Add multithreading doc

SUMMARY.md

@@ -67,6 +67,7 @@
 * [Networking](administration/networking.md)
 * [Memory Management](administration/memory-management.md)
 * [Monitoring](administration/monitoring.md)
+* [Multithreading](administration/multithreading.md)
 * [HTTP Proxy](administration/http-proxy.md)
 * [Hot Reload](administration/hot-reload.md)
 * [Troubleshooting](administration/troubleshooting.md)

administration/multithreading.md

@@ -0,0 +1,45 @@
+---
+description: Learn how to run Fluent Bit in multiple threads for improved scalability.
+---
+
+# Multithreading
+
+Fluent Bit has one event loop to handle critical operations, like managing
+timers, receiving internal messages, scheduling flushes, and handling retries.
+This event loop runs in Fluent Bit's main thread.
+
+To free up resources in the main thread, you can configure
+[inputs](../pipeline/inputs/README.md) and [outputs](../pipeline/outputs/README.md)
+to run in their own self-contained threads. However, inputs and outputs implement
+multithreading in distinct ways: inputs can run in **threaded** mode, and outputs
+can use one or more **workers**.
+
+## Inputs
+
+When inputs collect telemetry data, they can either perform this process
+inside Fluent Bit's main thread or inside a separate dedicated thread. You can
+configure this behavior by enabling or disabling the `threaded` setting.
+
+All inputs are capable of running in threaded mode, but certain inputs always
+run in threaded mode regardless of configuration. These always-threaded inputs are:
+
+- [Kubernetes Events](../pipeline/inputs/kubernetes-events.md)
+- [Node Exporter Metrics](../pipeline/inputs/node-exporter-metrics.md)
+- [Process Exporter Metrics](../pipeline/inputs/process-exporter-metrics.md)
+- [Windows Exporter Metrics](../pipeline/inputs/windows-exporter-metrics.md)
+
+Inputs are not internally aware of multithreading. If an input runs in threaded
+mode, Fluent Bit manages the logistics of that input's thread.
+
+## Outputs
+
+When outputs flush data, they can either perform this operation inside Fluent Bit's
+main thread or inside a separate dedicated thread called a _worker_. Each output
+can have one or more workers running in parallel, and each worker can handle multiple
+concurrent flushes. You can configure this behavior by changing the value of the
+`workers` setting.
+
+All outputs are capable of running in multiple workers, and each output has a
+a default value of `0`, `1`, or `2` workers. However, even if an output uses
+workers by default, you can safely reduce the number of workers below the default
+or disable workers entirely.