[BUG]: Document more differences between ddtrace.auto and patch_all

[amacfie-tc]

Tracer Version(s)

2.18

Python Version(s)

3.12

Pip Version(s)

24.3.1

Bug Report

The docs say that ddtrace.auto or patch_all can be used for instrumentation but there's no hint that exception replay only works for ddtrace.auto, not patch_all. This was a big time waster. Additionally, it would be nice to enable exception replay for patch_all.

Reproduction Code

No response

Error Logs

No response

Libraries in Use

No response

Operating System

No response

[mabdinur]

Hi @amacfie-tc,

Sorry about the confusion. ddtrace.patch_all(..) was introduced 8 years ago. Back then the ddtrace library only supported tracing and patch_all was designed to enable tracing integrations (ex: automatic tracing for fastapi, django, postgres, ...). ddtrace.auto was introduced in v2.2.0, it's primary purpose was to provide a programmatic way to configure/load all ddtrace products (tracing integrations via ddtrace.patch_all(...), exception replay, profiling, dynamic instrumentation, application security monitoring, etc.).

In all most all cases either ddtrace-run or import ddtrace.auto should be used. I will update the docs to make this more clear. We are currently considering deprecating ddtrace.patch_all(....). I will share this feedback with the team.

[amacfie-tc]

Hi @amacfie-tc,

Sorry about the confusion. ddtrace.patch_all(..) was introduced 8 years ago. Back then the ddtrace library only supported tracing and patch_all was designed to enable tracing integrations (ex: automatic tracing for fastapi, django, postgres, ...). ddtrace.auto was introduced in v2.2.0, it's primary purpose was to provide a programmatic way to configure/load all ddtrace products (tracing integrations via ddtrace.patch_all(...), exception replay, profiling, dynamic instrumentation, application security monitoring, etc.).

In all most all cases either ddtrace-run or import ddtrace.auto should be used. I will update the docs to make this more clear. We are currently considering deprecating ddtrace.patch_all(....). I will share this feedback with the team.

If we use ddtrace.auto, the way to exclude a library from instrumentation is via the DD_TRACE_<INTEGRATION>_ENABLED environment variables?

[mabdinur]

This map shows which integrations are enabled/disabled when patch_all is called by ddtrace-run or import ddtrace.auto:

dd-trace-py/ddtrace/_monkey.py

Line 27 in a225334

PATCH_MODULES = {

.

To override these defaults you can use DD_TRACE_<INTEGRATION>_ENABLED or DD_PATCH_MODULES if you would like to override multiple integrations.