag_file: "$2" |
| 140 | + - match: "*.dag_processing.last_run.seconds_ago.*" |
| 141 | + match_metric_type: gauge |
| 142 | + name: "af_agg_dag_processing_last_run_seconds" |
| 143 | + labels: |
| 144 | + airflow_id: "$1" |
| 145 | + dag_file: "$2" |
| 146 | + - match: "*.dag_processing.processor_timeouts" |
| 147 | + match_metric_type: gauge |
| 148 | + name: "af_agg_dag_processing_processor_timeouts" |
| 149 | + labels: |
| 150 | + airflow_id: "$1" |
| 151 | + - match: "*.executor.open_slots" |
| 152 | + match_metric_type: gauge |
| 153 | + name: "af_agg_executor_open_slots" |
| 154 | + labels: |
| 155 | + airflow_id: "$1" |
| 156 | + - match: "*.executor.queued_tasks" |
| 157 | + match_metric_type: gauge |
| 158 | + name: "af_agg_executor_queued_tasks" |
| 159 | + labels: |
| 160 | + airflow_id: "$1" |
| 161 | + - match: "*.executor.running_tasks" |
| 162 | + match_metric_type: gauge |
| 163 | + name: "af_agg_executor_running_tasks" |
| 164 | + labels: |
| 165 | + airflow_id: "$1" |
| 166 | + - match: "*.pool.open_slots.*" |
| 167 | + match_metric_type: gauge |
| 168 | + name: "af_agg_pool_open_slots" |
| 169 | + labels: |
| 170 | + airflow_id: "$1" |
| 171 | + pool_name: "$2" |
| 172 | + - match: "*.pool.queued_slots.*" |
| 173 | + match_metric_type: gauge |
| 174 | + name: "af_agg_pool_queued_slots" |
| 175 | + labels: |
| 176 | + airflow_id: "$1" |
| 177 | + pool_name: "$2" |
| 178 | + - match: "*.pool.running_slots.*" |
| 179 | + match_metric_type: gauge |
| 180 | + name: "af_agg_pool_running_slots" |
| 181 | + labels: |
| 182 | + airflow_id: "$1" |
| 183 | + pool_name: "$2" |
| 184 | + - match: "*.pool.starving_tasks.*" |
| 185 | + match_metric_type: gauge |
| 186 | + name: "af_agg_pool_starving_tasks" |
| 187 | + labels: |
| 188 | + airflow_id: "$1" |
| 189 | + pool_name: "$2" |
| 190 | + - match: "*.smart_sensor_operator.poked_tasks" |
| 191 | + match_metric_type: gauge |
| 192 | + name: "af_agg_smart_sensor_operator_poked_tasks" |
| 193 | + labels: |
| 194 | + airflow_id: "$1" |
| 195 | + - match: "*.smart_sensor_operator.poked_success" |
| 196 | + match_metric_type: gauge |
| 197 | + name: "af_agg_smart_sensor_operator_poked_success" |
| 198 | + labels: |
| 199 | + airflow_id: "$1" |
| 200 | + - match: "*.smart_sensor_operator.poked_exception" |
| 201 | + match_metric_type: gauge |
| 202 | + name: "af_agg_smart_sensor_operator_poked_exception" |
| 203 | + labels: |
| 204 | + airflow_id: "$1" |
| 205 | + - match: "*.smart_sensor_operator.exception_failures" |
| 206 | + match_metric_type: gauge |
| 207 | + name: "af_agg_smart_sensor_operator_exception_failures" |
| 208 | + labels: |
| 209 | + airflow_id: "$1" |
| 210 | + - match: "*.smart_sensor_operator.infra_failures" |
| 211 | + match_metric_type: gauge |
| 212 | + name: "af_agg_smart_sensor_operator_infra_failures" |
| 213 | + labels: |
| 214 | + airflow_id: "$1" |
| 215 | + |
| 216 | + # === Timers === |
| 217 | + - match: "*.dagrun.dependency-check.*" |
| 218 | + match_metric_type: observer |
| 219 | + name: "af_agg_dagrun_dependency_check" |
| 220 | + labels: |
| 221 | + airflow_id: "$1" |
| 222 | + dag_id: "$2" |
| 223 | + - match: "*.dag.*.*.duration" |
| 224 | + match_metric_type: observer |
| 225 | + name: "af_agg_dag_task_duration" |
| 226 | + labels: |
| 227 | + airflow_id: "$1" |
| 228 | + dag_id: "$2" |
| 229 | + task_id: "$3" |
| 230 | + - match: "*.dag_processing.last_duration.*" |
| 231 | + match_metric_type: observer |
| 232 | + name: "af_agg_dag_processing_duration" |
| 233 | + labels: |
| 234 | + airflow_id: "$1" |
| 235 | + dag_file: "$2" |
| 236 | + - match: "*.dagrun.duration.success.*" |
| 237 | + match_metric_type: observer |
| 238 | + name: "af_agg_dagrun_duration_success" |
| 239 | + labels: |
| 240 | + airflow_id: "$1" |
| 241 | + dag_id: "$2" |
| 242 | + - match: "*.dagrun.duration.failed.*" |
| 243 | + match_metric_type: observer |
| 244 | + name: "af_agg_dagrun_duration_failed" |
| 245 | + labels: |
| 246 | + airflow_id: "$1" |
| 247 | + dag_id: "$2" |
| 248 | + - match: "*.dagrun.schedule_delay.*" |
| 249 | + match_metric_type: observer |
| 250 | + name: "af_agg_dagrun_schedule_delay" |
| 251 | + labels: |
| 252 | + airflow_id: "$1" |
| 253 | + dag_id: "$2" |
| 254 | + - match: "*.scheduler.critical_section_duration" |
| 255 | + match_metric_type: observer |
| 256 | + name: "af_agg_scheduler_critical_section_duration" |
| 257 | + labels: |
| 258 | + airflow_id: "$1" |
| 259 | + - match: "*.dagrun.*.first_task_scheduling_delay" |
| 260 | + match_metric_type: observer |
| 261 | + name: "af_agg_dagrun_first_task_scheduling_delay" |
| 262 | + labels: |
| 263 | + airflow_id: "$1" |
| 264 | + dag_id: "$2" |
0 commit comments