Make signal handlers async-signal-safe

src/Proxy.cpp

@@ -20,26 +20,23 @@
 #include "RequestParser.h"
 #include "Backtrace.h"
 
-static bool Running = false;
-static bool Abort = false;
-static bool Stop = false;
+static volatile sig_atomic_t Running = 0;
+static volatile sig_atomic_t AbortSignal = 0;
+static volatile sig_atomic_t StopSignal = 0;
 
 static void abortHandler(int sig)
 {
-    if (!Abort) {
-        traceInfo(sig);
-    }
-    Abort = true;
-    if (!Running) {
-        abort();
+    // Signal handlers must be async-signal-safe: only set flags here.
+    if (!AbortSignal) {
+        AbortSignal = sig;
     }
 }
 
 static void stopHandler(int sig)
 {
-    Stop = true;
-    if (!Running) {
-        abort();
+    // Signal handlers must be async-signal-safe: only set flags here.
+    if (!StopSignal) {
+        StopSignal = sig;
     }
 }
 
@@ -141,18 +138,21 @@ int Proxy::run()
     logNotice("predixy running with Name:%s Workers:%d",
             mConf->name(),
             (int)mHandlers.size());
+    Running = 1;
     std::vector<std::shared_ptr<std::thread>> tasks;
     for (auto h : mHandlers) {
         std::shared_ptr<std::thread> t(new std::thread([=](){h->run();}));
         tasks.push_back(t);
     }
-    Running = true;
     bool stop = false;
     while (!stop) {
-        if (Abort) {
+        if (AbortSignal) {
+            int sig = AbortSignal;
+            AbortSignal = 0;
+            traceInfo(sig);
             stop = true;
             abort();
-        } else if (Stop) {
+        } else if (StopSignal) {
             fprintf(stderr, "predixy will quit ASAP Bye!\n");
             stop = true;
         }

test/run.sh

@@ -144,6 +144,7 @@ TESTS=(
     "test/request_parser_error_log.py"
     "test/response_parser_error_log.py"
     "test/request_parser_long_command.py"
+    "test/signal_handling.py"
     "test/pubsub_long_name.py"
     "test/pubsub_large_message.py"
     "test/transaction_forbid.py"

test/signal_handling.py

@@ -0,0 +1,90 @@
+#!/usr/bin/env python3
+#
+# Start a temporary predixy instance and verify SIGTERM shuts it down cleanly.
+#
+
+import os
+import signal
+import socket
+import subprocess
+import tempfile
+import time
+
+from test_util import parse_args, exit_with_result
+
+
+def wait_for_port(host, port, timeout=5.0):
+    deadline = time.time() + timeout
+    while time.time() < deadline:
+        try:
+            with socket.create_connection((host, port), timeout=0.5):
+                return True
+        except Exception:
+            time.sleep(0.05)
+    return False
+
+
+def start_predixy(root, redis_port):
+    predixy_bin = os.path.join(root, "src", "predixy")
+    if not os.path.exists(predixy_bin):
+        raise RuntimeError("predixy binary not found")
+
+    listen_sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+    listen_sock.bind(("127.0.0.1", 0))
+    listen_port = listen_sock.getsockname()[1]
+    listen_sock.close()
+
+    tmp_dir = tempfile.TemporaryDirectory()
+    conf_path = os.path.join(tmp_dir.name, "predixy_test.conf")
+    with open(conf_path, "w") as f:
+        f.write(
+            "Name PredixySignalTest\n"
+            f"Bind 127.0.0.1:{listen_port}\n"
+            "WorkerThreads 1\n"
+            "ClientTimeout 3\n"
+            "LogVerbSample 0\n"
+            "LogDebugSample 0\n"
+            "LogInfoSample 10000\n"
+            "LogNoticeSample 1\n"
+            "LogWarnSample 1\n"
+            "LogErrorSample 1\n"
+            "\n"
+            "StandaloneServerPool {\n"
+            "    RefreshMethod fixed\n"
+            "    Group test {\n"
+            f"        + 127.0.0.1:{redis_port}\n"
+            "    }\n"
+            "}\n"
+        )
+
+    proc = subprocess.Popen([predixy_bin, conf_path],
+                            stdout=subprocess.DEVNULL,
+                            stderr=subprocess.DEVNULL)
+    if not wait_for_port("127.0.0.1", listen_port, timeout=5.0):
+        proc.terminate()
+        tmp_dir.cleanup()
+        raise RuntimeError("predixy did not start")
+
+    return proc, tmp_dir
+
+
+def run_test(project_root, redis_port):
+    proc, tmp_dir = start_predixy(project_root, redis_port)
+    try:
+        proc.send_signal(signal.SIGTERM)
+        try:
+            proc.wait(timeout=5.0)
+        except subprocess.TimeoutExpired:
+            print("FAIL: predixy did not exit after SIGTERM")
+            proc.kill()
+            return False
+    finally:
+        tmp_dir.cleanup()
+    return True
+
+
+if __name__ == "__main__":
+    args = parse_args("Signal handling test", default_port=6380)
+    root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
+    success = run_test(root, args.port)
+    exit_with_result(success, "signal handling", "signal handling")