Use a timer to check for death of supervised node's parent

src/Net.cc

@@ -290,17 +290,6 @@ void net_run()
 	while ( iosource_mgr->Size() ||
 		(BifConst::exit_only_after_terminate && ! terminating) )
 		{
-		// Note: only simple + portable way of detecting loss of parent
-		// process seems to be polling for change in PPID.  There's platform
-		// specific ways if we do end up needing something more responsive
-		// and/or have to avoid overhead of polling, but maybe not worth
-		// the additional complexity:
-		//   Linux:   prctl(PR_SET_PDEATHSIG, ...)
-		//   FreeBSD: procctl(PROC_PDEATHSIG_CTL)
-		// TODO: make this a proper timer
-		if ( zeek::supervised_node && zeek::supervised_node->parent_pid != getppid() )
-			zeek_terminate_loop("supervised cluster node was orphaned");
-
 		double ts;
 		iosource::IOSource* src = iosource_mgr->FindSoonest(&ts);
 

src/Supervisor.cc

@@ -13,6 +13,7 @@
 #include "Reporter.h"
 #include "DebugLogger.h"
 #include "Val.h"
+#include "Net.h"
 #include "NetVar.h"
 #include "zeek-config.h"
 #include "util.h"
@@ -124,6 +125,29 @@ static std::string make_create_message(const Supervisor::NodeConfig& node)
 	return fmt("create %s %s", node.name.data(), json_str.data());
 	}
 
+ParentProcessCheckTimer::ParentProcessCheckTimer(double t, double arg_interval)
+	: Timer(t, TIMER_PPID_CHECK), interval(arg_interval)
+	{
+	}
+
+void ParentProcessCheckTimer::Dispatch(double t, int is_expire)
+	{
+	// Note: only simple + portable way of detecting loss of parent
+	// process seems to be polling for change in PPID.  There's platform
+	// specific ways if we do end up needing something more responsive
+	// and/or have to avoid overhead of polling, but maybe not worth
+	// the additional complexity:
+	//   Linux:   prctl(PR_SET_PDEATHSIG, ...)
+	//   FreeBSD: procctl(PROC_PDEATHSIG_CTL)
+	// Also note the Stem process has its own polling loop with similar logic.
+	if ( zeek::supervised_node->parent_pid != getppid() )
+		zeek_terminate_loop("supervised node was orphaned");
+
+	if ( ! is_expire )
+		timer_mgr->Add(new ParentProcessCheckTimer(network_time + interval,
+		                                           interval));
+	}
+
 Supervisor::Supervisor(Supervisor::Config cfg,
                        std::unique_ptr<bro::PipePair> pipe,
                        pid_t arg_stem_pid)
@@ -714,6 +738,7 @@ std::optional<Supervisor::SupervisedNode> Stem::Poll()
 		// the additional complexity:
 		//   Linux:   prctl(PR_SET_PDEATHSIG, ...)
 		//   FreeBSD: procctl(PROC_PDEATHSIG_CTL)
+		// Also note the similar polling methodology in ParentProcessCheckTimer.
 		DBG_STEM("Stem suicide");
 		Shutdown(13);
 		}

src/Supervisor.h

@@ -12,6 +12,7 @@
 #include <map>
 
 #include "iosource/IOSource.h"
+#include "Timer.h"
 #include "Pipe.h"
 #include "Flare.h"
 #include "NetVar.h"
@@ -19,6 +20,18 @@
 
 namespace zeek {
 
+class ParentProcessCheckTimer : public Timer {
+public:
+
+	ParentProcessCheckTimer(double t, double arg_interval);
+
+	void Dispatch(double t, int is_expire) override;
+
+protected:
+
+	double interval;
+};
+
 class Supervisor : public iosource::IOSource {
 public:
 

src/Timer.cc

@@ -37,6 +37,7 @@ const char* TimerNames[] = {
 	"TCPConnectionPartialClose",
 	"TCPConnectionResetTimer",
 	"TriggerTimer",
+	"ParentProcessIDCheck",
 	"TimerMgrExpireTimer",
 };
 

src/Timer.h

@@ -41,6 +41,7 @@ enum TimerType {
 	TIMER_TCP_PARTIAL_CLOSE,
 	TIMER_TCP_RESET,
 	TIMER_TRIGGER,
+	TIMER_PPID_CHECK,
 	TIMER_TIMERMGR_EXPIRE,
 };
 const int NUM_TIMER_TYPES = int(TIMER_TIMERMGR_EXPIRE) + 1;

src/main.cc

@@ -1481,6 +1481,9 @@ int main(int argc, char** argv)
 
 #endif
 
+		if ( zeek::supervised_node )
+			timer_mgr->Add(new zeek::ParentProcessCheckTimer(1, 1));
+
 		double time_net_start = current_time(true);;
 
 		uint64_t mem_net_start_total;