Introduce watchdog into ProcessGroupManager [3/N]: Move ownership to ProcessGroupManager

score/launch_manager/src/daemon/BUILD

@@ -32,10 +32,12 @@ cc_binary(
         "//score/launch_manager/src/daemon/src/process_group_manager:alive_monitor_thread",
         "//score/launch_manager/src/daemon/src/process_state_client:process_state_notifier",
         "//score/launch_manager/src/daemon/src/recovery_client",
+        "//score/launch_manager/src/daemon/src/watchdog:i_watchdog_if",
         "@score_baselibs//score/language/futurecpp",
     ] + select({
         "//config:lm_use_new_configuration": [
             "//score/launch_manager/src/daemon/src/configuration:flatbuffer_config_loader",
+            "//score/launch_manager/src/daemon/src/watchdog/details:watchdog_impl",
         ],
         "//conditions:default": [],
     }),

score/launch_manager/src/daemon/src/alive_monitor/details/daemon/AliveMonitorImpl.cpp

@@ -18,7 +18,6 @@
 #include <score/assert.hpp>
 
 #include "score/mw/launch_manager/alive_monitor/details/daemon/AliveMonitorImpl.hpp"
-#include "score/mw/launch_manager/watchdog/details/WatchdogImpl.hpp"
 
 namespace score
 {
@@ -31,21 +30,17 @@ namespace daemon
 
 #ifdef USE_NEW_CONFIGURATION
 AliveMonitorImpl::AliveMonitorImpl(SptrIRecoveryClient recovery_client,
-                                   UptrIWatchdogIf watchdog,
                                    UptrIProcessStateReceiver process_state_receiver,
                                    const Config& config)
     : m_recovery_client(recovery_client),
-      m_watchdog(std::move(watchdog)),
       m_process_state_receiver{std::move(process_state_receiver)},
       m_config(config)
 {
 }
 #else
 AliveMonitorImpl::AliveMonitorImpl(SptrIRecoveryClient recovery_client,
-                                   UptrIWatchdogIf watchdog,
                                    UptrIProcessStateReceiver process_state_receiver)
     : m_recovery_client(recovery_client),
-      m_watchdog(std::move(watchdog)),
       m_process_state_receiver{std::move(process_state_receiver)}
 {
 }
@@ -58,8 +53,7 @@ EInitCode AliveMonitorImpl::init() noexcept
     {
         m_osClock.startMeasurement();
 
-        m_daemon = std::make_unique<PhmDaemon>(m_osClock, std::move(m_watchdog),
-                                               std::move(m_process_state_receiver));
+        m_daemon = std::make_unique<PhmDaemon>(m_osClock, std::move(m_process_state_receiver));
     #ifdef USE_NEW_CONFIGURATION
         initResult = m_daemon->init(m_recovery_client, m_config);
     #else

score/launch_manager/src/daemon/src/alive_monitor/details/daemon/AliveMonitorImpl.hpp

@@ -26,16 +26,11 @@ namespace lcm {
 
 class IRecoveryClient;
 
-namespace watchdog {
-class IWatchdogIf;
-}
-
 namespace saf {
 
 namespace daemon {
 
 using SptrIRecoveryClient = std::shared_ptr<score::lcm::IRecoveryClient>;
-using UptrIWatchdogIf = std::unique_ptr<watchdog::IWatchdogIf>;
 using UptrIProcessStateReceiver = std::unique_ptr<score::lcm::IProcessStateReceiver>;
 using UptrPhmDaemon = std::unique_ptr<score::lcm::saf::daemon::PhmDaemon>;
 using OsClock = score::lcm::saf::timers::OsClockInterface;
@@ -47,12 +42,10 @@ class AliveMonitorImpl : public IAliveMonitor {
    public:
 #ifdef USE_NEW_CONFIGURATION
     AliveMonitorImpl(SptrIRecoveryClient recovery_client,
-                     UptrIWatchdogIf watchdog,
                      UptrIProcessStateReceiver process_state_receiver,
                      const Config& config);
 #else
-    AliveMonitorImpl(SptrIRecoveryClient recovery_client, 
-                     UptrIWatchdogIf watchdog, 
+    AliveMonitorImpl(SptrIRecoveryClient recovery_client,
                      UptrIProcessStateReceiver process_state_receiver);
 #endif
 
@@ -62,7 +55,6 @@ class AliveMonitorImpl : public IAliveMonitor {
 
    private:
     SptrIRecoveryClient m_recovery_client{nullptr};
-    UptrIWatchdogIf m_watchdog{nullptr};
     UptrPhmDaemon m_daemon{nullptr};
     OsClock m_osClock{};
     UptrIProcessStateReceiver m_process_state_receiver;

score/launch_manager/src/daemon/src/alive_monitor/details/daemon/BUILD

@@ -74,7 +74,6 @@ cc_library(
         "//score/launch_manager/src/daemon/src/alive_monitor/details/timers:cycle_timer",
         "//score/launch_manager/src/daemon/src/alive_monitor/details/timers:timers_os_clock",
         "//score/launch_manager/src/daemon/src/common:log",
-        "//score/launch_manager/src/daemon/src/watchdog:i_watchdog_if",
         "//score/launch_manager/src/lifecycle_client",
     ] + select({
         "//config:lm_use_new_configuration": [
@@ -106,7 +105,6 @@ cc_library(
     visibility = ["//score/launch_manager/src/daemon:__subpackages__"],
     deps = [
         ":i_health_monitor",
-        "//score/launch_manager/src/daemon/src/watchdog/details:watchdog_impl",
         "@score_baselibs//score/language/futurecpp",
     ],
 )

score/launch_manager/src/daemon/src/alive_monitor/details/daemon/PhmDaemon.cpp

@@ -31,22 +31,17 @@ namespace daemon
    true_no_defect) */
 /* RULECHECKER_comment(0, 4, check_incomplete_data_member_construction, "Default constructor is used for\
  processStateReader.", true_no_defect) */
-PhmDaemon::PhmDaemon(OsClock& f_osClock,
-                     std::unique_ptr<Watchdog> f_watchdog,
-                     std::unique_ptr<ProcessStateReceiver> f_process_state_receiver)
+PhmDaemon::PhmDaemon(OsClock& f_osClock, std::unique_ptr<ProcessStateReceiver> f_process_state_receiver)
     : osClock{f_osClock},
       cycleTimer{&osClock},
       swClusterHandlers{},
-      processStateReader{std::move(f_process_state_receiver)},
-      watchdog(std::move(f_watchdog))
+      processStateReader{std::move(f_process_state_receiver)}
 {
     static_cast<void>(f_osClock);
 }
 
 void PhmDaemon::performCyclicTriggers(void)
 {
-    bool isCriticalFailure{false};
-
     NanoSecondType syncTimestamp{timers::OsClock::getMonotonicSystemClock()};
     if (syncTimestamp == 0U)
     {
@@ -55,29 +50,13 @@ void PhmDaemon::performCyclicTriggers(void)
         syncTimestamp = UINT64_MAX;
     }
 
-    isCriticalFailure = (!processStateReader.distributeChanges(syncTimestamp));
-
-    if (!isCriticalFailure)
+    if (processStateReader.distributeChanges(syncTimestamp))
     {
         for (auto& phmHandler : swClusterHandlers)
         {
             phmHandler.performCyclicTriggers(syncTimestamp);
-            isCriticalFailure = isCriticalFailure || phmHandler.hasAnyRecoveryEnqueueFailed();
         }
     }
-
-    // watchdog is fired iff:
-    //  * isCriticalFailure is set (e.g. process state distribution error, recovery ring buffer full)
-    // else:
-    //  * watchdog is serviced
-    if (!isCriticalFailure)
-    {
-        watchdog->serviceWatchdog();
-    }
-    else
-    {
-        watchdog->fireWatchdogReaction();
-    }
 }
 
 #ifdef USE_NEW_CONFIGURATION

score/launch_manager/src/daemon/src/alive_monitor/details/daemon/PhmDaemon.hpp

@@ -25,7 +25,6 @@
 #include "score/mw/launch_manager/alive_monitor/details/ifexm/ProcessStateReader.hpp"
 #include "score/mw/launch_manager/alive_monitor/details/timers/CycleTimeValidator.hpp"
 #include "score/mw/launch_manager/alive_monitor/details/timers/CycleTimer.hpp"
-#include "score/mw/launch_manager/watchdog/IWatchdogIf.hpp"
 #ifdef USE_NEW_CONFIGURATION
 #include "score/mw/launch_manager/configuration/config.hpp"
 #endif
@@ -45,8 +44,6 @@ enum class EInitCode : std::int8_t
     kNotInitialized,                    ///< Init was not performed
     kCycleTimeInitFailed,              ///< Cyclic Timer initialization failed
     kConstructFlatCfgFactoryFailed,    ///< FlatCfgFactory failed loading SWCL configurations
-    kWatchdogInitFailed,               ///< Watchdog Initialization failed
-    kWatchdogEnableFailed,             ///< Enabling watchdog device failed
     kMachineConfigInitFailed,          ///< MachineConfigFactory failed loading the machine configuration
     kSignalHandlerRegistrationFailed,  ///< Failed to register signal handler for termination signals
     kGeneralError                      ///< General error
@@ -60,7 +57,6 @@ class PhmDaemon
 {
 public:
     using OsClock = score::lcm::saf::timers::OsClockInterface;
-    using Watchdog = watchdog::IWatchdogIf;
     using ProcessStateReceiver = score::lcm::IProcessStateReceiver;
     using RecoveryClient = score::lcm::IRecoveryClient;
     using MachineConfigFactory = factory::MachineConfigFactory;
@@ -77,12 +73,10 @@ class PhmDaemon
      as same as generated function", true_no_defect) */
     /// @brief Set the OS clock interface
     /// @param[in] f_osClock Access to the system clock (dependency injection possible in tests)
-    /// @param[in] f_watchdog watchdog implementation (dependency injection possible in tests)
     /// @param[in] f_process_state_receiver process state receiver implementation (dependency injection possible in tests)
     /* RULECHECKER_comment(3,1, check_expensive_to_copy_in_parameter, "Move only types cannot be passed by const ref",
        true_no_defect) */
     PhmDaemon(OsClock& f_osClock,
-              std::unique_ptr<Watchdog> f_watchdog,
               std::unique_ptr<ProcessStateReceiver> f_process_state_receiver);
 
     /* RULECHECKER_comment(0, 4, check_min_instructions, "Default destructor is not provided\
@@ -150,18 +144,6 @@ class PhmDaemon
             return EInitCode::kCycleTimeInitFailed;
         }
 
-        if (!watchdog->init(cycleTimeModified, machineConfig))
-        {
-            LM_LOG_ERROR() << "Phm Daemon: Initialization of watchdog failed!";
-            return EInitCode::kWatchdogInitFailed;
-        }
-
-        if (!watchdog->enable())
-        {
-            LM_LOG_ERROR() << "Phm Daemon: Enabling of watchdog failed!";
-            return EInitCode::kWatchdogEnableFailed;
-        }
-
         return EInitCode::kNoError;
     }
 
@@ -234,7 +216,6 @@ class PhmDaemon
         }
         LM_LOG_INFO() << "Phm Daemon: Received termination request - shutting down";
 
-        watchdog->disable();
         return true;
     }
 
@@ -267,9 +248,6 @@ class PhmDaemon
 
     /// @brief Process State Reader for PHM daemon
     ProcessStateReader processStateReader;
-
-    /// @brief Connection to watchdog devices
-    std::unique_ptr<Watchdog> watchdog;
 };
 
 }  // namespace daemon

score/launch_manager/src/daemon/src/alive_monitor/details/factory/BUILD

@@ -49,7 +49,7 @@ cc_library(
     }),
     include_prefix = "score/mw/launch_manager/alive_monitor/details/factory",
     strip_include_prefix = "/score/launch_manager/src/daemon/src/alive_monitor/details/factory",
-    visibility = ["//score/launch_manager/src/daemon/src/alive_monitor:__subpackages__"],
+    visibility = ["//score/launch_manager/src/daemon:__subpackages__"],
     deps = [
         ":static_config",
         "//score/launch_manager/src/daemon/src/alive_monitor/config",

score/launch_manager/src/daemon/src/common/constants.hpp

@@ -47,6 +47,9 @@ constexpr std::chrono::milliseconds kMaxRunningDelay{1000};  ///< report_running
 constexpr std::chrono::milliseconds kControlClientMaxIpcDelay{500};  ///< The maximum time to wait, when trying to communicate with LCM. When this time is exceeded kCommunicationError will be returned
 constexpr std::chrono::milliseconds kControlClientBgThreadSleepTime{100};
 
+constexpr std::int64_t kMainLoopCycleTimeMs{50};  ///< The period at which the main loop services the watchdog
+constexpr std::int64_t kMainLoopCycleTimeNs{kMainLoopCycleTimeMs * 1'000'000LL};
+
 enum class ControlClientLimits : uint16_t {
     kControlClientMaxInstances =
         256U,  ///< Maximum number of ControlClient instances that should be created by state manager. If state manager create more instances than kMaxInstances, those instances will always return kCommunicationError when used

score/launch_manager/src/daemon/src/configuration/config.hpp

@@ -13,11 +13,11 @@
 #ifndef CONFIG_HPP
 #define CONFIG_HPP
 
+#include <sys/types.h>
 #include <cstdint>
 #include <optional>
 #include <string>
 #include <string_view>
-#include <sys/types.h>
 #include <vector>
 
 namespace score::mw::launch_manager::configuration
@@ -291,4 +291,4 @@ class ConfigBuilder
 
 }  // namespace score::mw::launch_manager::configuration
 
-#endif // CONFIG_HPP
+#endif  // CONFIG_HPP

score/launch_manager/src/daemon/src/main.cpp

@@ -21,7 +21,7 @@
 #include "score/mw/launch_manager/common/log.hpp"
 
 #include "score/mw/launch_manager/alive_monitor/details/daemon/AliveMonitorImpl.hpp"
-#include "score/mw/launch_manager/watchdog/details/WatchdogImpl.hpp"
+#include "score/mw/launch_manager/watchdog/IWatchdogIf.hpp"
 #include "score/mw/launch_manager/process_group_manager/alive_monitor_thread.hpp"
 #include "score/mw/launch_manager/process_group_manager/process_group_manager.hpp"
 #include "score/mw/launch_manager/process_state_client/process_state_notifier.hpp"
@@ -205,21 +205,26 @@ int main([[maybe_unused]] int argc, [[maybe_unused]] const char* argv[])
 #endif
         LM_LOG_DEBUG() << "Launch Manager Started !!!!";
         std::shared_ptr<score::lcm::IRecoveryClient> recoveryClient{std::make_shared<score::lcm::RecoveryClient>()};
-        std::unique_ptr<score::lcm::watchdog::IWatchdogIf> watchdog{
-            std::make_unique<score::lcm::watchdog::WatchdogImpl>()};
         auto process_state_notifier = std::make_unique<score::lcm::internal::ProcessStateNotifier>();
         std::unique_ptr<score::lcm::saf::daemon::IAliveMonitor> healthMonitor{
             std::make_unique<score::lcm::saf::daemon::AliveMonitorImpl>(
-                recoveryClient, std::move(watchdog), process_state_notifier->constructReceiver()
+                recoveryClient, process_state_notifier->constructReceiver()
 #ifdef USE_NEW_CONFIGURATION
                 , *config_result
 #endif
                 )};
         std::unique_ptr<score::lcm::internal::IAliveMonitorThread> aliveMonitorThread{
             std::make_unique<score::lcm::internal::AliveMonitorThread>(std::move(healthMonitor))};
 
-        std::unique_ptr<ProcessGroupManager> process_group_manager = std::make_unique<ProcessGroupManager>(
-            std::move(aliveMonitorThread), recoveryClient, std::move(process_state_notifier));
+#ifdef USE_NEW_CONFIGURATION
+        auto watchdog = score::lcm::watchdog::createWatchdog();
+        auto process_group_manager = std::make_unique<ProcessGroupManager>(
+            std::move(aliveMonitorThread), recoveryClient, std::move(process_state_notifier), std::move(watchdog));
+#else
+        // Legacy configuration wires no watchdog; the watchdog usage in PGM is #ifdef-guarded off.
+        auto process_group_manager = std::make_unique<ProcessGroupManager>(
+            std::move(aliveMonitorThread), recoveryClient, std::move(process_state_notifier), nullptr);
+#endif
 
 #ifdef USE_NEW_CONFIGURATION
         if (process_group_manager->initialize(*config_result))

score/launch_manager/src/daemon/src/process_group_manager/BUILD

@@ -71,6 +71,7 @@ cc_library(
         "//score/launch_manager/src/daemon/src/process_group_manager/details:safe_process_map",
         "//score/launch_manager/src/daemon/src/process_state_client:iprocess_state_notifier",
         "//score/launch_manager/src/daemon/src/recovery_client",
+        "//score/launch_manager/src/daemon/src/watchdog:i_watchdog_if",
         "@score_baselibs//score/language/futurecpp",
     ] + select({
         "//config:lm_use_new_configuration": [