Feature: Adaptive Throttler

[kxrxh]

Problem

Fixed-rate throttlers don't work well on resource-constrained edge devices. They either crash when memory runs out or waste performance when resources are available.

Proposal

Add an AdaptiveThrottler that monitors system resources (memory, CPU, goroutines) and automatically adjusts processing speed based on what's actually available. No manual tuning needed.

Use Cases

• Edge/fog computing with limited resources
• Long-running streams with variable load
• Autonomous systems needing self-regulating backpressure

I'm interested in implementing this. Would make go-streams much more practical for real-world edge deployments.
Happy to discuss the approach if you think this fits with the project!

[reugn]

@kxrxh, a custom adaptive throttling mechanism can be used in a transforming flow or in a source or sink. Could you elaborate on your idea for implementing an adaptive throttler flow?

[kxrxh]

@reugn So, I am thinking of implementing it as a Flow operator (like your Throttler).

Two parts:

• AdaptiveThrottler (Flow): same in/out channel pattern; enforces a per-period quota derived from a dynamic “currentRate.”
• ResourceMonitor: periodically samples memory (runtime.ReadMemStats), CPU, and goroutines; exposes current stats for the throttler to read.

Adaptation: if memory/CPU exceed thresholds, multiplicative decrease (e.g., -20%); if below, increase proportional to headroom; apply smoothing and clamp within [MinThroughput, MaxThroughput].

Quota: elements are counted per period (e.g., 1s). When quota is hit, sending waits for a quota-reset signal (ticker resets counter).

Backpressure: bounded buffer; when full, upstream blocks (or we can support a discard mode like the fixed Throttler).