fix: keep llama.cpp logs routed to the latest logger after recreating Llama

llama/addon/globals/addonLog.cpp

@@ -6,6 +6,7 @@
 
 AddonThreadSafeLogCallbackFunction addonThreadSafeLoggerCallback;
 bool addonJsLoggerCallbackSet = false;
+AddonThreadSafeLogCallbackFunctionContext* addonCurrentLoggerContext = nullptr;
 std::atomic<int> addonLoggerLogLevel(5);
 std::atomic<std::optional<int>> addonLoggerLogLevelOverride(std::nullopt);
 std::atomic<int> addonLastLoggerLogLevel(6);
@@ -103,17 +104,26 @@ void addonLlamaCppLogCallback(ggml_log_level level, const char* text, void* user
 }
 
 Napi::Value setLogger(const Napi::CallbackInfo& info) {
-    if (addonJsLoggerCallbackSet) {
-        addonJsLoggerCallbackSet = false;
-        addonThreadSafeLoggerCallback.Release();
-    }
-
     if (info.Length() < 1 || !info[0].IsFunction()) {
+        if (addonJsLoggerCallbackSet) {
+            addonJsLoggerCallbackSet = false;
+            addonCurrentLoggerContext = nullptr;
+            addonThreadSafeLoggerCallback.Release();
+        }
+
         return info.Env().Undefined();
     }
 
     auto addonLoggerJSCallback = info[0].As<Napi::Function>();
     AddonThreadSafeLogCallbackFunctionContext* context = new Napi::Reference<Napi::Value>(Napi::Persistent(info.This()));
+
+    // install the new logger before releasing the previous one, so that the previous logger's
+    // deferred finalizer cannot disable the new logger (the finalizer only disables logging
+    // when its context is still the current one)
+    const bool hadPreviousLogger = addonJsLoggerCallbackSet;
+    auto previousLoggerCallback = addonThreadSafeLoggerCallback;
+
+    addonCurrentLoggerContext = context;
     addonThreadSafeLoggerCallback = AddonThreadSafeLogCallbackFunction::New(
         info.Env(),
         addonLoggerJSCallback,
@@ -122,7 +132,10 @@ Napi::Value setLogger(const Napi::CallbackInfo& info) {
         1,
         context,
         [](Napi::Env, void*, AddonThreadSafeLogCallbackFunctionContext* ctx) {
-            addonJsLoggerCallbackSet = false;
+            if (addonCurrentLoggerContext == ctx) {
+                addonJsLoggerCallbackSet = false;
+                addonCurrentLoggerContext = nullptr;
+            }
 
             delete ctx;
         }
@@ -132,6 +145,9 @@ Napi::Value setLogger(const Napi::CallbackInfo& info) {
     // prevent blocking the main node process from exiting due to active resources
     addonThreadSafeLoggerCallback.Unref(info.Env());
 
+    if (hadPreviousLogger)
+        previousLoggerCallback.Release();
+
     return info.Env().Undefined();
 }
 

test/modelDependent/llama/logger.test.ts

@@ -0,0 +1,38 @@
+import {describe, expect, test} from "vitest";
+import {getModelFile} from "../../utils/modelFiles.js";
+import {getLlama, LlamaLogLevel} from "../../../src/index.js";
+
+describe("llama", () => {
+    describe("logger", () => {
+        test("a custom logger keeps receiving llama.cpp log lines after a second getLlama() call", {timeout: 1000 * 60 * 60 * 2}, async () => {
+            const modelPath = await getModelFile("bge-small-en-v1.5-q8_0.gguf");
+            const lines1: string[] = [];
+            const lines2: string[] = [];
+
+            const llama1 = await getLlama({
+                logLevel: LlamaLogLevel.debug,
+                logger(level, message) {
+                    lines1.push(message);
+                }
+            });
+            await llama1.loadModel({modelPath});
+
+            const llama2 = await getLlama({
+                logLevel: LlamaLogLevel.debug,
+                logger(level, message) {
+                    lines2.push(message);
+                }
+            });
+            await new Promise((resolve) => setTimeout(resolve, 100));
+            await llama2.loadModel({modelPath});
+
+            // llama.cpp emits native log lines through the logger of the most recently created instance.
+            // Only TS-side logs would reach `lines2` if the native log sink was broken.
+            expect(lines1.some((line) => line.includes("llama_model_loader"))).toBe(true);
+            expect(lines2.some((line) => line.includes("llama_model_loader"))).toBe(true);
+
+            await llama2.dispose();
+            await llama1.dispose();
+        });
+    });
+});