Airflow monitoring

SkyWalking ingests Apache Airflow 3.x metrics from Airflow's native OpenTelemetry exporter via the OpenTelemetry receiver, aggregates them with MAL, and shows them in Horizon UI under Workflow Scheduler → Airflow.

Data flow

Airflow exports metrics via native OpenTelemetry (otel_on / OTEL_EXPORTER_OTLP_*) from scheduler and triggerer.
OpenTelemetry Collector receives OTLP metrics and forwards them to SkyWalking OAP.
OAP aggregates metrics with MAL.
Horizon UI displays them under Workflow Scheduler → Airflow.

Setup

1. Enable Airflow OpenTelemetry metrics

Install the OTel extra and point export at your Collector. See Airflow metrics documentation.

bash

pip install 'apache-airflow[otel]'

export OTEL_EXPORTER_OTLP_ENDPOINT=http://otel-collector:4318
export OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
export OTEL_RESOURCE_ATTRIBUTES=cluster=prod-airflow

Set AIRFLOW__METRICS__OTEL_ON=True on scheduler and triggerer.

Required OTLP resource attributes:

Attribute	Purpose
`cluster`	Names the Service (`airflow::{cluster}`)
`host.name`	Identifies scheduler or triggerer (UI Components tab)

2. OpenTelemetry Collector

Forward OTLP metrics to OAP. Example pipeline:

yaml

receivers:
  otlp:
    protocols:
      http:
        endpoint: 0.0.0.0:4318
      grpc:
        endpoint: 0.0.0.0:4317

processors:
  batch:

exporters:
  otlp:
    endpoint: oap:11800
    tls:
      insecure: true

service:
  pipelines:
    metrics:
      receivers: [otlp]
      processors: [batch]
      exporters: [otlp]

Full example: cluster/otel-collector-config.yaml. Do not hard-code service or instance names in Collector processors — derive them from Airflow's resource attributes.

3. SkyWalking OAP

Ensure airflow/* is in SW_OTEL_RECEIVER_ENABLED_OTEL_METRICS_RULES (enabled by default).

Entity model

SkyWalking entity	Mapping
Service	`airflow::{cluster}` from resource `cluster`
Instance (OAP) / Components (UI)	Scheduler or triggerer from `host.name`

Components vs SkyWalking Instance vs Airflow Task Instance

OAP stores scheduler/triggerer hosts as Instance; Horizon UI labels the tab Components so operators do not confuse it with an Airflow Task Instance (one task run inside a DAG run).

Term	Meaning
SkyWalking Service	One Airflow cluster (`airflow::{cluster}`)
Components (UI) / Instance (OAP)	Long-running scheduler or triggerer (`host.name`)
Airflow Task Instance	Single task execution — not on this dashboard

Service panels aggregate cluster-wide samples. Component panels are scoped per host.name.

On Kubernetes, run a Collector sidecar per monitored pod; Airflow pushes to localhost:4318 and the sidecar forwards to OAP. Set cluster via OTEL_RESOURCE_ATTRIBUTES; host.name comes from the pod hostname.

Supported metrics

MAL rules: otel-rules/airflow/airflow-service.yaml and airflow-instance.yaml. Asset counters use airflow.asset.* (Airflow 3.x only). Data source: Airflow native OpenTelemetry export.

Airflow Service Supported Metrics

Monitoring Panel	Unit	Metric Name	Description
Tasks Executable	count	meter_airflow_scheduler_tasks_executable	Tasks ready for execution
Running Tasks	count	meter_airflow_executor_running_tasks	Tasks currently running on executor
Queued Tasks	count	meter_airflow_executor_queued_tasks	Queued tasks on executor
Scheduled Slots	count	meter_airflow_pool_scheduled_slots	Scheduled but not yet running slots in pool
Executor Open Slots	count	meter_airflow_executor_open_slots	Open executor slots
DAG File Queue Size	count	meter_airflow_dag_file_queue_size	DAG files pending scan
DAG Import Errors	count	meter_airflow_dag_import_errors	DAG files that failed to parse
DAG Bag Size	count	meter_airflow_dagbag_size	DAGs found in the last scheduler scan
DAG Total Parse Time	seconds	meter_airflow_dag_total_parse_time	Time to scan and import queued DAG files
DAG File Refresh Errors	count/min	meter_airflow_dag_file_refresh_error	DAG file load failures per minute
Asset Updates	count/min	meter_airflow_asset_updates	Updated assets per minute

Airflow Instance Supported Metrics

Monitoring Panel	Unit	Metric Name	Description
Pool Open / Deferred / Running Slots	count	meter_airflow_instance_pool_open_slots, meter_airflow_instance_pool_deferred_slots, meter_airflow_instance_pool_running_slots	Pool capacity on the scheduler
Running Tasks / Scheduled Slots	count	meter_airflow_instance_executor_running_tasks, meter_airflow_instance_pool_scheduled_slots	Executor queue depth and pool slots waiting to run
Scheduler Heartbeat	count/min	meter_airflow_instance_scheduler_heartbeat	Scheduler heartbeats per minute
Executor Open / Queued Slots	count	meter_airflow_instance_executor_open_slots, meter_airflow_instance_executor_queued_tasks	Executor capacity and queue depth on the scheduler
Asset Updates	count/min	meter_airflow_instance_asset_updates	Asset updates on this host
Asset Triggered DagRuns	count/min	meter_airflow_instance_asset_triggered_dagruns	DagRuns triggered by assets
Triggerer Heartbeat	count/min	meter_airflow_instance_triggerer_heartbeat	Triggerer heartbeats per minute
Triggers Running / Capacity Left	count	meter_airflow_instance_triggers_running, meter_airflow_instance_triggerer_capacity_left	Live deferrable trigger load on the triggerer
Triggers Blocked / Failed / Succeeded	count/min	meter_airflow_instance_triggers_blocked_main_thread, meter_airflow_instance_triggers_failed, meter_airflow_instance_triggers_succeeded	Deferred trigger outcomes on the triggerer

Panel-to-metric mapping details: SWIP-7.

Verification

Run the e2e suites (same as CI):

Suite	Command	Checks
Mock (full MAL contract)	`e2e run -c test/e2e-v2/cases/airflow/mock/e2e.yaml`	29
Cluster (native OTel smoke)	`e2e run -c test/e2e-v2/cases/airflow/cluster/e2e.yaml`	16

CI runs both via .github/workflows/skywalking.yaml (Airflow and Airflow Cluster matrix jobs) after make docker.all.

Details: e2e README.

Horizon UI

Open Workflow Scheduler → Airflow after OAP ingests metrics.

When Airflow is linked to K8S_SERVICE via service hierarchy, start from the 3D Infrastructure Map (middleware tier) and drill down into Kubernetes Services:

Service — cluster KPIs and DAG processing trends (11 panels).

Components — per-host scheduler (six widgets) or triggerer (three widgets); the template defines nine widgets and each host shows only metrics present in OAP for that role.

Kubernetes Services — service, instances, endpoints, and topology for the linked K8S_SERVICE:

Customization

Override MAL rules under otel-rules/airflow/ or extend Horizon UI dashboards. Restart OAP after rule changes.