airflow-bash-ssh-to-orchestra

Airflow Bash/SSH Operators → Orchestra

Overview

Airflow's BashOperator, SSHOperator, and WinRMOperator run shell commands — locally or on remote hosts. Orchestra doesn't have a local shell execution model; instead:

• Remote shell commands → LINUX_SSH + LINUX_SSH_EXECUTE_COMMAND or WINDOWS_SSH + WINDOWS_SSH_COMMAND — for mapping ssh_conn_id=/winrm_conn_id= to the right Orchestra connection, see airflow-connections-to-orchestra
• Python scripts via bash → PYTHON + PYTHON_EXECUTE_SCRIPT (the better pattern)
• Containerised workloads → AWS_ECS, GKE, AZURE_CONTAINER_APPS, or AKS depending on cloud
• Pure local glue (echo, mkdir, env checks) → fold into adjacent tasks or drop

---

Decision Tree

Is it a dbt command (dbt run/build/test/seed/snapshot)?
  → YES: use dbt-core-airflow-to-orchestra skill instead — DBT_CORE_EXECUTE,
         even if it arrived via SSHOperator against a remote dbt server.
         Do NOT convert it to LINUX_SSH.
  → NO: continue

Does it run a Python script?
  → YES: PYTHON + PYTHON_EXECUTE_SCRIPT

Does it SSH to a remote server to run commands?
  → Linux: LINUX_SSH + LINUX_SSH_EXECUTE_COMMAND
  → Windows: WINDOWS_SSH + WINDOWS_SSH_COMMAND

Is it a Kubernetes pod?
  → AWS: AWS_EKS + AWS_EKS_RUN_JOB
  → GCP: GKE + GKE_RUN_JOB
  → Azure: AZURE_KUBERNETES_SERVICE + AKS_RUN_JOB

Is it a Docker container (ECS/Cloud Run/Container Apps)?
  → AWS: AWS_ECS + AWS_ECS_RUN_TASK
  → GCP: GCP_CLOUD_RUN + GCP_CLOUD_RUN_EXECUTE_JOB
  → Azure: AZURE_CONTAINER_APPS + ACA_RUN_JOB

Is it purely local glue (echo, mkdir, env check)?
  → Assess whether it can be dropped or folded into adjacent task

---

BashOperator (Python script) → PYTHON

The most common pattern: BashOperator(bash_command="python scripts/my_script.py") maps directly to PYTHON_EXECUTE_SCRIPT.

# Airflow
run_script = BashOperator(
    task_id="run_etl_script",
    bash_command="cd /opt/airflow && python scripts/etl.py --env prod",
)