server health for aardvark

arangod/Agency/CleanOutServer.cpp

@@ -129,17 +129,21 @@ bool CleanOutServer::start() const {
   if (res.accepted && res.indices.size()==1 && res.indices[0]) {
     
     LOG_TOPIC(INFO, Logger::AGENCY) << "Pending: Clean out server " + _server;
+  LOG(WARN) << __FILE__<<__LINE__ ;
 
     // Check if we can get things done in the first place
     if (!checkFeasibility()) {
-      finish("DBServers/" + _server);
-          return false;
+      finish("DBServers/" + _server, false);
+      LOG(WARN) << __FILE__<<__LINE__ ;
+      return false;
     }
 
-    
+      LOG(WARN) << __FILE__<<__LINE__ ;
+
 
     // Schedule shard relocations
     scheduleMoveShards();
+  LOG(WARN) << __FILE__<<__LINE__ ;
     
     return true;
     
@@ -244,9 +248,9 @@ bool CleanOutServer::checkFeasibility () const {
     }
     
     LOG_TOPIC(ERR, Logger::AGENCY)
-      << "Cannot accomodate all shards " << collections.str()
-      << " with replication factors " << factors.str()
-      << " after cleaning out server " << _server;
+      << "Cannot accomodate shards " << collections.str()
+      << "with replication factors " << factors.str()
+      << "after cleaning out server " << _server;
     return false;
   }
 

arangod/Agency/Supervision.cpp

@@ -206,22 +206,23 @@ std::vector<check_t> Supervision::checkCoordinators() {
   return ret;
 }
 
-
-bool Supervision::doChecks(bool timedout) {
+bool Supervision::updateSnapshot() {
 
   if (_agent == nullptr || this->isStopping()) {
     return false;
   }
-
   _snapshot = _agent->readDB().get(_agencyPrefix);
-
-  LOG_TOPIC(DEBUG, Logger::AGENCY) << "Sanity checks";
-  /*std::vector<check_t> ret = */checkDBServers();
-  checkCoordinators();
-
   return true;
 }
 
+bool Supervision::doChecks(bool timedout) {
+
+  checkDBServers();
+  checkCoordinators();
+  return true;
+  
+}
+
 void Supervision::run() {
 
   CONDITION_LOCKER(guard, _cv);
@@ -256,6 +257,7 @@ void Supervision::run() {
     }
     
     // Do supervision
+    updateSnapshot();
     doChecks(timedout);
     workJobs();
     
@@ -267,36 +269,45 @@ void Supervision::workJobs() {
 
   Node::Children const& todos = _snapshot(toDoPrefix).children();
   Node::Children const& pends = _snapshot(pendingPrefix).children();
+
   if (!todos.empty()) {
     for (auto const& todoEnt : todos) {
       Node const& job = *todoEnt.second;
-      LOG(WARN) << __FILE__<<__LINE__ << job.toJson();
       
-      std::string jobType = job("type").getString(),
-        jobId = job("jobId").getString(),
-        creator = job("creator").getString();
-      if (jobType == "failedServer") {
-        FailedServer fs(_snapshot, _agent, jobId, creator, _agencyPrefix);
-      } else if (jobType == "cleanOutServer") {
-        CleanOutServer cos(_snapshot, _agent, jobId, creator, _agencyPrefix);
+      try {
+        std::string jobType = job("type").getString(),
+          jobId = job("jobId").getString(),
+          creator = job("creator").getString();
+        if (jobType == "failedServer") {
+          FailedServer fs(_snapshot, _agent, jobId, creator, _agencyPrefix);
+        } else if (jobType == "cleanOutServer") {
+          CleanOutServer cos(_snapshot, _agent, jobId, creator, _agencyPrefix);
+        }
+      } catch (std::exception const& e) {
+        LOG_TOPIC(ERR, Logger::AGENCY) << e.what() << " " << __FILE__ << __LINE__;
       }
     }
   }
+
   if (!pends.empty()) {
     for (auto const& pendEnt : pends) {
       Node const& job = *pendEnt.second;
-      LOG(WARN) << __FILE__<<__LINE__ << job.toJson();
 
-      std::string jobType = job("type").getString(),
-        jobId = job("jobId").getString(),
-        creator = job("creator").getString();
-      if (jobType == "failedServer") {
-        FailedServer fs(_snapshot, _agent, jobId, creator, _agencyPrefix);
-      } else if (jobType == "cleanOutServer") {
-        CleanOutServer cos(_snapshot, _agent, jobId, creator, _agencyPrefix);
+      try {
+        std::string jobType = job("type").getString(),
+          jobId = job("jobId").getString(),
+          creator = job("creator").getString();
+        if (jobType == "failedServer") {
+          FailedServer fs(_snapshot, _agent, jobId, creator, _agencyPrefix);
+        } else if (jobType == "cleanOutServer") {
+          CleanOutServer cos(_snapshot, _agent, jobId, creator, _agencyPrefix);
+        }
+      } catch (std::exception const& e) {
+        LOG_TOPIC(ERR, Logger::AGENCY) << e.what() << " " << __FILE__ << __LINE__;
       }
     }
   }
+  
 }
 
 // Start thread

arangod/Agency/Supervision.h

@@ -141,6 +141,9 @@ class Supervision : public arangodb::Thread {
   /// @brief Perform sanity checking
   bool doChecks(bool);
 
+  /// @brief update my local agency snapshot
+  bool updateSnapshot();
+
   Agent* _agent; /**< @brief My agent */
   Node _snapshot;
 

js/actions/api-cluster.js

@@ -196,7 +196,7 @@ actions.defineHttp({
     var DBserver = req.parameters.DBserver;
     var coord = { coordTransactionID: ArangoClusterInfo.uniqid() };
     var options = { coordTransactionID: coord.coordTransactionID, timeout:10 };
-    var op = ArangoClusterComm.asyncRequest("GET","server:"+local,"_system",
+    var op = ArangoClusterComm.asyncRequest("GET","server:"+DBserver,"_system",
                                             "/_admin/statistics","",{},options);
     var r = ArangoClusterComm.wait(op);
     res.contentType = "application/json; charset=utf-8";