make each ManagedRpcServer its own Task

4 files changed, 23 insertions, 17 deletions

diff --git a/slobrok/src/vespa/slobrok/server/managed_rpc_server.cpp b/slobrok/src/vespa/slobrok/server/managed_rpc_server.cpp
index 065dac04304..3220a9e5d69 100644
--- a/slobrok/src/vespa/slobrok/server/managed_rpc_server.cpp
+++ b/slobrok/src/vespa/slobrok/server/managed_rpc_server.cpp
@@ -2,6 +2,7 @@
 
 #include "managed_rpc_server.h"
 #include "i_rpc_server_manager.h"
+#include "random.h"
 #include <vespa/fnet/frt/supervisor.h>
 #include <vespa/fnet/frt/target.h>
 
@@ -16,17 +17,18 @@ ManagedRpcServer::ManagedRpcServer(const std::string & name,
                                    const std::string & spec,
                                    IRpcServerManager &manager)
     : NamedService(name, spec),
+      FNET_Task(manager.getSupervisor()->GetScheduler()),
       _mmanager(manager),
       _monitor(*this, *manager.getSupervisor()),
       _monitoredServer(nullptr),
       _checkServerReq(nullptr)
 {
+    double seconds = randomIn(0.2, 0.8);
+    LOG(debug, "first healthcheck for %s in %g seconds", name.c_str(), seconds);
+    Schedule(seconds);
 }
 
-
-void
-ManagedRpcServer::healthCheck()
-{
+void ManagedRpcServer::PerformTask() {
     if (_monitoredServer == nullptr) {
         _monitoredServer = _mmanager.getSupervisor()->GetTarget(_spec.c_str());
     }
@@ -35,12 +37,19 @@ ManagedRpcServer::healthCheck()
         _checkServerReq->SetMethodName("slobrok.callback.listNamesServed");
         _monitoredServer->InvokeAsync(_checkServerReq, 25.0, this);
     }
+    double seconds = randomIn(2.5, 2.9);
+    LOG(debug, "next healthcheck for %s in %g seconds", getName().c_str(), seconds);
+    Schedule(seconds);
 }
 
+void ManagedRpcServer::healthCheck() {
+    ScheduleNow();
+}
 
 ManagedRpcServer::~ManagedRpcServer()
 {
     LOG(debug, "(role[%s].~ManagedRpcServer)", _name.c_str());
+    Kill();
     cleanupMonitor();
 }
 
diff --git a/slobrok/src/vespa/slobrok/server/managed_rpc_server.h b/slobrok/src/vespa/slobrok/server/managed_rpc_server.h
index 072aebd7850..13bf2ba8686 100644
--- a/slobrok/src/vespa/slobrok/server/managed_rpc_server.h
+++ b/slobrok/src/vespa/slobrok/server/managed_rpc_server.h
@@ -23,6 +23,7 @@ class IRpcServerManager;
 
 class ManagedRpcServer: public NamedService,
                         public FRT_IRequestWait,
+                        public FNET_Task,
                         public IMonitoredServer
 {
 public:
@@ -32,6 +33,7 @@ public:
     ~ManagedRpcServer();
 
     void healthCheck();
+    void PerformTask() override;
 
 private:
     IRpcServerManager      &_mmanager;
diff --git a/slobrok/src/vespa/slobrok/server/selfcheck.cpp b/slobrok/src/vespa/slobrok/server/selfcheck.cpp
index 075db42e0d6..a9fab650772 100644
--- a/slobrok/src/vespa/slobrok/server/selfcheck.cpp
+++ b/slobrok/src/vespa/slobrok/server/selfcheck.cpp
@@ -18,8 +18,7 @@ SelfCheck::SelfCheck(FNET_Scheduler *sched,
                      RpcServerMap& rpcsrvmap,
                      RpcServerManager& rpcsrvman)
     : FNET_Task(sched),
-      _rpcsrvmap(rpcsrvmap), _rpcsrvmanager(rpcsrvman),
-      _checkIndex(0)
+      _rpcsrvmap(rpcsrvmap), _rpcsrvmanager(rpcsrvman)
 {
     // start within 1 second
     double seconds = randomIn(0.123, 1.000);
@@ -38,18 +37,16 @@ void
 SelfCheck::PerformTask()
 {
     std::vector<const NamedService *> mrpcsrvlist = _rpcsrvmap.allManaged();
-    if (_checkIndex < mrpcsrvlist.size()) {
-        const NamedService *r = mrpcsrvlist[_checkIndex++];
+
+    for (size_t i = 0; i < mrpcsrvlist.size(); ++i) {
+        const NamedService *r = mrpcsrvlist[i];
         ManagedRpcServer *m = _rpcsrvmap.lookupManaged(r->getName());
-        LOG(debug, "managed: %s -> %s", m->getName().c_str(), m->getSpec().c_str());
         LOG_ASSERT(r == m);
-        m->healthCheck();
-    } else {
-        _checkIndex = 0;
+        LOG(debug, "managed: %s -> %s", m->getName().c_str(), m->getSpec().c_str());
     }
-    // reschedule more often with more services, on average 1s per loop:
-    double seconds = randomIn(0.5, 1.5) / (1 + mrpcsrvlist.size());
-    LOG(debug, "next selfcheck in %g seconds", seconds);
+    // reschedule in 1-2 seconds:
+    double seconds = randomIn(0.987, 2.000);
+    LOG(debug, "selfcheck AGAIN in %g seconds", seconds);
     Schedule(seconds);
 }
 
diff --git a/slobrok/src/vespa/slobrok/server/selfcheck.h b/slobrok/src/vespa/slobrok/server/selfcheck.h
index 9bb96808666..52eced4700b 100644
--- a/slobrok/src/vespa/slobrok/server/selfcheck.h
+++ b/slobrok/src/vespa/slobrok/server/selfcheck.h
@@ -2,7 +2,6 @@
 #pragma once
 
 #include <vespa/fnet/task.h>
-#include <vector>
 
 namespace slobrok {
 
@@ -23,7 +22,6 @@ class SelfCheck : public FNET_Task
 private:
     RpcServerMap &_rpcsrvmap;
     RpcServerManager &_rpcsrvmanager;
-    size_t _checkIndex;
 
     SelfCheck(const SelfCheck &);            // Not used
     SelfCheck &operator=(const SelfCheck &); // Not used