> ## Documentation Index
> Fetch the complete documentation index at: https://docs.strait.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Adaptive Concurrency

> Dynamic worker pool sizing based on system load and throughput.

Adaptive concurrency dynamically adjusts the worker pool size between configurable min/max bounds based on observed throughput and error rates. This ensures optimal resource utilization while protecting downstream services from being overwhelmed during traffic spikes.

## Overview

The adaptive concurrency feature allows Strait to scale its execution capacity up or down automatically. Instead of a fixed number of workers, the system observes the health of the job execution environment and adjusts the concurrency level in real-time. This prevents the "thundering herd" problem and ensures that the system remains responsive under heavy load.

## How It Works

The system monitors job completion rates, error rates, and queue depth to determine the ideal concurrency level for the current environment.

* **Scaling Up**: When throughput is high and error rates remain low, the system gradually increases the number of concurrent workers to maximize processing speed.
* **Scaling Down**: When errors spike or throughput drops significantly, the system aggressively reduces concurrency to alleviate pressure on the system.
* **Algorithm**: It uses an Additive Increase, Multiplicative Decrease (AIMD) algorithm, similar to TCP congestion control, to find the stable operating point for the current workload.

## Configuration

The following environment variables control the behavior of the adaptive concurrency engine:

| Variable                   | Default              | Description                               |
| :------------------------- | :------------------- | :---------------------------------------- |
| `ADAPTIVE_CONCURRENCY_MIN` | 1                    | The minimum number of concurrent workers. |
| `ADAPTIVE_CONCURRENCY_MAX` | `WORKER_CONCURRENCY` | The maximum number of concurrent workers. |

## When to Use

Adaptive concurrency is best suited for workloads with variable load patterns or those that interact with external APIs that have rate limits.

* **Variable Load**: Automatically scales down during quiet periods to save resources and scales up during spikes.
* **Downstream Protection**: Prevents overloading downstream services or databases by backing off when error rates increase.
* **Resource Efficiency**: Allows for more efficient resource usage by only running the necessary number of workers for the current demand.

## Related Concepts

* [Circuit Breaker](/concepts/resilience)
* [Environment Variables](/configuration/environment-variables)