py-zipkin

py_zipkin

py_zipkin provides a context manager/decorator along with some utilities to facilitate the usage of Zipkin in Python applications.

Install

pip install py_zipkin

Usage

py_zipkin requires a transport_handler object that handles logging zipkin messages to a central logging service such as kafka or scribe.

py_zipkin.zipkin.zipkin_span is the main tool for starting zipkin traces or logging spans inside an ongoing trace. zipkin_span can be used as a context manager or a decorator.

Usage #1: Start a trace with a given sampling rate

from py_zipkin.zipkin import zipkin_span

def some_function(a, b):
    with zipkin_span(
        service_name='my_service',
        span_name='my_span_name',
        transport_handler=some_handler,
        port=42,
        sample_rate=0.05, # Value between 0.0 and 100.0
    ):
        do_stuff(a, b)

Usage #2: Trace a service call

The difference between this and Usage #1 is that the zipkin_attrs are calculated separately and passed in, thus negating the need of the sample_rate param.

# Define a pyramid tween
def tween(request):
    zipkin_attrs = some_zipkin_attr_creator(request)
    with zipkin_span(
        service_name='my_service',
        span_name='my_span_name',
        zipkin_attrs=zipkin_attrs,
        transport_handler=some_handler,
        port=22,
    ) as zipkin_context:
        response = handler(request)
        zipkin_context.update_binary_annotations(
            some_binary_annotations)
        return response

Usage #3: Log a span inside an ongoing trace

This can be also be used inside itself to produce continuously nested spans.

@zipkin_span(service_name='my_service', span_name='some_function')
def some_function(a, b):
    return do_stuff(a, b)

Other utilities

zipkin_span.update_binary_annotations() can be used inside a zipkin trace to add to the existing set of binary annotations.

def some_function(a, b):
    with zipkin_span(
        service_name='my_service',
        span_name='some_function',
        transport_handler=some_handler,
        port=42,
        sample_rate=0.05,
    ) as zipkin_context:
        result = do_stuff(a, b)
        zipkin_context.update_binary_annotations({'result': result})

zipkin_span.add_sa_binary_annotation() can be used to add a binary annotation to the current span with the key 'sa'. This function allows the user to specify the destination address of the service being called (useful if the destination doesn't support zipkin). See http://zipkin.io/pages/data_model.html for more information on the 'sa' binary annotation.

NOTE: the V2 span format only support 1 "sa" endpoint (represented by remoteEndpoint) so add_sa_binary_annotation now raises ValueError if you try to set multiple "sa" annotations for the same span.

def some_function():
    with zipkin_span(
        service_name='my_service',
        span_name='some_function',
        transport_handler=some_handler,
        port=42,
        sample_rate=0.05,
    ) as zipkin_context:
        make_call_to_non_instrumented_service()
        zipkin_context.add_sa_binary_annotation(
            port=123,
            service_name='non_instrumented_service',
            host='12.34.56.78',
        )

create_http_headers_for_new_span() creates a set of HTTP headers that can be forwarded in a request to another service.

headers = {}
headers.update(create_http_headers_for_new_span())
http_client.get(
    path='some_url',
    headers=headers,
)

Transport

py_zipkin (for the moment) thrift-encodes spans. The actual transport layer is pluggable, though.

The recommended way to implement a new transport handler is to subclass py_zipkin.transport.BaseTransportHandler and implement the send and get_max_payload_bytes methods.

send receives an already encoded thrift list as argument. get_max_payload_bytes should return the maximum payload size supported by your transport, or None if you can send arbitrarily big messages.

The simplest way to get spans to the collector is via HTTP POST. Here's an example of a simple HTTP transport using the requests library. This assumes your Zipkin collector is running at localhost:9411.

NOTE: older versions of py_zipkin suggested implementing the transport handler as a function with a single argument. That's still supported and should work with the current py_zipkin version, but it's deprecated.

import requests

from py_zipkin.transport import BaseTransportHandler


class HttpTransport(BaseTransportHandler):

    def get_max_payload_bytes(self):
        return None

    def send(self, encoded_span):
        # The collector expects a thrift-encoded list of spans.
        requests.post(
            'http://localhost:9411/api/v1/spans',
            data=encoded_span,
            headers={'Content-Type': 'application/x-thrift'},
        )

If you have the ability to send spans over Kafka (more like what you might do in production), you'd do something like the following, using the kafka-python package:

from kafka import SimpleProducer, KafkaClient

from py_zipkin.transport import BaseTransportHandler


class KafkaTransport(BaseTransportHandler):

    def get_max_payload_bytes(self):
        # By default Kafka rejects messages bigger than 1000012 bytes.
        return 1000012

    def send(self, message):
        kafka_client = KafkaClient('{}:{}'.format('localhost', 9092))
        producer = SimpleProducer(kafka_client)
        producer.send_messages('kafka_topic_name', message)

Using in multithreading environments

If you want to use py_zipkin in a cooperative multithreading environment, e.g. asyncio, you need to explicitly pass an instance of py_zipkin.storage.Stack as parameter context_stack for zipkin_span and create_http_headers_for_new_span. By default, py_zipkin uses a thread local storage for the attributes, which is defined in py_zipkin.storage.ThreadLocalStack.

Additionally, you'll also need to explicitly pass an instance of py_zipkin.storage.SpanStorage as parameter span_storage to zipkin_span.

from py_zipkin.zipkin import zipkin_span
from py_zipkin.storage import Stack
from py_zipkin.storage import SpanStorage


def my_function():
    context_stack = Stack()
    span_storage = SpanStorage()
    await my_function(context_stack, span_storage)

async def my_function(context_stack, span_storage):
    with zipkin_span(
        service_name='my_service',
        span_name='some_function',
        transport_handler=some_handler,
        port=42,
        sample_rate=0.05,
        context_stack=context_stack,
        span_storage=span_storage,
    ):
        result = do_stuff(a, b)

Firehose mode [EXPERIMENTAL]

"Firehose mode" records 100% of the spans, regardless of sampling rate. This is useful if you want to treat these spans differently, e.g. send them to a different backend that has limited retention. It works in tandem with normal operation, however there may be additional overhead. In order to use this, you add a firehose_handler just like you add a transport_handler.

This feature should be considered experimental and may be removed at any time without warning. If you do use this, be sure to send asynchronously to avoid excess overhead for every request.

License

Copyright (c) 2018, Yelp, Inc. All Rights reserved. Apache v2

1.2.8 (2023-03-23)

• Add back exports in py_zipkin.encoding
• Fix mypy tests

1.2.7 (2023-02-06)

• Drop support for Python 3.6

1.2.6 (2023-02-06)

• Drop support for V1_THRIFT encoding

1.0.0 (2022-06-09)

• Droop Python 2.7 support (minimal supported python version is 3.5)
• Recompile protobuf using version 3.19

0.21.0 (2021-03-17)

• The default encoding is now V2 JSON. If you want to keep the old V1 thrift encoding you'll need to specify it.

0.20.2 (2021-03-11)

• Don't crash when annotating exceptions that cannot be str()'d

0.20.1 (2020-10-27)

• Support PRODUCER and CONSUMER spans

0.20.0 (2020-03-09)

• Add create_http_headers helper

0.19.0 (2020-02-28)

• Add zipkin_span.add_annotation() method
• Add autoinstrumentation for python Threads
• Allow creating a copy of Tracer
• Add extract_zipkin_attrs_from_headers() helper

0.18.7 (2020-01-15)

• Expose encoding.create_endpoint helper

0.18.6 (2019-09-23)

• Ensure tags are strings when using V2_JSON encoding

0.18.5 (2019-08-08)

• Add testing.MockTransportHandler module

0.18.4 (2019-08-02)

• Fix thriftpy2 import to allow cython module

0.18.3 (2019-05-15)

• Fix unicode bug when decoding thrift tag strings

0.18.2 (2019-03-26)

• Handled exception while emitting trace and log the error
• Ensure tracer is cleared regardless span of emit outcome

0.18.1 (2019-02-22)

• Fix ThreadLocalStack() bug introduced in 0.18.0

0.18.0 (2019-02-13)

• Fix multithreading issues
• Added Tracer module

0.17.1 (2019-02-05)

• Ignore transport_handler overrides in an inner span since that causes spans to be dropped.

0.17.0 (2019-01-25)

• Support python 3.7
• py-zipkin now depends on thriftpy2 rather than thriftpy. They can coexist in the same codebase, so it should be safe to upgrade.

0.16.1 (2018-11-16)

• Handle null timestamps when decoding thrift traces

0.16.0 (2018-11-13)

• py_zipkin is now able to convert V1 thrift spans to V2 JSON

0.15.1 (2018-10-31)

• Changed DeprecationWarnings to logging.warning

0.15.0 (2018-10-22)

• Added support for V2 JSON encoding.
• Fixed TransportHandler bug that was affecting also V1 JSON.

0.14.1 (2018-10-09)

• Fixed memory leak introduced in 0.13.0.

0.14.0 (2018-10-01)

• Support JSON encoding for V1 spans.
• Allow overriding the span_name after creation.

0.13.0 (2018-06-25)

• Removed deprecated zipkin_logger.debug() interface.
• py_zipkin.stack was renamed as py_zipkin.storage. If you were importing this module, you'll need to update your code.

0.12.0 (2018-05-29)

• Support max payload size for transport handlers.
• Transport handlers should now be implemented as classes extending py_zipkin.transport.BaseTransportHandler.

0.11.2 (2018-05-23)

• Don't overwrite passed in annotations

0.11.1 (2018-05-23)

• Add binary annotations to the span even if the request is not being sampled. This fixes binary annotations for firehose spans.

0.11.0 (2018-02-08)

• Add support for "firehose mode", which logs 100% of the spans regardless of sample rate.

0.10.1 (2018-02-05)

• context_stack will now default to ThreadLocalStack() if passed as None

0.10.0 (2018-02-05)

• Add support for using explicit in-process context storage instead of using thread_local. This allows you to use py_zipkin in cooperative multitasking environments e.g. asyncio
• py_zipkin.thread_local is now deprecated. Instead use py_zipkin.stack.ThreadLocalStack()
• TraceId and SpanId generation performance improvements.
• 128-bit TraceIds now start with an epoch timestamp to support easy interop with AWS X-Ray

0.9.0 (2017-07-31)

• Add batch span sending. Note that spans are now sent in lists.

0.8.3 (2017-07-10)

• Be defensive about having logging handlers configured to avoid throwing NullHandler attribute errors

0.8.2 (2017-06-30)

• Don't log ss and sr annotations when in a client span context
• Add error binary annotation if an exception occurs

0.8.1 (2017-06-16)

• Fixed server send timing to more accurately reflect when server send actually occurs.
• Replaced logging_start annotation with logging_end

0.8.0 (2017-06-01)

• Added 128-bit trace id support
• Added ability to explicitly specify host for a span
• Added exception handling if host can't be determined automatically
• SERVER_ADDR ('sa') binary annotations can be added to spans
• py36 support

0.7.1 (2017-05-01)

• Fixed a bug where update_binary_annotations would fail for a child span in a trace that is not being sampled

0.7.0 (2017-03-06)

• Simplify update_binary_annotations for both root and non-root spans

0.6.0 (2017-02-03)

• Added support for forcing zipkin_span to report timestamp/duration. Changes API of zipkin_span, but defaults back to existing behavior.

0.5.0 (2017-02-01)

• Properly set timestamp/duration on server and local spans
• Updated thrift spec to include these new fields
• The zipkin_span entrypoint should be backwards compatible

0.4.4 (2016-11-29)

• Add optional annotation for when Zipkin logging starts

0.4.3 (2016-11-04)

• Fix bug in zipkin_span decorator

0.4.2 (2016-11-01)

• Be defensive about transport_handler when logging spans.

0.4.1 (2016-10-24)

• Add ability to override span_id when creating new ZipkinAttrs.

0.4.0 (2016-10-20)

• Added start and stop functions as friendlier versions of the enter and exit functions.

0.3.1 (2016-09-30)

• Adds new param to thrift.create_endpoint allowing creation of thrift Endpoint objects on a proxy machine representing another host.

0.2.1 (2016-09-30)

• Officially "release" v0.2.0. Accidentally pushed a v0.2.0 without the proper version bump, so v0.2.1 is the new real version. Please use this instead of v0.2.0.

0.2.0 (2016-09-30)

• Fix problem where if zipkin_attrs and sample_rate were passed, but zipkin_attrs.is_sampled=True, new zipkin_attrs were being generated.

0.1.2 (2016-09-29)

• Fix sampling algorithm that always sampled for rates > 50%

0.1.1 (2016-07-05)

• First py_zipkin version with context manager/decorator functionality.