Let's talk about usage of confluent-kafka-python Producer together with Tornado web server.

We know that Tornado is non blocking-server and has event loop for processing socket events.
We also know that librdkafka has several threads and its own event loop. And confluent-kafka-python is build on top of librdkafka.

When we call Producer.produce() from something inside Tornado it executes in main thread. Right? Then librdkafka pass data (with callback, probably) to another thread (event loop thread). And when message succefully recieved by Kafka, callback executed.

What happens when librdkafka call callback function from its event loop? Does it lock Python main thread with GIL?