Don't read from timerfd and eventfd on each EventLoop tick (#9192)

Motivation:

We do not need to issue a read on timerfd and eventfd when the EventLoop wakes up if we register these as Edge-Triggered. This removes the overhead of 2 syscalls and so helps to reduce latency.

Modifications:

- Ensure we register the timerfd and eventfd with EPOLLET flag
- If eventfd_write fails with EAGAIN, call eventfd_read and try eventfd_write again as we only use it as wake-up mechanism.

Result:

Less syscalls and so reducing overhead.

Co-authored-by: Carl Mastrangelo <carl@carlmastrangelo.com>

transport-native-epoll/src/main/c/netty_epoll_native.c

@@ -116,10 +116,27 @@ static jint netty_epoll_native_timerFd(JNIEnv* env, jclass clazz) {
 }
 
 static void netty_epoll_native_eventFdWrite(JNIEnv* env, jclass clazz, jint fd, jlong value) {
-    jint eventFD = eventfd_write(fd, (eventfd_t) value);
+    uint64_t val;
 
-    if (eventFD < 0) {
+    for (;;) {
-        netty_unix_errors_throwChannelExceptionErrorNo(env, "eventfd_write() failed: ", errno);
+        jint ret = eventfd_write(fd, (eventfd_t) value);
+
+        if (ret < 0) {
+            // We need to read before we can write again, let's try to read and then write again and if this
+            // fails we will bail out.
+            //
+            // See http://man7.org/linux/man-pages/man2/eventfd.2.html.
+            if (errno == EAGAIN) {
+                if (eventfd_read(fd, &val) == 0 || errno == EAGAIN) {
+                    // Try again
+                    continue;
+                }
+                netty_unix_errors_throwChannelExceptionErrorNo(env, "eventfd_read(...) failed: ", errno);
+            } else {
+                netty_unix_errors_throwChannelExceptionErrorNo(env, "eventfd_write(...) failed: ", errno);
+            }
+        }
+        break;
     }
 }
 

transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollEventLoop.java

@@ -98,12 +98,16 @@ class EpollEventLoop extends SingleThreadEventLoop {
             this.epollFd = epollFd = Native.newEpollCreate();
             this.eventFd = eventFd = Native.newEventFd();
             try {
-                Native.epollCtlAdd(epollFd.intValue(), eventFd.intValue(), Native.EPOLLIN);
+                // It is important to use EPOLLET here as we only want to get the notification once per
+                // wakeup and don't call eventfd_read(...).
+                Native.epollCtlAdd(epollFd.intValue(), eventFd.intValue(), Native.EPOLLIN | Native.EPOLLET);
             } catch (IOException e) {
                 throw new IllegalStateException("Unable to add eventFd filedescriptor to epoll", e);
             }
             this.timerFd = timerFd = Native.newTimerFd();
             try {
+                // It is important to use EPOLLET here as we only want to get the notification once per
+                // wakeup and don't call read(...).
                 Native.epollCtlAdd(epollFd.intValue(), timerFd.intValue(), Native.EPOLLIN | Native.EPOLLET);
             } catch (IOException e) {
                 throw new IllegalStateException("Unable to add timerFd filedescriptor to epoll", e);
@@ -402,12 +406,10 @@ class EpollEventLoop extends SingleThreadEventLoop {
     private void processReady(EpollEventArray events, int ready) {
         for (int i = 0; i < ready; i ++) {
             final int fd = events.fd(i);
-            if (fd == eventFd.intValue()) {
+            if (fd == eventFd.intValue() || fd == timerFd.intValue()) {
-                // consume wakeup event.
+                // Just ignore as we use ET mode for the eventfd and timerfd.
-                Native.eventFdRead(fd);
+                //
-            } else if (fd == timerFd.intValue()) {
+                // See also https://stackoverflow.com/a/12492308/1074097
-                // consume wakeup event, necessary because the timer is added with ET mode.
-                Native.timerFdRead(fd);
             } else {
                 final long ev = events.events(i);