# Using a DAG to write custom metrics in CloudWatch
You can use the following code example to write a directed acyclic graph (DAG) that runs a `PythonOperator` to retrieve OS-level metrics for an Amazon MWAA environment. The DAG then publishes the data as custom metrics to Amazon CloudWatch.
Custom OS-level metrics provide you with additional visibility about how your environment workers are utilizing resources such as virtual memory and CPU. You can use this information to select the [environment class](environment-class.md) that best suits your workload.
**Topics**
+ [Version](#samples-custom-metrics-version)
+ [Prerequisites](#samples-custom-metrics-prereqs)
+ [Permissions](#samples-custom-metrics-permissions)
+ [Dependencies](#samples-custom-metrics-dependencies)
+ [Code example](#samples-custom-metrics-code)
## Version
You can use the code example on this page with **Apache Airflow v2** in [Python 3.10](https://peps.python.org/pep-0619/) and **Apache Airflow v3** in [Python 3.11](https://peps.python.org/pep-0664/).
## Prerequisites
To use the code example on this page, you need the following:
+ An [Amazon MWAA environment](get-started.md).
## Permissions
No additional permissions are required to use the code example on this page.
## Dependencies
+ No additional dependencies are required to use the code example on this page.
## Code example
1. In your command prompt, navigate to the folder where your DAG code is stored. For example:
```
cd dags
```
1. Copy the contents of the following code example and save it locally as `dag-custom-metrics.py`. Replace `MWAA-ENV-NAME` with your environment name.
```
from airflow import DAG
from airflow.operators.python_operator import PythonOperator
from airflow.utils.dates import days_ago
from datetime import datetime
import os,json,boto3,psutil,socket
def publish_metric(client,name,value,cat,unit='None'):
environment_name = os.getenv("MWAA_ENV_NAME")
value_number=float(value)
hostname = socket.gethostname()
ip_address = socket.gethostbyname(hostname)
print('writing value',value_number,'to metric',name)
response = client.put_metric_data(
Namespace='MWAA-Custom',
MetricData=[
{
'MetricName': name,
'Dimensions': [
{
'Name': 'Environment',
'Value': environment_name
},
{
'Name': 'Category',
'Value': cat
},
{
'Name': 'Host',
'Value': ip_address
},
],
'Timestamp': datetime.now(),
'Value': value_number,
'Unit': unit
},
]
)
print(response)
return response
def python_fn(**kwargs):
client = boto3.client('cloudwatch')
cpu_stats = psutil.cpu_stats()
print('cpu_stats', cpu_stats)
virtual = psutil.virtual_memory()
cpu_times_percent = psutil.cpu_times_percent(interval=0)
publish_metric(client=client, name='virtual_memory_total', cat='virtual_memory', value=virtual.total, unit='Bytes')
publish_metric(client=client, name='virtual_memory_available', cat='virtual_memory', value=virtual.available, unit='Bytes')
publish_metric(client=client, name='virtual_memory_used', cat='virtual_memory', value=virtual.used, unit='Bytes')
publish_metric(client=client, name='virtual_memory_free', cat='virtual_memory', value=virtual.free, unit='Bytes')
publish_metric(client=client, name='virtual_memory_active', cat='virtual_memory', value=virtual.active, unit='Bytes')
publish_metric(client=client, name='virtual_memory_inactive', cat='virtual_memory', value=virtual.inactive, unit='Bytes')
publish_metric(client=client, name='virtual_memory_percent', cat='virtual_memory', value=virtual.percent, unit='Percent')
publish_metric(client=client, name='cpu_times_percent_user', cat='cpu_times_percent', value=cpu_times_percent.user, unit='Percent')
publish_metric(client=client, name='cpu_times_percent_system', cat='cpu_times_percent', value=cpu_times_percent.system, unit='Percent')
publish_metric(client=client, name='cpu_times_percent_idle', cat='cpu_times_percent', value=cpu_times_percent.idle, unit='Percent')
return "OK"
with DAG(dag_id=os.path.basename(__file__).replace(".py", ""), schedule_interval='*/5 * * * *', catchup=False, start_date=days_ago(1)) as dag:
t = PythonOperator(task_id="memory_test", python_callable=python_fn, provide_context=True)
```
1. Run the following AWS CLI command to copy the DAG to your environment's bucket, then trigger the DAG using the Apache Airflow UI.
```
aws s3 cp your-dag.py s3://your-environment-bucket/dags/
```
1. If the DAG runs successfully, you get something similar to the following in your Apache Airflow logs:
```
[2022-08-16, 10:54:46 UTC] {{logging_mixin.py:109}} INFO - cpu_stats scpustats(ctx_switches=3253992384, interrupts=1964237163, soft_interrupts=492328209, syscalls=0)
[2022-08-16, 10:54:46 UTC] {{logging_mixin.py:109}} INFO - writing value 16024199168.0 to metric virtual_memory_total
[2022-08-16, 10:54:46 UTC] {{logging_mixin.py:109}} INFO - {'ResponseMetadata': {'RequestId': 'fad289ac-aa51-46a9-8b18-24e4e4063f4d', 'HTTPStatusCode': 200, 'HTTPHeaders': {'x-amzn-requestid': 'fad289ac-aa51-46a9-8b18-24e4e4063f4d', 'content-type': 'text/xml', 'content-length': '212', 'date': 'Tue, 16 Aug 2022 17:54:45 GMT'}, 'RetryAttempts': 0}}
[2022-08-16, 10:54:46 UTC] {{logging_mixin.py:109}} INFO - writing value 14356287488.0 to metric virtual_memory_available
[2022-08-16, 10:54:46 UTC] {{logging_mixin.py:109}} INFO - {'ResponseMetadata': {'RequestId': '6ef60085-07ab-4865-8abf-dc94f90cab46', 'HTTPStatusCode': 200, 'HTTPHeaders': {'x-amzn-requestid': '6ef60085-07ab-4865-8abf-dc94f90cab46', 'content-type': 'text/xml', 'content-length': '212', 'date': 'Tue, 16 Aug 2022 17:54:45 GMT'}, 'RetryAttempts': 0}}
[2022-08-16, 10:54:46 UTC] {{logging_mixin.py:109}} INFO - writing value 1342296064.0 to metric virtual_memory_used
[2022-08-16, 10:54:46 UTC] {{logging_mixin.py:109}} INFO - {'ResponseMetadata': {'RequestId': 'd5331438-5d3c-4df2-bc42-52dcf8d60a00', 'HTTPStatusCode': 200, 'HTTPHeaders': {'x-amzn-requestid': 'd5331438-5d3c-4df2-bc42-52dcf8d60a00', 'content-type': 'text/xml', 'content-length': '212', 'date': 'Tue, 16 Aug 2022 17:54:45 GMT'}, 'RetryAttempts': 0}}
...
[2022-08-16, 10:54:46 UTC] {{python.py:152}} INFO - Done. Returned value was: OK
[2022-08-16, 10:54:46 UTC] {{taskinstance.py:1280}} INFO - Marking task as SUCCESS. dag_id=dag-custom-metrics, task_id=memory_test, execution_date=20220816T175444, start_date=20220816T175445, end_date=20220816T175446
[2022-08-16, 10:54:46 UTC] {{local_task_job.py:154}} INFO - Task exited with return code 0
```