timeexecution 8.4.0

Python project

Time Execution

This package is designed to record application metrics into specific backends. With the help of Grafana or Kibana you can easily use these metrics to create meaningful monitoring dashboards.

Features

• Sending data to multiple backends (e.g. ElasticSearch)
• Custom backends
• Hooks to include additional data per metric.

Available backends

• Elasticsearch client >=8,<9
• Elasticsearch server >=7,<9

Note: In previous versions, this package supported other backends out of the box, namely InfluxDB and Kafka. Although, these have been removed.

Installation

If you want to use it with the ElasticSearchBackend:

\$ pip install timeexecution[elasticsearch]

or if you prefer to have all backends available and easily switch between them:

\$ pip install timeexecution[all]

Configuration

The package can be configured with the follwing settings:

• origin: A string that will be included in field origin in all metrics. This is particularly useful in an environment where the same backend (e.g. an Elasticsearch index) is shared by multiple applications or micro-services, so each application uses it's own origin identifier.
• backends: Specify the backend where to send metrics.
• hooks: Hooks allow you to include additional fields as part of the metric data. Learn more about how to use hooks
• duration_field - the field to be used to store the duration measured. If no value is provided, the default will be value.

Usage

To use this package you decorate the functions you want to time its execution. Every wrapped function will create a metric consisting of 3 default values:

• name - The name of the series the metric will be stored in. By default, timeexecution will use the fully qualified name of the decorated method or function (e.g. src.api.views.ExampleView.get).
• value - The time it took in ms for the wrapped function to complete.
• hostname - The hostname of the machine the code is running on.

See the following example

from time_execution import settings, time_execution
from time_execution.backends.elasticsearch import ElasticsearchBackend

# Setup the desired backend
elasticsearch_backend = ElasticsearchBackend('elasticsearch', index='metrics')

# Configure the time_execution decorator
settings.configure(backends=[elasticsearch_backend])

# Wrap the methods where u want the metrics
@time_execution
def hello():
    return 'World'

# Now when we call hello() and we will get metrics in our backends
hello()

This will result in an entry in Elasticsearch:

[
    {
        "_index": "metrics-2016.01.28",
        "_type": "metric",
        "_id": "AVKIp9DpnPWamvqEzFB3",
        "_score": null,
        "_source": {
            "timestamp": "2016-01-28T14:34:05.416968",
            "hostname": "dfaa4928109f",
            "name": "__main__.hello",
            "value": 312
        },
        "sort": [
            1453991645416
        ]
    }
]

It's also possible to use a thread. It will basically add metrics to a queue, and these will be then sent in bulk to the configured backend. This setup is useful to avoid the impact of network latency or backend performance.

For example:

from time_execution import settings, time_execution
from time_execution.backends.threaded import ThreadedBackend

# Setup threaded backend which will be run on separate thread
threaded_backend = ThreadedBackend(
    backend=ElasticsearchBackend,
    backend_kwargs={
        "host" : "elasticsearch",
        "index": "metrics",
    }
)

# there is also possibility to configure backend by import path, like:
threaded_backend = ThreadedBackend(
    backend="time_execution.backends.elasticsearch.ElasticsearchBackend",
    #: any other configuration belongs to backend
    backend_kwargs={
        "hosts" : "elasticsearch",
        "topic": "metrics"
    }
)

# Configure the time_execution decorator
settings.configure(backends=[threaded_backend])

# Wrap the methods where u want the metrics
@time_execution
def hello():
    return 'World'

# Now when we call hello() we put metrics in queue to send it either in some
# configurable time later or when queue will reach configurable limit.
hello()

It's also possible to decorate coroutines or awaitables in Python >=3.5.

For example:

import asyncio
from time_execution import time_execution_async

# ... Setup the desired backend(s) as described above ...

# Wrap the methods where you want the metrics
@time_execution_async
async def hello():
    await asyncio.sleep(1)
    return 'World'

# Now when we schedule hello() we will get metrics in our backends
loop = asyncio.get_event_loop()
loop.run_until_complete(hello())

Hooks

time_execution supports hooks where you can change the metric before its being sent to the backend.

With a hook you can add additional and change existing fields. This can be useful for cases where you would like to add a column to the metric based on the response of the wrapped function.

Simple callable hook

A function which is called after calling the decorated function and before sending the metrics. It receives the following arguments:

• response: the returned value of the wrapped function
• exception: exception raised the wrapped function, if any
• metric: a dictionary containing the data to be sent to the backend
• func: the decorated function itself
• func_args: original args received by the wrapped function.
• func_kwargs: original kwargs received by the wrapped function.

From within a hook you can change the name if you want the metrics to be split into multiple series.

See the following example how to setup hooks.

# Now lets create a hook
def my_hook(response, exception, metric, func, func_args, func_kwargs):
    status_code = getattr(response, 'status_code', None)
    if status_code:
        return dict(
            name='{}.{}'.format(metric['name'], status_code),
            extra_field='foo bar'
        )

# Configure the time_execution decorator, but now with hooks
settings.configure(backends=[backend], hooks=[my_hook])

There is also possibility to create decorator with custom set of hooks. It is needed for example to track celery tasks.

from multiprocessing import current_process
# Hook for celery tasks
def celery_hook(response, exception, metric, func, func_args, func_kwargs):
    """
    Add celery worker-specific details into response.
    """
    p = current_process()
    hook = {
        'name': metric.get('name'),
        'value': metric.get('value'),
        'success': exception is None,
        'process_name': p.name,
        'process_pid': p.pid,
    }
    return hook

# Create time_execution decorator with extra hooks
time_execution_celery = time_execution(extra_hooks=[celery_hook])

@celery.task
@time_execution_celery
def celery_task(self, **kwargs):
    return True

# Or do it in place where it is needed
@celery.task
@time_execution(extra_hooks=[celery_hook])
def celery_task(self, **kwargs):
    return True

# Or override default hooks by custom ones. Just setup `disable_default_hooks` flag
@celery.task
@time_execution(extra_hooks=[celery_hook], disable_default_hooks=True)
def celery_task(self, **kwargs):
    return True

Generator hook

A generator hook works similarly to a simple hook, but also allows to run an arbitrary code just before the decorated function is called. It receives func, func_args, and func_kwargs as its arguments, whilst response, exception, and metric are sent later at the yield site:

def generator_hook(func, func_args, func_kwargs) -> Generator[
    None,
    Tuple[Any, Optional[BaseException], Dict[str, Any]],  # response, exception, and metrics
    Optional[Dict[str, Any]],
]:
    print("This runs before the decorated function")
    (response, exception, metrics) = yield
    print("This runs after the decorated function")
    ...

Manually sending metrics

You can also send any metric you have manually to the backend. These will not add the default values and will not hit the hooks.

See the following example.

from time_execution import write_metric

loadavg = os.getloadavg()
write_metric('cpu.load.1m', value=loadavg[0])
write_metric('cpu.load.5m', value=loadavg[1])
write_metric('cpu.load.15m', value=loadavg[2])

Custom Backend

Writing a custom backend is very simple, all you need to do is create a class with a [write]{.title-ref} method. It is not required to extend [BaseMetricsBackend]{.title-ref} but, in order to easily upgrade, we recommend you do.

from time_execution.backends.base import BaseMetricsBackend


class MetricsPrinter(BaseMetricsBackend):
    def write(self, name, **data):
        print(name, data)

Example scenario

In order to read the metrics, e.g. using ElasticSearch as a backend, the following lucene query could be used:

name:"__main__.hello" AND hostname:dfaa4928109f

For more advanced query syntax, please have a look at the Lucene documentation and the ElasticSearch Query DSL reference.

Contribute

You have something to contribute? Great! There are a few things that may come in handy.

Testing in this project is done via tox with the use of docker.

There is a Makefile with a few targets that we use often:

• make test
• make format
• make lint
• make build

make test command will run tests for the python versions specified in tox.ini spinning up all necessary services via docker.

In some cases (on Ubuntu 18.04) the Elasticsearch Docker image might not be able to start and will exit with the following error:

max virtual memory areas vm.max_map_count [65530] is too low, increase to at least [262144]

This can be solved by adding the following line to `/etc/sysctl.conf`:

vm.max_map_count=262144