failed servers are excluded from new shard creation

arangod/Agency/Job.h

@@ -45,8 +45,8 @@ static std::string const pendingPrefix  = "/Target/Pending/";
 static std::string const failedPrefix   = "/Target/Failed/";
 static std::string const finishedPrefix = "/Target/Finished/";
 static std::string const toDoPrefix     = "/Target/ToDo/";
-static std::string const cleanedPrefix  = "Target/CleanedServers";
-static std::string const failedServersPrefix  = "Target/FailedServers";
+static std::string const cleanedPrefix  = "/Target/CleanedServers";
+static std::string const failedServersPrefix  = "/Target/FailedServers";
 static std::string const planColPrefix  = "/Plan/Collections/";
 static std::string const curColPrefix   = "/Current/Collections/";
 static std::string const blockedServersPrefix = "/Supervision/DBServers/";

arangod/Agency/Supervision.cpp

@@ -612,7 +612,7 @@ bool Supervision::updateAgencyPrefix (size_t nTries, int intervalSec) {
   while (!this->isStopping()) {
     _snapshot = _agent->readDB().get("/");
     if (_snapshot.children().size() > 0) {
-      _agencyPrefix = std::string("/") + _snapshot.children().begin()->first;
+      _agencyPrefix = "/arango";//std::string("/") + _snapshot.children().begin()->first;
       LOG_TOPIC(DEBUG, Logger::AGENCY) << "Agency prefix is " << _agencyPrefix;
       return true;
     }

arangod/Cluster/ClusterInfo.cpp

@@ -2051,6 +2051,8 @@ void ClusterInfo::loadCurrentCoordinators() {
 ////////////////////////////////////////////////////////////////////////////////
 
 static std::string const prefixCurrentDBServers = "Current/DBServers";
+static std::string const prefixTargetCleaned = "Target/CleanedOutServers";
+static std::string const prefixTargetFailed = "Target/FailedServers";
 
 void ClusterInfo::loadCurrentDBServers() {
   uint64_t storedVersion = _DBServersProt.version;
@@ -2062,17 +2064,50 @@ void ClusterInfo::loadCurrentDBServers() {
 
   // Now contact the agency:
   AgencyCommResult result = _agency.getValues(prefixCurrentDBServers);
+  AgencyCommResult failed = _agency.getValues(prefixTargetFailed);
+  AgencyCommResult cleaned = _agency.getValues(prefixTargetCleaned);
 
   if (result.successful()) {
 
     velocypack::Slice currentDBServers =
       result.slice()[0].get(std::vector<std::string>(
             {AgencyComm::prefix(), "Current", "DBServers"}));
+    velocypack::Slice failedDBServers =
+      failed.slice()[0].get(std::vector<std::string>(
+            {AgencyComm::prefix(), "Target", "FailedServers"}));
+    velocypack::Slice cleanedDBServers =
+      cleaned.slice()[0].get(std::vector<std::string>(
+            {AgencyComm::prefix(), "Target", "CleanedOutServers"}));
 
     if (currentDBServers.isObject()) {
       decltype(_DBServers) newDBServers;
 
       for (auto const& dbserver : VPackObjectIterator(currentDBServers)) {
+        if (failedDBServers.isArray()) {
+          bool found = false;
+          for (auto const& failedServer : VPackArrayIterator(failedDBServers)) {
+            if (dbserver.key == failedServer) {
+              found = true;
+              continue;
+            }
+          }
+          if (found) {
+            continue;
+          }
+        }
+        if (cleanedDBServers.isArray()) {
+          bool found = false;
+          for (auto const& cleanedServer : VPackArrayIterator(cleanedDBServers)) {
+            if (dbserver.key == cleanedServer) {
+              found = true;
+              continue;
+            }
+          }
+          if (found) {
+            continue;
+          }
+        }
+
         newDBServers.emplace(
           std::make_pair(dbserver.key.copyString(), dbserver.value.copyString()));
       }

arangod/Cluster/ClusterMethods.cpp

@@ -1979,6 +1979,7 @@ std::map<std::string, std::vector<std::string>> distributeShards(
   std::map<std::string, std::vector<std::string>> shards;
 
   ClusterInfo*  ci = ClusterInfo::instance();
+  ci->loadCurrentDBServers();
   if (dbServers.size() == 0) {
     dbServers = ci->getCurrentDBServers();
     if (dbServers.empty()) {

scripts/startLocalCluster.sh

@@ -83,6 +83,7 @@ for aid in `seq 0 $(( $NRAGENTS - 1 ))`; do
         --server.threads 16 \
         --agency.compaction-step-size $COMP \
         --log.force-direct true \
+        --log.level agency=debug \
         > cluster/$port.stdout 2>&1 &
 done