Python client library for SignalFx

This is a programmatic interface in Python for SignalFx's metadata and ingest APIs. It is meant to provide a base for communicating with SignalFx APIs that can be easily leveraged by scripts and applications to interact with SignalFx or report metric and event data to SignalFx. It is also the base for metric reporters that integrate with common Python-based metric collections tools or libraries.

Installation

To install with pip:

$ pip install signalfx

To install from source:

$ git clone https://github.com/signalfx/signalfx-python.git
$ cd signalfx-python/
$ pip install -e .

Usage

This client library provides programmatic access to SignalFx's APIs:

the data ingest API;
the metadata REST API;
the SignalFlow API.

You start by instantiating a signalfx.SignalFx() object, which then gives you access to the API client that you want:

import signalfx

sfx = signalfx.SignalFx(api_endpoint='https://api.{REALM}.signalfx.com',
        ingest_endpoint='https://ingest.{REALM}.signalfx.com',
        stream_endpoint='https://stream.{REALM}.signalfx.com')


# For the ingest API
ingest = sfx.ingest('ORG_TOKEN')

# For the REST API
rest = sfx.rest('API_TOKEN')

# For the SignalFlow API
flow = sfx.signalflow('ACCESS_TOKEN')

If no endpoints are set manually, this library uses the us0 realm by default. If you are not in this realm, you will need to explicitly set the endpoint urls above. To determine if you are in a different realm and need to explicitly set the endpoints, check your profile page in the SignalFx web application. You will also need to specify an access token when requesting one of those clients. For the ingest client, you need to specify your organization access token (which can be obtained from the SignalFx organization you want to report data into). For the REST API, you must use your user access token. For the SignalFlow client, either an organization access token or a user access token may be used. For more information on access tokens, see the API's Authentication documentation.

Reporting data

Basic usage of the library for reporting data goes as follows:

import signalfx

with signalfx.SignalFx().ingest('ORG_TOKEN') as sfx:
    sfx.send(
        gauges=[
          {'metric': 'myfunc.time',
           'value': 532,
           'timestamp': 1442960607000},
          ...
        ],
        counters=[
          {'metric': 'myfunc.calls',
           'value': 42,
           'timestamp': 1442960607000},
          ...
        ],
        cumulative_counters=[
          {'metric': 'myfunc.calls_cumulative',
           'value': 10,
           'timestamp': 1442960607000},
          ...
        ])

If you're sending data from multiple places in your code, you should create your ingest client once and use it throughout your application. Each ingest client instance has an internal queue of datapoints and events that need to be sent to SignalFx, as well as an internal thread draining that queue. When you no longer need the client instance, make sure you call .stop() on it to ensure the queue is fully drained.

import signalfx

sfx = signalfx.SignalFx().ingest('ORG_TOKEN')
try:
    sfx.send(...)
    sfx.send(...)
finally:
    # Make sure that everything gets sent.
    sfx.stop()

The timestamp must be a millisecond precision timestamp; the number of milliseconds elapsed since Epoch. The timestamp field is optional, but strongly recommended. If not specified, it will be set by SignalFx's ingest servers automatically; in this situation, the timestamp of your datapoints will not accurately represent the time of their measurement (network latency, batching, etc. will all impact when those datapoints actually make it to SignalFx).

When sending datapoints with multiple calls to send(), it is recommended to re-use the same SignalFx client object for each send() call.

If you must use multiple client objects for the same token, which is not recommended, it is important to call stop() after making all send() calls. Each SignalFx client object uses a background thread to send datapoints without blocking the caller. Calling stop() will gracefully flush the thread's send queue and close its TCP connections.

Sending multi-dimensional data

Reporting dimensions for the data is also optional, and can be accomplished by specifying a dimensions parameter on each datapoint containing a dictionary of string to string key/value pairs representing the dimensions:

import signalfx

with signalfx.SignalFx().ingest('ORG_TOKEN') as sfx:
    sfx.send(
        gauges=[
          {
            'metric': 'myfunc.time',
            'value': 532,
            'timestamp': 1442960607000,
            'dimensions': {'host': 'server1', 'host_ip': '1.2.3.4'}
          },
          ...
        ], ...)

See examples/generic_usecase.py for a complete code sample showing how to send data to SignalFx.

Sending events

Events can be sent to SignalFx via the send_event() function. The event type must be specified, and dimensions and extra event properties can be supplied as well.

import signalfx

with signalfx.SignalFx().ingest('ORG_TOKEN') as sfx:
    sfx.send_event(
        event_type='deployments',
        dimensions={
            'host': 'myhost',
            'service': 'myservice',
            'instance': 'myinstance'},
        properties={
            'version': '2015.04.29-01'})

Metric metadata and tags

The library includes functions to search, retrieve, and update metric metadata and tags. Deleting tags is also supported.

import signalfx

with signalfx.SignalFx().rest('ORG_TOKEN') as sfx:
    sfx.update_tag('tag_name',
                   description='An example tag',
                   custom_properties={'version': 'some_number'})

AWS integration

Optionally, the client may be configured to append additional dimensions to all metrics and events sent to SignalFx. One use case for this is to append the AWS unique ID of the current host as an extra dimension. For example,

import signalfx
from signalfx.aws import AWS_ID_DIMENSION, get_aws_unique_id

sfx = signalfx.SignalFx().ingest('ORG_TOKEN')

# This dimension will be added to all datapoints sent.
sfx.add_dimensions({AWS_ID_DIMENSION: get_aws_unique_id()})

try:
    sfx.send(
        gauges=[
          {
            'metric': 'myfunc.time',
            'value': 532,
            'timestamp': 1442960607000
            'dimensions': {'host': 'server1', 'host_ip': '1.2.3.4'}
          },
        ])
finally:
    sfx.stop()

Pyformance reporter

pyformance is a Python library that provides CodaHale-style metrics in a very Pythonic way. We offer a reporter that can report the pyformance metric registry data directly to SignalFx.

from signalfx.pyformance import (count_calls, count_calls_with_dims,
                                 gauge, SignalFxReporter)

@count_calls
def callme():
    # whatever
    pass

@count_calls_with_dims(dimension_key="dimension_value")
def callme_with_dims():
    # whatever
    pass

sfx = SignalFxReporter(token='ORG_TOKEN')
sfx.start()

callme()
callme()
callme_with_dims()
callme_with_dims()
gauge('test').set_value(42)

See examples/pyformance_usecase.py for a complete code example using Pyformance.

Executing SignalFlow computations

SignalFlow is SignalFx's real-time analytics computation language. The SignalFlow API allows SignalFx users to execute real-time streaming analytics computations on the SignalFx platform. For more information, head over to our Developers documentation:

The SignalFlow client accepts either an Organization Access Token or a User API Token. Executing a SignalFlow program is very simple with this client library:

import signalfx

program = "data('cpu.utilization').mean().publish()"
with signalfx.SignalFx().signalflow('ACCESS_TOKEN') as flow:
    print('Executing {0} ...'.format(program))
    computation = flow.execute(program)
    for msg in computation.stream():
        if isinstance(msg, signalfx.signalflow.messages.DataMessage):
            print('{0}: {1}'.format(msg.logical_timestamp_ms, msg.data))
        if isinstance(msg, signalfx.signalflow.messages.EventMessage):
            print('{0}: {1}'.format(msg.timestamp_ms, msg.properties))

Metadata about the streamed timeseries is received from .stream(), but it is automatically intercepted by the client library and made available through the Computation object returned by execute():

if isinstance(msg, signalfx.signalflow.messages.DataMessage):
    for tsid, value in msg.data.items():
        metadata = computation.get_metadata(tsid)
        # Display metadata and datapoint value as desired

For more examples of how to execute SignalFlow computation with this library, interpret and use the returned stream messages, you can look at the simple example in examples/signalflow/basic.py or at the SignalFlow CLI and its implementation which uses this library.

Building a Pandas DataFrame from SignalFlow output

With the SignalFlow output being programmatically accessible, it's easy to convert this data into any form that you need for further use or analysis. One such use case is to build a Pandas DataFrame with the computation's output. For a complete example of how to do this, see examples/signalflow/dataframe.py.

Known Issues

Sending only 1 datapoint and not seeing it in the chart

The reason you are not seeing the metrics in the chart is because the script that is calling the Python client module is exiting right after calling the send method. The Python client library is mainly targeted towards sending a continuous stream of metrics and was implemented to be asynchronous.

To work around this problem (most common in short-lived scripts for example), register an atexit function to cleanly stop the datapoint sending thread when your program exits:

import atexit
import signalfx

sfx = signalfx.SignalFx().ingest('ORG_TOKEN')
atexit.register(sfx.stop)

SSLError when working with tags, metrics, dimensions, metrictimeseries, organization

ERROR:root:Posting to SignalFx failed.
SSLError: hostname 'api.signalfx.com' doesn't match either of '*.signalfuse.com', 'signalfuse.com'.

Root Cause: SignalFx's API endpoints (api.signalfx.com, ingest.signalfx.com and stream.signalfx.com) have SSL SNI enabled and the urllib3 module in Python versions prior to 2.7.8 had a bug that causes the above issue. This was fixed in later versions of Python; we recommend using Python 2.7.9 or newer when using this library.

Exceeding int64 integer sizes

The protocol buffer used to transmit data through the ingest API restricts integers and longs to (-(2**63) to (2**63)-1). long values in Python 2.x and int values in 3.x can exceed these values. Any value or property value less than -(2**63) or greater than (2**63)-1 will raise a ValueError exception.

danny-cheung/signalfx-python