swactor/tests/stress/saturation.rs

//! Saturation stress tests - find where the runtime breaks.
//!
//! These tests intentionally push past limits to document failure modes.

use super::{BlackHole, Counter, Msg, Stress, StressResult};
use std::time::Duration;
use swactor::runtime::{Runtime, RuntimeConfig};

/// FIXME: does this even make sense to test?
/// Blast the router inbox until it overflows.
/// Documents: What happens when router can't keep up?
#[test]
#[cfg(feature = "stress")]
fn router_inbox_overflow() {
    println!("\n>>> STRESS: Router Inbox Overflow");

    let config = RuntimeConfig {
        max_actors: 10,
        router_max_messages: 100, // Tiny buffer
        actor_max_messages: 1000,
        num_threads: 1,
    };
    let runtime = Runtime::new(config);
    let sink = runtime.spawn(BlackHole).unwrap();

    // Blast messages without processing
    let mut result = StressResult::new("router_inbox_overflow");
    let start = std::time::Instant::now();

    for _ in 0..10_000 {
        result.operations += 1;
        if runtime.send_to::<Msg>(sink, Msg).is_ok() {
            result.successes += 1;
        } else {
            result.failures += 1;
        }
    }

    result.duration = start.elapsed();
    result.note(format!("Router buffer: 100, Messages sent: 10,000"));
    result.note(format!(
        "Expected: ~99% failure rate (buffer fills immediately)"
    ));
    result.print();

    // Verify we actually saw failures
    assert!(result.failures > 0, "Expected router to reject messages");
    println!(">>> PASS: Router correctly rejects messages when full\n");
}

/// FIXME: This test makes no sense until we make sure panics happen when
/// actor inbox buffers are full.
/// Blast a single actor's inbox until it overflows.
/// Documents: What happens when actor can't keep up?
#[test]
#[cfg(feature = "stress")]
fn actor_inbox_overflow() {
    println!("\n>>> STRESS: Actor Inbox Overflow");

    let config = RuntimeConfig {
        max_actors: 10,
        router_max_messages: 100_000, // Large router buffer
        actor_max_messages: 100,      // Tiny actor inbox
        num_threads: 1,
    };
    let runtime = Runtime::new(config);
    let sink = runtime.spawn(BlackHole).unwrap();

    // Process router registration
    runtime.tick();

    // Now blast messages - router will accept them but actor inbox will fill
    let mut sent = 0u64;
    for _ in 0..10_000 {
        if runtime.send_to::<Msg>(sink, Msg).is_ok() {
            sent += 1;
        }
        // Tick occasionally to let router deliver
        if sent % 100 == 0 {
            runtime.tick();
        }
    }

    // The router accepted messages, but many were dropped at actor inbox
    // We can't easily count these drops from outside, but we can document the behavior
    println!("  Router accepted {} messages", sent);
    println!("  Actor inbox capacity: 100");
    println!("  Note: Messages beyond inbox capacity are silently dropped");
    println!(">>> This is a known limitation - bounded queues drop overflow\n");
}

/// FIXME: Does it make sense to have this as a test? Yes the runtime
/// fails if you try and spawn actors when the queue is full.
/// Spawn actors until the queue rejects.
/// Documents: What happens when actor queue fills?
#[test]
#[cfg(feature = "stress")]
fn actor_queue_overflow() {
    println!("\n>>> STRESS: Actor Queue Overflow");

    let config = RuntimeConfig {
        max_actors: 100, // Small actor queue
        router_max_messages: 10_000,
        actor_max_messages: 100,
        num_threads: 1,
    };
    let runtime = Runtime::new(config);

    let mut result = StressResult::new("actor_queue_overflow");
    let start = std::time::Instant::now();

    // Try to spawn 500 actors into 100-slot queue
    for _ in 0..500 {
        result.operations += 1;
        match runtime.spawn(BlackHole) {
            Ok(_) => result.successes += 1,
            Err(_) => result.failures += 1,
        }
    }

    result.duration = start.elapsed();
    result.note(format!("Queue capacity: 100, Spawn attempts: 500"));
    result.note(format!(
        "Expected: ~80% failure rate (queue fills after ~100)"
    ));
    result.print();

    // Note: Router also takes a slot, so we expect ~99 actors max
    assert!(
        result.successes <= 100,
        "Spawned more actors than queue capacity"
    );
    assert!(result.failures > 0, "Expected spawn failures");
    println!(">>> PASS: Actor queue correctly rejects when full\n");
}

/// FIXME: IS this actually testing what it should be?
/// Sustained overload - run at 2x capacity for extended period.
/// Documents: Does the system degrade gracefully or crash?
#[test]
#[cfg(feature = "stress")]
fn sustained_overload() {
    println!("\n>>> STRESS: Sustained Overload");

    let config = RuntimeConfig {
        max_actors: 100,
        router_max_messages: 1000,
        actor_max_messages: 100,
        num_threads: 1,
    };
    let runtime = Runtime::new(config);

    // Spawn some actors
    let mut actors = Vec::new();
    for _ in 0..50 {
        if let Ok(addr) = runtime.spawn(Counter::new()) {
            actors.push(addr);
        }
    }

    // Process registrations
    for _ in 0..200 {
        runtime.tick();
    }

    let result = Stress::new("sustained_overload")
        .for_duration(Duration::from_secs(2))
        .run(|| {
            // Send to random actor
            let idx = (std::time::Instant::now().elapsed().as_nanos() as usize) % actors.len();
            let success = runtime.send_to::<Msg>(actors[idx], Msg).is_ok();

            // Process some (but not all) - simulating overload
            runtime.tick();

            success
        });

    result.print();
    println!(">>> System survived sustained overload without panic\n");
}

/// FIXME: The logic for timing recovery does not make sense
/// Burst traffic - idle to 100x normal, back to idle.
/// Documents: Recovery behavior after traffic spikes.
#[test]
#[cfg(feature = "stress")]
fn burst_traffic() {
    println!("\n>>> STRESS: Burst Traffic");

    let config = RuntimeConfig {
        max_actors: 100,
        router_max_messages: 10_000,
        actor_max_messages: 1000,
        num_threads: 1,
    };
    let runtime = Runtime::new(config);
    let sink = runtime.spawn(Counter::new()).unwrap();

    // Process registration
    for _ in 0..10 {
        runtime.tick();
    }

    let mut total_sent = 0u64;
    let mut total_failed = 0u64;

    // 5 burst cycles
    for cycle in 0..5 {
        // Burst: send 1000 messages as fast as possible
        let mut burst_sent = 0;
        let mut burst_failed = 0;
        for _ in 0..1000 {
            if runtime.send_to::<Msg>(sink, Msg).is_ok() {
                burst_sent += 1;
            } else {
                burst_failed += 1;
            }
        }
        total_sent += burst_sent;
        total_failed += burst_failed;

        // Recovery: process until queue is drained
        let recovery_start = std::time::Instant::now();
        for _ in 0..5000 {
            runtime.tick();
        }
        let recovery_time = recovery_start.elapsed();

        println!(
            "  Cycle {}: sent={}, failed={}, recovery={:?}",
            cycle + 1,
            burst_sent,
            burst_failed,
            recovery_time
        );
    }

    println!(
        "\n  Total sent: {}, Total failed: {}",
        total_sent, total_failed
    );
    println!(">>> Burst traffic test complete\n");
}

/// Find the message drop cliff - at what load factor do drops spike?
#[test]
#[cfg(feature = "stress")]
fn message_drop_curve() {
    println!("\n>>> STRESS: Message Drop Curve");
    println!("  Testing drop rate at various load factors...\n");

    // Test at different load factors (messages per tick)
    for msgs_per_tick in [1, 5, 10, 20, 50, 100] {
        let config = RuntimeConfig {
            max_actors: 10,
            router_max_messages: 1000,
            actor_max_messages: 500,
            num_threads: 1,
        };
        let runtime = Runtime::new(config);
        let sink = runtime.spawn(BlackHole).unwrap();

        // Warmup
        for _ in 0..10 {
            runtime.tick();
        }

        let mut sent = 0u64;
        let mut failed = 0u64;

        // Run for fixed iterations
        for _ in 0..100 {
            // Send burst
            for _ in 0..msgs_per_tick {
                if runtime.send_to::<Msg>(sink, Msg).is_ok() {
                    sent += 1;
                } else {
                    failed += 1;
                }
            }
            // Process one tick
            runtime.tick();
        }

        let drop_rate = if sent + failed > 0 {
            (failed as f64 / (sent + failed) as f64) * 100.0
        } else {
            0.0
        };

        println!(
            "  msgs/tick={:3}  sent={:5}  failed={:5}  drop_rate={:.1}%",
            msgs_per_tick, sent, failed, drop_rate
        );
    }

    println!("\n>>> Message drop curve test complete\n");
}