Systemd watchdog notification (#57)

* feat: feed systemd watchdog

* add libsystemd-dev as build dependency

* fix: put watchdog feed in the right place

* ci: install libsystemd-dev in ubuntu

* ci: don't run alpine build

* add watchdog parameter to systemd unit

* keep feeding watchdog during shutdown

* chore: update watchdog only when needed

* chore: remove unused CI definition completely

.github/workflows/build.yml

@@ -15,7 +15,7 @@ jobs:
       - name: Install Dependencies
         run: |
           apt update
-          apt install autoconf autogen libtool systemd -y
+          apt install autoconf autogen libtool systemd libsystemd-dev -y
       - name: Build
         run: |
           ./autogen.sh && ./configure CFLAGS='-g -O2' --sysconfdir=/etc --localstatedir=/var --libdir=/usr/lib64 --prefix=/usr
@@ -24,18 +24,3 @@ jobs:
         run: |
           make mainloop_test
           ./mainloop_test
-
-  alpine-linux:
-    name: alpine 3.18 (musl)
-    runs-on: ubuntu-22.04
-    container: alpine:3.18
-    steps:
-      - uses: actions/checkout@v1
-        with:
-          submodules: recursive
-      - name: install tools
-        run: apk update && apk add autoconf automake build-base git libtool linux-headers openssl-dev perl-extutils-pkgconfig python3 py3-future
-      - name: configure
-        run: ./autogen.sh && ./configure CFLAGS='-g -O2' --sysconfdir=/etc --localstatedir=/var --libdir=/usr/lib --disable-systemd --prefix=/usr
-      - name: build
-        run: make

README.md

@@ -34,6 +34,7 @@ The following libraries need also to be installed :
     $ sudo apt install libtool
     $ sudo apt install autoconf
     $ sudo apt install pkg-config
+    $ sudo apt install libsystemd-dev
 
 #### Build ####
 

configure.ac

@@ -63,6 +63,9 @@ AM_CONDITIONAL([HAVE_GTEST], [test x$HAVE_GTEST = xtrue])
 AC_SUBST([GTEST_LIBS])
 AC_SUBST([GTEST_CFLAGS])
 
+AC_CHECK_LIB([systemd], [main], [], [AC_MSG_ERROR([systemd library not found])])
+AC_SUBST([LIBS], ["-lsystemd"])
+
 #####################################################################
 # --with-
 #####################################################################

debian/control

@@ -13,6 +13,7 @@ Build-Depends:
  libgtest-dev,
  libc6,
  libstdc++6,
+ libsystemd-dev,
 Standards-Version: 4.1.3
 Homepage: https://github.com/Auterion/mavlink-router
 

mavlink-router.service.in

@@ -6,6 +6,7 @@ Type=simple
 ExecStart=@bindir@/mavlink-routerd
 Restart=always
 Nice=-20
+WatchdogSec=5s
 
 [Install]
 WantedBy=multi-user.target

src/mavlink-router/mainloop.cpp

@@ -17,13 +17,15 @@
  */
 #include "mainloop.h"
 
+#include <chrono>
 #include <assert.h>
 #include <fcntl.h>
 #include <signal.h>
 #include <sys/epoll.h>
 #include <sys/stat.h>
 #include <sys/timerfd.h>
 #include <unistd.h>
+#include <systemd/sd-daemon.h>
 
 #include <memory>
 #include <vector>
@@ -280,6 +282,10 @@ void Mainloop::handle_tcp_connection()
 
 int Mainloop::loop()
 {
+    sd_notify(0, "READY=1");
+    const int watchdog_interval_us = _watchdogIntervalUs();
+    auto last_watchdog_update = std::chrono::steady_clock::now();
+
     if (epollfd < 0)
         return EXIT_FAILURE;
 
@@ -290,6 +296,16 @@ int Mainloop::loop()
 
     while (!_should_exit.load(std::memory_order_relaxed)) {
         run_single(-1);
+
+        // Watchdog update
+        if (watchdog_interval_us > 0) {
+        const auto now = std::chrono::steady_clock::now();
+            if (std::chrono::duration_cast<std::chrono::microseconds>(now - last_watchdog_update)
+                    .count() > watchdog_interval_us) {
+                last_watchdog_update = now;
+                sd_notify(0, "WATCHDOG=1");
+            }
+        }
     }
 
     // This is a bit weird, but models previous behavior: run event handling a
@@ -298,12 +314,22 @@ int Mainloop::loop()
     if (_log_endpoint) {
         _log_endpoint->stop();
 
-        usec_t now = now_usec();
-        usec_t deadline = now + TIMEOUT_LOG_SHUTDOWN_US;
+        auto now = std::chrono::steady_clock::now();
+        const auto deadline = now + std::chrono::microseconds(TIMEOUT_LOG_SHUTDOWN_US);
 
         while (now < deadline) {
-            run_single((deadline - now) / 1000);
-            now = now_usec();
+            run_single(std::chrono::duration_cast<std::chrono::milliseconds>(deadline - now).count());
+
+            // Watchdog update
+            if (watchdog_interval_us > 0) {
+                if (std::chrono::duration_cast<std::chrono::microseconds>(now - last_watchdog_update)
+                        .count() > watchdog_interval_us) {
+                    last_watchdog_update = now;
+                    sd_notify(0, "WATCHDOG=1");
+                }
+            }
+
+            now = std::chrono::steady_clock::now();
         }
 
         _log_endpoint->stop();
@@ -934,6 +960,15 @@ void Mainloop::_handle_pipe()
     }
 }
 
+int Mainloop::_watchdogIntervalUs()
+{
+  const char* watchdog_usec_env = getenv("WATCHDOG_USEC");
+  if (watchdog_usec_env) {
+    return atoi(watchdog_usec_env) / 2;
+  }
+  return 0;
+}
+
 MainloopSignalHandlers::MainloopSignalHandlers(Mainloop* mainloop)
 {
     if (mainloop_instance) {

src/mavlink-router/mainloop.h

@@ -151,6 +151,7 @@ class Mainloop {
     bool _log_aggregate_timeout(void *data);
     void _init_pipe();
     void _handle_pipe();
+    static int _watchdogIntervalUs();
 
     static Mainloop* instance;
 };