feat(pipeline): detect node crashes and tally dropped frames

Register a KPN event handler in both scene_analyze and scene_preview:
  - Overflow events accumulate per-node dropped-frame counts, printed on exit.
  - A Closed event from any node other than result_sink at EOF means a stage
    died; trip an atomic so the main loop bails out instead of hanging on
    'done' forever, and exit non-zero.

Bumps external/KPN to the commit that exposes set_event_handler / NodeEvent.

src/main.cpp

@@ -48,8 +48,11 @@
 #include <chrono>
 #include <cstring>
 #include <iostream>
+#include <map>
+#include <mutex>
 #include <stdexcept>
 #include <string>
+#include <string_view>
 #include <thread>
 
 // ── CLI parsing ───────────────────────────────────────────────────────────────
@@ -192,17 +195,53 @@ int main(int argc, char** argv) {
     );
 #endif
 
+    // ── Pipeline observability (KPN event handler) ────────────────────────────
+    // Tally dropped frames per node (channel overflow) and detect a node that
+    // stops unexpectedly. A Closed event from any node other than result_sink at
+    // EOF means a stage crashed — without this the main loop below would hang on
+    // `done` forever, so we trip a flag to unblock it and exit non-zero.
+    std::mutex                  event_mtx;
+    std::map<std::string, long> overflow_counts;
+    std::atomic<bool>           node_crashed{false};
+
+    net.set_event_handler(
+        [&](std::string_view node_name, kpn::NodeEvent ev,
+            std::chrono::steady_clock::time_point) {
+            if (ev == kpn::NodeEvent::Overflow) {
+                std::lock_guard<std::mutex> lk(event_mtx);
+                ++overflow_counts[std::string(node_name)];
+            } else {  // NodeEvent::Closed
+                // result_sink closing once EOF has been signalled is the normal
+                // shutdown path, not a crash.
+                if (node_name == "result_sink" && done.load(std::memory_order_acquire))
+                    return;
+                std::cerr << "[main] node '" << node_name
+                          << "' stopped unexpectedly — aborting pipeline\n";
+                node_crashed.store(true, std::memory_order_release);
+            }
+        });
+
     // ── Run ───────────────────────────────────────────────────────────────────
-    // Note: StaticNetwork does not expose set_error_handler; node exceptions
-    // are printed to stderr by the KPN run_loop and the network continues.
     std::cerr << "[main] starting pipeline…\n";
     net.start();
 
-    // Main thread waits until ResultSinkFunc signals EOF completion
+    // Main thread waits until ResultSinkFunc signals EOF completion, or a node
-    while (!done.load(std::memory_order_acquire))
+    // crash trips node_crashed.
+    while (!done.load(std::memory_order_acquire) &&
+           !node_crashed.load(std::memory_order_acquire))
         std::this_thread::sleep_for(std::chrono::milliseconds(100));
 
     net.stop();
     net.print_diagnostics();
-    return 0;
+
+    {
+        std::lock_guard<std::mutex> lk(event_mtx);
+        if (!overflow_counts.empty()) {
+            std::cerr << "[main] dropped frames (channel overflow):\n";
+            for (const auto& [name, count] : overflow_counts)
+                std::cerr << "  " << name << ": " << count << "\n";
+        }
+    }
+
+    return node_crashed.load(std::memory_order_acquire) ? 1 : 0;
 }

src/scene_preview.cpp

@@ -35,8 +35,11 @@
 #include <chrono>
 #include <cstring>
 #include <iostream>
+#include <map>
+#include <mutex>
 #include <stdexcept>
 #include <string>
+#include <string_view>
 #include <thread>
 
 static Config parse_args(int argc, char** argv) {
@@ -143,12 +146,36 @@ int main(int argc, char** argv) {
         kpn::edge(tracker.output<"annotation">(),  sink.input<"annotation">())
     );
 
+    // ── Pipeline observability (KPN event handler) ────────────────────────────
+    // Tally dropped frames per node (channel overflow) and detect a node that
+    // stops unexpectedly so the preview loop below can bail out instead of
+    // spinning on a dead pipeline.
+    std::mutex                  event_mtx;
+    std::map<std::string, long> overflow_counts;
+    std::atomic<bool>           node_crashed{false};
+
+    net.set_event_handler(
+        [&](std::string_view node_name, kpn::NodeEvent ev,
+            std::chrono::steady_clock::time_point) {
+            if (ev == kpn::NodeEvent::Overflow) {
+                std::lock_guard<std::mutex> lk(event_mtx);
+                ++overflow_counts[std::string(node_name)];
+            } else {  // NodeEvent::Closed
+                if (node_name == "result_sink" && done.load(std::memory_order_acquire))
+                    return;
+                std::cerr << "[main] node '" << node_name
+                          << "' stopped unexpectedly — aborting pipeline\n";
+                node_crashed.store(true, std::memory_order_release);
+            }
+        });
+
     // ── Run ───────────────────────────────────────────────────────────────────
     std::cerr << "[main] starting pipeline — press q or Esc to quit early\n";
     net.start();
 
-    // Main thread drives the display window; returns false on EOF or q/Esc
+    // Main thread drives the display window; returns false on EOF or q/Esc.
-    while (preview.step()) {
+    // Bail out early if a node crashes.
+    while (!node_crashed.load(std::memory_order_acquire) && preview.step()) {
         cv::waitKey(1);  // pump OS events between frames
     }
 
@@ -156,5 +183,15 @@ int main(int argc, char** argv) {
     sink_fn.flush();  // write whatever was accumulated (no-op if EOF already flushed)
     cv::destroyAllWindows();
     net.print_diagnostics();
-    return 0;
+
+    {
+        std::lock_guard<std::mutex> lk(event_mtx);
+        if (!overflow_counts.empty()) {
+            std::cerr << "[main] dropped frames (channel overflow):\n";
+            for (const auto& [name, count] : overflow_counts)
+                std::cerr << "  " << name << ": " << count << "\n";
+        }
+    }
+
+    return node_crashed.load(std::memory_order_acquire) ? 1 : 0;
 }