Changed the locking strategy in agent creation.

The code in this part was uselessly taking exclusive root locks, leading to locking
fights when starting all drive as the same time in bigger systems.

objectstore/BackendPopulator.cpp

@@ -29,18 +29,29 @@ BackendPopulator::BackendPopulator(cta::objectstore::Backend & be,
     const std::string &agentType, const cta::log::LogContext & lc): m_backend(be), m_agentReference(agentType, lc.logger()),
     m_lc(lc) {
   cta::objectstore::RootEntry re(m_backend);
-  cta::objectstore::ScopedExclusiveLock rel(re);
-  re.fetch();
+  re.fetchNoLock();
+  cta::objectstore::EntryLogSerDeser cl("user0", "systemhost", time(NULL));
+  // We might have to create the agent register (but this is unlikely)
+  try {
+    re.getAgentRegisterAddress();
+  } catch (...) {
+    RootEntry re2(m_backend);
+    ScopedExclusiveLock rel(re2);
+    re2.fetch();
+    re2.addOrGetAgentRegisterPointerAndCommit(m_agentReference, cl);
+  }
   Agent agent(m_agentReference.getAgentAddress(), m_backend);
   agent.initialize();
-  cta::objectstore::EntryLogSerDeser cl("user0", "systemhost", time(NULL));
-  re.addOrGetAgentRegisterPointerAndCommit(m_agentReference, cl);
-  rel.release();
   agent.insertAndRegisterSelf();
-  rel.lock(re);
-  re.fetch();
-  re.addOrGetDriveRegisterPointerAndCommit(m_agentReference, cl);
-  rel.release();
+  // Likewise, make sure the drive register is around.
+  try {
+    re.getDriveRegisterAddress();
+  } catch (...) {
+    RootEntry re2(m_backend);
+    ScopedExclusiveLock rel(re2);
+    re2.fetch();
+    re2.addOrGetDriveRegisterPointerAndCommit(m_agentReference, cl);
+  }
 }
 
 //------------------------------------------------------------------------------