/opentelemetry-python-distro

Primary LanguagePythonApache License 2.0Apache-2.0

Lumigo OpenTelemetry Distro for Python 🌠

Tracer Testing Version

The Lumigo OpenTelemetry Distribution for Python is a package that provides no-code distributed tracing for containerized applications.

The Lumigo OpenTelemetry Distribution for Python is made of several upstream OpenTelemetry packages, with additional automated quality-assurance and customizations that optimize for no-code injection, meaning that you should need to update exactly zero lines of code in your application in order to make use of the Lumigo OpenTelemetry Distribution. (See the No-code activation section for auto-instrumentation instructions)

Logging support

The Lumigo OpenTelemetry Distribution also allows logging span-correlated records. See the configuration section for details on how to enable this feature. When using the logging feature, the same set of rules for secret masking applies on the content of the log message, with only LUMIGO_SECRET_MASKING_REGEX being considered.

Note: If you are looking for the Lumigo Python tracer for AWS Lambda functions, lumigo-tracer is the package you should use instead.

Setup

Adding the Lumigo OpenTelemetry Distro for Python to your application is a three-step process:

  1. Add the Lumigo OpenTelemetry Distro for Python as dependency
  2. Provide configurations through environment variables
  3. Activate the tracer, which can also be achieved through environment variables

Add lumigo_opentelemetry as dependency

The lumigo_opentelemetry package needs to be a dependency of your application. In most cases, you will add lumigo_opentelemetry as a line in requirements.txt:

lumigo_opentelemetry

Or, you may use pip:

pip install lumigo_opentelemetry

Environment-based configuration

Configure the LUMIGO_TRACER_TOKEN environment variable with the token value generated for you by the Lumigo platform, under Settings --> Tracing --> Manual tracing:

LUMIGO_TRACER_TOKEN=<token>

Replace <token> below with the token generated for you by the Lumigo platform.

It is also strongly suggested that you set the OTEL_SERVICE_NAME environment variable with, as value, the service name you have chosen for your application:

OTEL_SERVICE_NAME=<service name>

Replace <service name> with the desired name of the service.

Note: While you are providing environment variables for configuration, consider also providing the one needed for no-code tracer activation :-)

Tracer activation

There are two ways to activate the lumigo_opentelemetry package: one based on importing the package in code (manual activation), and the other via the environment (no-code activation). The no-code activation approach is the preferred one.

No-code activation

Note: The instructions in this section are mutually exclusive with those provided in the Manual activation section.

Set the following environment variable:

AUTOWRAPT_BOOTSTRAP=lumigo_opentelemetry

Manual activation

Note: The instructions in this section are mutually exclusive with those provided in the No-code activation section.

Import lumigo_opentelemetry at the beginning of your main file:

import lumigo_opentelemetry

Configuration

OpenTelemetry configurations

The Lumigo OpenTelemetry Distro for Python is made of several upstream OpenTelemetry packages, together with additional logic and, as such, the environment variables that work with "vanilla" OpenTelemetry work also with the Lumigo OpenTelemetry Distro for Python. Specifically supported are:

Lumigo-specific configurations

The lumigo_opentelemetry package additionally supports the following configuration options as environment variables:

  • LUMIGO_TRACER_TOKEN: [Required] Required configuration to send data to Lumigo; you will find the right value in Lumigo under Settings -> Tracing -> Manual tracing.

  • LUMIGO_DEBUG=true: Enables debug logging

  • LUMIGO_DEBUG_SPANDUMP: path to a local file where to write a local copy of the spans that will be sent to Lumigo; this option handy for local testing but should not be used in production unless you are instructed to do so by Lumigo support.

  • LUMIGO_SECRET_MASKING_REGEX=["regex1", "regex2"]: Prevents Lumigo from sending keys that match the supplied regular expressions. All regular expressions are case-insensitive. By default, Lumigo applies the following regular expressions: [".*pass.*", ".*key.*", ".*secret.*", ".*credential.*", ".*passphrase.*"].

    • We support more granular masking using the following parameters. If not given, the above configuration is the fallback: LUMIGO_SECRET_MASKING_REGEX_HTTP_REQUEST_BODIES, LUMIGO_SECRET_MASKING_REGEX_HTTP_REQUEST_HEADERS, LUMIGO_SECRET_MASKING_REGEX_HTTP_RESPONSE_BODIES, LUMIGO_SECRET_MASKING_REGEX_HTTP_RESPONSE_HEADERS, LUMIGO_SECRET_MASKING_REGEX_HTTP_QUERY_PARAMS, LUMIGO_SECRET_MASKING_REGEX_ENVIRONMENT.

  • LUMIGO_SWITCH_OFF=true: This option disables the Lumigo OpenTelemetry distro entirely; no instrumentation will be injected, no tracing data will be collected.

  • LUMIGO_REPORT_DEPENDENCIES=false: This option disables the built-in dependency reporting to Lumigo SaaS. For more information, refer to the Automated dependency reporting section.

  • LUMIGO_AUTO_FILTER_EMPTY_SQS: This option enables the automatic filtering of empty SQS messages from being sent to Lumigo SaaS. For more information, refer to the Filtering out empty SQS messages section.

  • LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX='["regex1", "regex2"]': This option enables the filtering of client and server endpoints through regular expression searches. Fine-tune your settings via the following environment variables, which work in conjunction with LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX for a specific span type:

    • LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX_SERVER applies the regular expression search exclusively to server spans. Searching is performed against the following attributes on a span: url.path and http.target.
    • LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX_CLIENT applies the regular expression search exclusively to client spans. Searching is performed against the following attributes on a span: url.full and http.url.

    For more information check out Filtering http endpoints.

Logging instrumentation

  • LUMIGO_ENABLE_LOGS - Default: false. When set to true, turns on the logging instrumentation to capture log-records logged by Python's logging builtin library and send them to Lumigo. Emitted logs will also get injected with the active span context (see list).

  • LUMIGO_DEBUG_LOGDUMP - similar to LUMIGO_DEBUG_SPANDUMP, only for logs instead of spans. Effective only when LUMIGO_ENABLE_LOGS is set to true.

Execution Tags

Execution Tags allow you to dynamically add dimensions to your invocations so that they can be identified, searched for, and filtered in Lumigo. For example: in multi-tenanted systems, execution tags are often used to mark with the identifiers of the end-users that trigger them for analysis (e.g., Explore view) and alerting purposes.

Creating Execution Tags

In the Lumigo OpenTelemetry Distro for Python, execution tags are represented as span attributes and, specifically, as span attributes with the lumigo.execution_tags. prefix. For example, you could add an execution tag as follows:

from opentelemetry.trace import get_current_span

get_current_span().set_attribute('lumigo.execution_tags.foo','bar')

Notice that, using OpenTelemetry's get_current_span() API, you do not need to keep track of the current span, you can get it at any point of your program execution.

In OpenTelemetry, span attributes can be strings, numbers (double precision floating point or signed 64 bit integer), booleans (a.k.a. "primitive types"), and arrays of one primitive type (e.g., an array of string, and array of numbers or an array of booleans). In Lumigo, booleans and numbers are transformed to strings.

IMPORTANT: If you use the Span.set_attribute API multiple times on the same span to set values for the same key multiple values, you may override previous values rather than adding to them:

from opentelemetry.trace import get_current_span

get_current_span().set_attribute('lumigo.execution_tags.foo','bar')
get_current_span().set_attribute('lumigo.execution_tags.foo','baz')

In the snippets above, the foo execution tag will have in Lumigo only the baz value! Multiple values for an execution tag are supported as follows:

from opentelemetry.trace import get_current_span

get_current_span().set_attribute('lumigo.execution_tags.foo',['bar', 'baz'])

Tuples are also supported to specify multiple values for an execution tag:

from opentelemetry.trace import get_current_span

get_current_span().set_attribute('lumigo.execution_tags.bar',('baz','xyz',))

The snippets above will produce in Lumigo the foo tag having both bar and baz values. Another option to set multiple values is setting execution Tags in different spans of an invocation.

Execution Tags in different spans of an invocation

In Lumigo, multiple spans may be merged together into one invocation, which is the entry that you see, for example, in the Explore view. The invocation will include all execution tags on all its spans, and merge their values:

from opentelemetry import trace

trace.get_current_span().set_attribute('lumigo.execution_tags.foo','bar')

tracer = trace.get_tracer(__name__)

with tracer.start_as_current_span('child_span') as child_span:
    child_span.set_attribute('lumigo.execution_tags.foo','baz')

In the examples above, the invocation in Lumigo resulting from executing the code will have both bar and baz values associated with the foo execution tag. Which spans are merged in the same invocation depends on the parent-child relations among those spans. Explaining this topic is outside the scope of this documentation; a good first read to get deeper into the topic is the Traces documentation of OpenTelemetry. In case your execution tags on different spans appear on different invocations than what you would expect, get in touch with Lumigo support.

Execution Tag Limitations

  • Up to 50 execution tag keys per invocation in Lumigo, irrespective of how many spans are part of the invocation or how many values each execution tag has.
  • The key of an execution tag cannot contain the . character; for example: lumigo.execution_tags.my.tag is not a valid tag. The OpenTelemetry Span.set_attribute() API will not fail or log warnings, but that will be displayed as my in Lumigo.
  • Each execution tag key can be at most 50 characters long; the lumigo.execution_tags. prefix does not count against the 50 characters limit.
  • Each execution tag value can be at most 70 characters long.

Programmatic Errors

Programmatic Errors allow you to customize errors, monitor and troubleshoot issues that should not necessarily interfere with the service. For example, an application tries to remove a user who doesn't exist. These custom errors can be captured by adding just a few lines of additional code to your application.

Programmatic Errors indicating that a non-fatal error occurred, such as an application error. You can log programmatic errors, track custom error issues, and trigger Alerts.

Creating a Programmatic Error

Programmatic errors are created by adding span events with a custom attribute being set with the key name lumigo.type.

For example, you could add a programmatic error as follows:

from opentelemetry.trace import get_current_span

get_current_span().add_event('<error-message>', {'lumigo.type': '<error-type>'})

Supported runtimes

  • cpython: 3.7.x, 3.8.x, 3.9.x, 3.10.x, 3.11.x

Supported packages

Instrumentation Package Supported Versions
3.7 3.8 3.9 3.10 3.11
botocore boto3 1.17.22~1.33.13 1.17.22~1.34.144 1.17.22~1.34.144 1.17.22~1.34.144 1.17.22~1.34.144
django django 3.2.1~3.2.25 3.2.1~3.2.25 3.2.1~3.2.25 3.2.1~3.2.25 3.2.1~3.2.25
3.2 4.0.1~4.2.14 4.0.1~4.2.14 4.0.1~4.2.14 4.0.1~4.2.14
3.2 3.2 5.0.1~5.1.1 5.0.1~5.1.1
4.0 4.0 3.2 3.2
4.0.a1 4.0.a1 4.0 4.0
4.0.b1 4.0.b1 4.0.a1 4.0.a1
4.0.rc1 4.0.rc1 4.0.b1 4.0.b1
4.1 4.1 4.0.rc1 4.0.rc1
4.1.a1 4.1.a1 4.1 4.1
4.1.b1 4.1.b1 4.1.a1 4.1.a1
4.1.rc1 4.1.rc1 4.1.b1 4.1.b1
4.1rc1 4.1rc1 4.1.rc1 4.1.rc1
4.2 4.2 4.1rc1 4.1rc1
4.2.a1 4.2.a1 4.2 4.2
4.2.b1 4.2.b1 4.2.a1 4.2.a1
4.2.rc1 4.2.rc1 4.2.b1 4.2.b1
4.2rc1 4.2rc1 4.2.rc1 4.2.rc1
4.2rc1 4.2rc1
5.0 5.0
5.0rc1 5.0rc1
fastapi uvicorn 0.11.3~0.22.0 0.11.3~0.22.0 0.11.3~0.22.0 0.11.3~0.22.0 0.12.0~0.22.0
0.24.0~0.30.1 0.24.0~0.30.1 0.24.0~0.30.1 0.24.0~0.30.1
fastapi 0.56.1~0.100.0 0.56.1~0.100.0 0.56.1~0.100.0 0.56.1~0.100.0 0.56.1~0.100.0
0.100.0b2~0.103.2 0.100.0b2~0.111.0.dev1 0.100.0b2~0.111.0.dev1 0.100.0b2~0.111.0.dev1 0.100.0b2~0.111.0.dev1
flask flask 2.0.0~2.2.5 2.0.0~2.2.5 2.0.0~2.2.5 2.0.0~2.2.5 2.0.0~2.2.5
grpcio grpcio 1.45.0~1.62.2 1.45.0~1.65.0rc2 1.45.0~1.65.0rc2 1.45.0~1.65.0rc2 1.49.0~1.65.0rc2
kafka_python kafka_python 2.0.0~2.0.2 2.0.0~2.0.2 2.0.0~2.0.2 2.0.0~2.0.2 2.0.0~2.0.2
pika pika 1.0.0 1.0.0 1.0.0 1.0.0 1.0.0
1.0.1~1.3.0 1.0.1~1.3.0 1.0.1~1.3.0 1.0.1~1.3.0 1.0.1~1.3.0
1.3.0rc5~1.3.2 1.3.0rc5~1.3.2 1.3.0rc5~1.3.2 1.3.0rc5~1.3.2 1.3.0rc5~1.3.2
psycopg psycopg 3.1.1~3.1.20 3.1.1~3.2.1 3.1.1~3.2.1 3.1.1~3.2.1 3.1.1~3.2.1
3.1 3.1 3.1 3.1 3.1
psycopg-binary 3.1.1~3.1.20 3.1.1~3.2.1 3.1.1~3.2.1 3.1.1~3.2.1 3.1.4~3.2.1
3.1 3.1 3.1 3.1
psycopg2 psycopg2 2.7.5~2.9.9 2.8.1~2.9.9 2.8.1~2.9.9 2.8.1~2.8.6 2.9.5~2.9.9
2.8 2.8 2.8 2.9.5~2.9.9
2.9 2.9 2.9 2.8
psycopg2-binary 2.7.5~2.9.9 2.8.1~2.9.9 2.8.1~2.9.9 2.8.1~2.8.6 2.9.5~2.9.9
2.8 2.8 2.8 2.9.5~2.9.9
2.9 2.9 2.9 2.8
pymongo pymongo 3.1.1~3.3.1 3.1.1~3.3.1 3.1.1~3.3.1 3.1.1~3.3.1 3.1.1~3.3.1
3.5.0~3.13.0 3.5.0~3.13.0 3.5.0~3.13.0 3.5.0~3.13.0 3.5.0~3.13.0
4.0.1~4.7.3 4.0.1~4.8.0b0 4.0.1~4.8.0b0 4.0.1~4.8.0b0 4.0.1~4.8.0b0
3.1 3.1 3.1 3.1 3.1
3.2 3.2 3.2 3.2 3.2
4.0 4.0 4.0 4.0 4.0
pymysql pymysql 0.9.0~0.10.1 0.9.0~0.10.1 0.9.0~0.10.1 0.9.0~0.10.1 0.9.0~0.10.1
1.0.0~1.0.3 1.0.0~1.0.3 1.0.0~1.0.3 1.0.0~1.0.3 1.0.0~1.0.3
1.1.0~1.1.1 1.1.0~1.1.1 1.1.0~1.1.1 1.1.0~1.1.1 1.1.0~1.1.1
redis redis 4.1.1~4.2.0 4.1.1~4.2.0 4.1.1~4.2.0 4.1.1~4.2.0 4.1.1~4.2.0
4.2.1~4.6.0 4.2.1~4.6.0 4.2.1~4.6.0 4.2.1~4.6.0 4.2.1~4.6.0
5.0.0~5.1.0a1 5.0.0~5.1.0b7 5.0.0~5.1.0b7 5.0.0~5.1.0b7 5.0.0~5.1.0b7

Automated dependency reporting

To provide better support and better data-driven product decisions with respect to which packages to support next, the Lumigo OpenTelemetry Distro for Python will report to Lumigo SaaS on startup the packages and their versions used in this application, together with the OpenTelemetry resource data to enable analytics in terms of which platforms use which dependencies.

The data uploaded to Lumigo is a set of key-value pairs with package name and version. Similar is available through the tracing data sent to Lumigo, except that this aims at covering dependencies for which the Lumigo OpenTelemetry Distro for Python does not have instrumentation (yet?). Lumigo's only goal for these analytics data is to be able to give you the instrumentations you need without you needing to tell us!

The dependencies data is sent only when a LUMIGO_TRACER_TOKEN is present in the process environment, and it can be opted out via the LUMIGO_REPORT_DEPENDENCIES=false environment variable.

Baseline setup

The Lumigo OpenTelemetry Distro will automatically create the following OpenTelemetry constructs provided to a TraceProvider.

Resource attributes

SDK resource attributes

  • The attributes from the default resource:

    • telemetry.sdk.language: python
    • telemetry.sdk.name: opentelemetry
    • telemetry.sdk.version: depends on the version of the opentelemetry-sdk included in the dependencies

  • The lumigo.distro.version containing the version of the Lumigo OpenTelemetry Distro for Python as specified in the VERSION file

Process resource attributes

  • The following process.runtime.* attributes as specified in the Process Semantic Conventions:

    • process.runtime.description
    • process.runtime.name
    • process.runtime.version

  • A non-standard process.environ resource attribute, containing a stringified representation of the process environment, with environment variables scrubbed based on the LUMIGO_SECRET_MASKING_REGEX configuration.

Amazon ECS resource attributes

If the instrumented Python application is running on the Amazon Elastic Container Service (ECS):

  • cloud.provider attribute with value aws
  • cloud.platform with value aws_ecs
  • container.name with the hostname of the ECS Task container
  • container.id with the ID of the Docker container (based on the cgroup id)

If the ECS task uses the ECS agent v1.4.0, and has therefore access to the Task metadata endpoint version 4, the following experimental attributes as specified in the AWS ECS Resource Attributes specification:

  • aws.ecs.container.arn
  • aws.ecs.cluster.arn
  • aws.ecs.launchtype
  • aws.ecs.task.arn
  • aws.ecs.task.family
  • aws.ecs.task.revision

Kubernetes resource attributes

  • k8s.pod.uid with the Pod identifier, supported for both cgroups v1 and v2

Span exporters

SDK configuration

  • The following SDK environment variables are supported:

    • OTEL_SPAN_ATTRIBUTE_VALUE_LENGTH_LIMIT
    • OTEL_ATTRIBUTE_VALUE_LENGTH_LIMIT

    ** If the OTEL_SPAN_ATTRIBUTE_VALUE_LENGTH_LIMIT environment variable is not set, the span attribute size limit will be taken from OTEL_ATTRIBUTE_VALUE_LENGTH_LIMIT environment variable. The default size limit when both are not set is 2048.

Advanced use cases

Access to the TracerProvider

The Lumigo OpenTelemetry Distro provides access to the TracerProvider it configures (see the Baseline setup section for more information) through the tracer_provider attribute of the lumigo_opentelemetry package:

from lumigo_opentelemetry import tracer_provider

# Do here stuff like adding span processors

Ensure spans are flushed to Lumigo before shutdown

For short-running processes, the BatchProcessor configured by the Lumigo OpenTelemetry Distro may not ensure that the tracing data are sent to Lumigo (see the Baseline setup section for more information). Through the access to the tracer_provider, however, it is possible to ensure that all spans are flushed to Lumigo as follows:

from lumigo_opentelemetry import tracer_provider

# Do some logic

tracer_provider.force_flush()

# Now the Python process can terminate, with all the spans closed so far sent to Lumigo

Consuming SQS messages with Boto3 receive_message

Messaging instrumentations that retrieve messages from queues tend to be counter-intuitive for end-users: when retrieving one or more messages from the queue, one would naturally expect that all calls done using data from those messages, e.g., sending their content to a database or another queue, would result in spans that are children of the describing the retrieving of those messages.

Consider the following scenario, which is supported by the boto3 SQS receive_message instrumentation of the Lumigo OpenTelemetry Distro for Python:

from opentelemetry import trace

tracer = trace.get_tracer(__name__)

response = client.receive_message(...)  # Instrumentation creates a `span_0` span

for message in response.get("Messages", []):
  # The SQS.ReceiveMessage span is active in this scope
  with tracer.start_as_current_span("span_1"):  # span_0 is the parent of span_1
    do_something()

Without the scope provided by the iterator over response["Messages"], span_1 would be without a parent span, and that would result in a separate invocation and a separate transaction in Lumigo.

Filtering out empty SQS messages

A common pattern in SQS-based applications is to continuously poll an SQS queue for messages, and to process them as they arrive. In order not to clutter the Lumigo platform with empty SQS polling messages, the default behavior is to filter them out from being sent to Lumigo.

You can change this behavior by setting the boolean environment variable LUMIGO_AUTO_FILTER_EMPTY_SQS to false. The possible variations are:

  • LUMIGO_AUTO_FILTER_EMPTY_SQS=true filter out empty SQS polling messages
  • LUMIGO_AUTO_FILTER_EMPTY_SQS=false do not filter out empty SQS polling messages
  • No environment variable set (default): filter out empty SQS polling messages

Filtering http endpoints

You can selectively filter spans based on HTTP server/client endpoints for various components, not limited to web frameworks.

Global filtering

Set the LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX environment variable to a list of regex strings. Spans with matching server/client endpoints will not be traced.

Specific Filtering

For exclusive server (inbound) or client (outbound) span filtering, use the environment variables:

  • LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX_SERVER
  • LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX_CLIENT

Notes:

  • the environment variable must be a valid JSON array of strings, so if you want to match endpoint with the hostname google.com the environment variable value should be ["google\\.com"].
  • If we are filtering out an HTTP call to an opentelemetry traced component, every subsequent invocation made by that component won't be traced either.

Examples:

  • Filtering out every incoming HTTP request to the /login endpoint (will also match requests such as /login?user=foo, /login/bar))):
    • LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX_SERVER=["\\/login"]
  • Filtering out every outgoing HTTP request to the google.com domain (will also match requests such as google.com/foo, bar.google.com):
    • LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX_CLIENT=["google\\.com"]'
  • Filtering out every outgoing HTTP request to https://www.google.com (will also match requests such as https://www.google.com/, https://www.google.com/foo)
    • LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX_CLIENT=["https:\\/\\/www\\.google\\.com"]
  • Filtering out every HTTP request (incoming or outgoing) with the word login:
    • LUMIGO_FILTER_HTTP_ENDPOINTS_REGEX=["login"]

Contributing

For guidelines on contributing, please see CONTRIBUTING.md.