Research integrating auto- and manual instrumentations

[Oberon00]

The https://github.com/open-telemetry/opentelemetry-auto-instr-python/ (currently actually https://github.com/lightstep/opentelemetry-auto-instr-python/) repository contains an auto-injecting & auto-instrumenting agent (injection = executing code automatically at interpreter startup; instrumentation = modifying code to automatically create traces at runtime). We also currently have some integrations in our ext/ folder (e.g. for WSGI, requests, mongodb) and in PRs (MySQL, PostreSQL, flask, ...). We should look into how we can integrate these with the agent. Probably we should start by looking at https://github.com/lightstep/opentelemetry-auto-instr-python/tree/master/oteltrace/contrib and comparing our existing (or proposed) integrations with the packages there.

It seems that the interface that the agent uses is a single parameterless patch function: https://github.com/lightstep/opentelemetry-auto-instr-python/blob/9c401c58a54640891c9c59f9f69663288e93e6e7/oteltrace/monkey.py#L186-L194

Basically we can start out with that exact interface and look whether our or the auto-integration looks like a better starting point. Then we can integrate any parts that are still better from one in the other and make sure it follows the auto-instrumentation interface and OpenTracing semantic conventions.

Some integrations, e.g., wsgi, may need special handling (there is no equivalent to patch in ext-wsgi, but it might be usable from contrib.django.patch.)

[toumorokoshi]

@ocelotl has made a lot of progress in this area, and I think feedback can also be sent over to his first PR with a proof of concept: #327

[c24t]

See open-telemetry/opentelemetry-python-contrib#1, which adds datadog's python integrations to the new repo.