From 6795243d5f4e4b412300ddf86612d062ac88766d Mon Sep 17 00:00:00 2001
From: reverse <mike.chevronnet@gmail.com>
Date: Wed, 12 Aug 2026 12:43:41 +0000
Subject: [PATCH] core: slow-command snotice (slow_command_ms) + a watchdog
 thread (watchdog_ms) so a blocked core thread is visible instead of a silent
 freeze

---
 echoircd.conf.example |  4 ++++
 src/ircd.rs           | 17 ++++++++++++++++-
 src/main.rs           | 26 ++++++++++++++++++++++++--
 3 files changed, 44 insertions(+), 3 deletions(-)

diff --git a/echoircd.conf.example b/echoircd.conf.example
index e1a1527..94e0152 100644
--- a/echoircd.conf.example
+++ b/echoircd.conf.example
@@ -132,6 +132,10 @@ amu_target    = both
 # connectclass_required = yes    # refuse clients that match no allow class (default no)
 
 # --- global connection limits (per-class recvq/hardsendq/softsendq override these) ---
+# --- core-thread health (single-threaded core: nothing slow may run inline) ---
+# slow_command_ms = 200   # snotice when one event takes at least this long (0 = off)
+# watchdog_ms     = 5000  # a thread warns to the log if the core is stuck this long (0 = off)
+
 # max_line  = 16384            # max bytes in one line / receive queue (default 16 KiB)
 # max_sendq = 1048576          # max queued output before a slow client is dropped (1 MiB)
 
diff --git a/src/ircd.rs b/src/ircd.rs
index fb8e1bd..39945a1 100644
--- a/src/ircd.rs
+++ b/src/ircd.rs
@@ -4,7 +4,10 @@
 
 use std::collections::HashMap;
 use std::net::{SocketAddr, TcpStream};
+use std::sync::atomic::{AtomicU64, Ordering};
 use std::sync::mpsc::{Receiver, Sender};
+use std::sync::Arc;
+use std::time::Instant;
 
 use crate::command::Command;
 use crate::config::Config;
@@ -118,14 +121,26 @@ impl Ircd {
     /// server down with it. State touched before the panic may be left inconsistent,
     /// so this is a last-resort safety net, not a licence to panic — the untrusted
     /// parsers are still written so they can't panic in the first place.
-    pub fn run(mut self, rx: Receiver<Event>) {
+    /// `busy` is a shared marker the watchdog thread samples: it holds the ms-since-
+    /// `base` at which the current event started (0 = idle), so a stuck handler is
+    /// visible from outside. Events slower than `slow_command_ms` are also snoticed.
+    pub fn run(mut self, rx: Receiver<Event>, busy: Arc<AtomicU64>, base: Instant) {
+        let slow_ms = self.server.conf_num("slow_command_ms", 200u64);
         for ev in rx {
+            busy.store((base.elapsed().as_millis() as u64).max(1), Ordering::Relaxed);
+            let start = Instant::now();
             if std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| self.handle_event(ev)))
                 .is_err()
             {
                 // the default panic hook already logged the details to stderr
                 eprintln!("[core] recovered from a panicking event handler; continuing");
             }
+            busy.store(0, Ordering::Relaxed);
+            let ms = start.elapsed().as_millis() as u64;
+            if slow_ms != 0 && ms >= slow_ms {
+                self.server
+                    .snotice(&format!("slow event: a command took {ms}ms on the core thread"));
+            }
         }
     }
 
diff --git a/src/main.rs b/src/main.rs
index 132b5ad..067b610 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -4,10 +4,11 @@
 #![forbid(unsafe_code)]
 
 use std::net::TcpListener;
-use std::sync::atomic::AtomicU64;
+use std::sync::atomic::{AtomicU64, Ordering};
 use std::sync::mpsc;
 use std::sync::Arc;
 use std::thread;
+use std::time::{Duration, Instant};
 
 use echoircd::config::Config;
 use echoircd::ircd::{Event, Ircd};
@@ -67,7 +68,28 @@ fn main() {
     let core_cfg = cfg.clone();
     let core_tx = tx.clone(); // the core self-injects events (DNS results)
     let core_counter = counter.clone();
-    let core = thread::spawn(move || Ircd::new(core_cfg, core_tx, core_counter).run(rx));
+    // watchdog: the core stores when it started the current event into `core_busy`
+    // (0 = idle); a separate thread warns if it stays stuck past `watchdog_ms`.
+    let wd_base = Instant::now();
+    let core_busy = Arc::new(AtomicU64::new(0));
+    let watchdog_ms = raw_num("watchdog_ms", 5000) as u64; // 0 = off
+    if watchdog_ms > 0 {
+        let (wb, base) = (core_busy.clone(), wd_base);
+        thread::spawn(move || loop {
+            thread::sleep(Duration::from_millis(1000));
+            let cur = wb.load(Ordering::Relaxed);
+            if cur != 0 {
+                let stuck = (base.elapsed().as_millis() as u64).saturating_sub(cur);
+                if stuck > watchdog_ms {
+                    eprintln!(
+                        "[watchdog] core thread stuck ~{stuck}ms on one event — a handler is blocking the whole server"
+                    );
+                }
+            }
+        });
+    }
+    let (busy, base) = (core_busy, wd_base);
+    let core = thread::spawn(move || Ircd::new(core_cfg, core_tx, core_counter).run(rx, busy, base));
 
     // background timer: drives ping/idle timeouts
     let tick_tx = tx.clone();