Skip to main content
Help us improve PyPI by participating in user testing. All experience levels needed!

StackImpact Python Profiler

Project description

Overview

StackImpact is a production-grade performance profiler built for both production and development environments. It gives developers continuous and historical code-level view of application performance that is essential for locating CPU, memory allocation and I/O hot spots as well as latency bottlenecks. Included runtime metrics and error monitoring complement profiles for extensive performance analysis. Learn more at stackimpact.com.

dashboard

dashboard

Features

  • Continuous hot spot profiling of CPU usage, memory allocation and blocking calls.
  • TensorFlow profiling.
  • Error and exception monitoring.
  • Health monitoring including CPU, memory, garbage collection and other runtime metrics.
  • Alerts on profile anomalies.
  • Team access.

Learn more on the features page (with screenshots).

How it works

The StackImpact profiler agent is imported into a program and used as a normal package. When the program runs, various sampling profilers are started and stopped automatically by the agent and/or programmatically using the agent methods. The agent periodically reports recorded profiles and metrics to the StackImpact Dashboard. If an application has multiple processes, also referred to as workers, instances or nodes, only one or two processes will have active agents at any point of time.

Documentation

See full documentation for reference.

Supported environment

  • Linux, OS X or Windows. Python version 2.7, 3.4 or higher.
  • Memory allocation profiler and some GC metrics are only available for Python 3.
  • Profilers only support Linux and OS X.
  • Time (blocking call) profiler supports threads and gevent.
  • On unix systems the profilers use the following signals: SIGPROF, SIGALRM, SIGUSR2. Only SIGUSR2 is handled transparently, i.e. it should not conflict with previousely registered handlers.

Getting started

Create StackImpact account

Sign up for a free trial account at stackimpact.com (also with GitHub login).

Installing the agent

Install the Python agent by running

pip install stackimpact

And import the package in your application

import stackimpact

Configuring the agent

Start the agent in the main thread by specifying the agent key and application name. The agent key can be found in your account’s Configuration section.

agent = stackimpact.start(
    agent_key = 'agent key here',
    app_name = 'MyPythonApp')

Add the agent initialization to the worker code, e.g. wsgi.py, if applicable.

All initialization options:

  • agent_key (Required) The access key for communication with the StackImpact servers.
  • app_name (Required) A name to identify and group application data. Typically, a single codebase, deployable unit or executable module corresponds to one application.
  • app_version (Optional) Sets application version, which can be used to associate profiling information with the source code release.
  • app_environment (Optional) Used to differentiate applications in different environments.
  • host_name (Optional) By default, host name will be the OS hostname.
  • auto_profiling (Optional) If set to False, disables automatic profiling and reporting. Focused or manual profiling should be used instead. Useful for environments without support for timers or background tasks.
  • debug (Optional) Enables debug logging.
  • cpu_profiler_disabled, allocation_profiler_disabled, block_profiler_disabled, error_profiler_disabled (Optional) Disables respective profiler when True.
  • include_agent_frames (Optional) Set to True to not exclude agent stack frames from profile call graphs.
  • auto_destroy (Optional) Set to False to disable agent’s exit handlers. If necessary, call destroy() to gracefully shutdown the agent.

Focused profiling

Use agent.profile(name) to instruct the agent when to start and stop profiling. The agent decides if and which profiler is activated. Normally, this method should be used in repeating code, such as request or event handlers. In addition to more precise profiling, timing information will also be reported for the profiled spans. Usage example:

span = agent.profile('span1');

# your code here

span.stop();

Alternatively, a with statement can be used:

with agent.profile('span1'):
    # your code ehere

Manual profiling

Manual profiling should not be used in production!

By default, the agent starts and stops profiling automatically. Manual profiling allows to start and stop profilers directly. It is suitable for profiling short-lived programs and should not be used for long-running production applications. Automatic profiling should be disabled with auto_profiling: False.

# Start CPU profiler.
agent.start_cpu_profiler();
# Stop CPU profiler and report the recorded profile to the Dashboard.
agent.stop_cpu_profiler();
# Start blocking call profiler.
agent.start_block_profiler();
# Stop blocking call profiler and report the recorded profile to the Dashboard.
agent.stop_block_profiler();
# Start heap allocation profiler.
agent.start_allocation_profiler();
# Stop heap allocation profiler and report the recorded profile to the Dashboard.
agent.stop_allocation_profiler();
# Start TensorFlow profiler.
agent.start_tf_profiler();
# Stop TensorFlow profiler and report the recorded profile to the Dashboard.
agent.stop_tf_profiler();

Analyzing performance data in the Dashboard

Once your application is restarted, you can start observing continuous CPU, memory, I/O, and other hot spot profiles, execution bottlenecks as well as process metrics in the Dashboard.

Troubleshooting

To enable debug logging, add debug = True to startup options. If the debug log doesn’t give you any hints on how to fix a problem, please report it to our support team in your account’s Support section.

Overhead

The agent overhead is measured to be less than 1% for applications under high load. For applications that are horizontally scaled to multiple processes, StackImpact agents are only active on a small subset of the processes at any point of time, therefore the total overhead is much lower.

Project details


Release history Release notifications

This version
History Node

1.2.3

History Node

1.2.2

History Node

1.2.1

History Node

1.2.0

History Node

1.1.5

History Node

1.1.4

History Node

1.1.3

History Node

1.1.2

History Node

1.1.1

History Node

1.1.0

History Node

1.0.5

History Node

1.0.4

History Node

1.0.3

History Node

1.0.2

History Node

1.0.1

History Node

1.0.0

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Filename, size & hash SHA256 hash help File type Python version Upload date
stackimpact-1.2.3.tar.gz (22.1 kB) Copy SHA256 hash SHA256 Source None May 9, 2018

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging CloudAMQP CloudAMQP RabbitMQ AWS AWS Cloud computing Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page