B #6772: Fix for NUMA and CPU Pinning Discrepancies During VM Save and Live Migration

Signed-off-by: Kristian Feldsam <feldsam@gmail.com>

Author: feldsam

include/History.h

@@ -114,11 +114,13 @@ class History:public ObjectSQL, public ObjectXML
     std::string  deployment_file;
     std::string  context_file;
     std::string  token_file;
+    std::string  migrate_file;
 
     // Remote paths
     std::string  checkpoint_file;
     std::string  rdeployment_file;
     std::string  system_dir;
+    std::string  rmigrate_file;
 
     /**
      *  Writes the history record in the DB

include/VirtualMachine.h

@@ -585,6 +585,19 @@ class VirtualMachine : public PoolObjectSQL
         return history->token_file;
     }
 
+    /**
+     *  Returns the migrate filename. The migrate file is in the form:
+     *          $ONE_LOCATION/var/vms/$VM_ID/migrate.$SEQ
+     *  or, in case that OpenNebula is installed in root
+     *          /var/lib/one/vms/$VM_ID/migrate.$SEQ
+     *  The hasHistory() function MUST be called before this one.
+     *    @return the migrate file path
+     */
+    const std::string & get_migrate_file() const
+    {
+        return history->migrate_file;
+    };
+
     /**
      *  Returns the remote deployment filename. The file is in the form:
      *          $DS_LOCATION/$SYSTEM_DS/$VM_ID/deployment.$SEQ
@@ -596,6 +609,17 @@ class VirtualMachine : public PoolObjectSQL
         return history->rdeployment_file;
     };
 
+    /**
+     *  Returns the remote migrate filename. The file is in the form:
+     *          $DS_LOCATION/$SYSTEM_DS/$VM_ID/migrate.$SEQ
+     *  The hasHistory() function MUST be called before this one.
+     *    @return the migrate filename
+     */
+    const std::string & get_rmigrate_file() const
+    {
+        return history->rmigrate_file;
+    };
+
     /**
      *  Returns the checkpoint filename for the current host. The checkpoint file
      *  is in the form:

include/VirtualMachineManager.h

@@ -376,7 +376,9 @@ class VirtualMachineManager :
             const std::string& tmpl,
             int ds_id,
             int sgid = -1,
-            int nicid = -1);
+            int nicid = -1,
+            const std::string& lmfile = "",
+            const std::string& rmfile = "");
 
 public:
     /**

src/lcm/LifeCycleActions.cc

@@ -350,7 +350,11 @@ void LifeCycleManager::trigger_migrate(int vid, const RequestAttributes& ra,
 
             if ( vm->get_hid() != vm->get_previous_hid() )
             {
-                hpool->del_capacity(vm->get_previous_hid(), sr);
+                HostShareCapacity prev_sr;
+                Template tmpl;
+                vm->get_previous_capacity(prev_sr, tmpl);
+
+                hpool->del_capacity(vm->get_previous_hid(), prev_sr);
 
                 vm->release_previous_vnc_port();
             }
@@ -1038,6 +1042,8 @@ void LifeCycleManager::clean_up_vm(VirtualMachine * vm, bool dispose,
                                    int& image_id, int uid, int gid, int req_id, Template& quota_tmpl)
 {
     HostShareCapacity sr;
+    HostShareCapacity prev_sr;
+    Template tmpl;
 
     time_t the_time = time(0);
 
@@ -1245,11 +1251,13 @@ void LifeCycleManager::clean_up_vm(VirtualMachine * vm, bool dispose,
             case VirtualMachine::MIGRATE:
                 vm->set_running_etime(the_time);
 
+                vm->get_previous_capacity(prev_sr, tmpl);
+
                 vm->set_previous_etime(the_time);
                 vm->set_previous_vm_info();
                 vm->set_previous_running_etime(the_time);
 
-                hpool->del_capacity(vm->get_previous_hid(), sr);
+                hpool->del_capacity(vm->get_previous_hid(), prev_sr);
 
                 vmpool->update_previous_history(vm);
 
@@ -1268,11 +1276,13 @@ void LifeCycleManager::clean_up_vm(VirtualMachine * vm, bool dispose,
             case VirtualMachine::SAVE_MIGRATE:
                 vm->set_running_etime(the_time);
 
+                vm->get_previous_capacity(prev_sr, tmpl);
+
                 vm->set_previous_etime(the_time);
                 vm->set_previous_vm_info();
                 vm->set_previous_running_etime(the_time);
 
-                hpool->del_capacity(vm->get_previous_hid(), sr);
+                hpool->del_capacity(vm->get_previous_hid(), prev_sr);
 
                 vmpool->update_previous_history(vm);
 

src/lcm/LifeCycleStates.cc

@@ -44,16 +44,6 @@ void LifeCycleManager::start_prolog_migrate(VirtualMachine* vm)
 
     vm->set_state(VirtualMachine::PROLOG_MIGRATE);
 
-    vm->set_previous_etime(the_time);
-
-    vm->set_previous_running_etime(the_time);
-
-    vmpool->update_previous_history(vm);
-
-    vm->set_prolog_stime(the_time);
-
-    vmpool->update_history(vm);
-
     if ( vm->get_hid() != vm->get_previous_hid() )
     {
         Template tmpl;
@@ -64,6 +54,16 @@ void LifeCycleManager::start_prolog_migrate(VirtualMachine* vm)
         vm->release_previous_vnc_port();
     }
 
+    vm->set_previous_etime(the_time);
+
+    vm->set_previous_running_etime(the_time);
+
+    vmpool->update_previous_history(vm);
+
+    vm->set_prolog_stime(the_time);
+
+    vmpool->update_history(vm);
+
     vmpool->update(vm);
 
     //----------------------------------------------------
@@ -290,6 +290,9 @@ void LifeCycleManager::trigger_deploy_success(int vid)
 
             vm->set_running_stime(the_time);
 
+            Template tmpl;
+            vm->get_previous_capacity(sr, tmpl);
+
             vmpool->update_history(vm.get());
 
             vm->set_previous_etime(the_time);
@@ -298,8 +301,6 @@ void LifeCycleManager::trigger_deploy_success(int vid)
 
             vmpool->update_previous_history(vm.get());
 
-            vm->get_capacity(sr);
-
             hpool->del_capacity(vm->get_previous_hid(), sr);
 
             vm->set_state(VirtualMachine::RUNNING);

src/rm/RequestManagerVirtualMachine.cc

@@ -1160,14 +1160,6 @@ void VirtualMachineMigrate::request_execute(xmlrpc_c::paramList const& paramList
         return;
     }
 
-    if (live && vm->is_pinned())
-    {
-        att.resp_msg = "VM with a pinned NUMA topology cannot be live-migrated";
-        failure_response(ACTION, att);
-
-        return;
-    }
-
     // Get System DS information from current History record
     c_ds_id  = vm->get_ds_id();
     c_tm_mad = vm->get_tm_mad();

src/vm/History.cc

@@ -128,6 +128,11 @@ void History::non_persistent_data()
 
     token_file = os.str();
 
+    os.str("");
+    os << vm_lhome << "/migrate." << seq;
+
+    migrate_file = os.str();
+
     // ----------- Remote Locations ------------
     os.str("");
     os << ds_location << "/" << ds_id << "/" << oid;
@@ -141,6 +146,11 @@ void History::non_persistent_data()
     os << system_dir << "/deployment." << seq;
 
     rdeployment_file = os.str();
+
+    os.str("");
+    os << system_dir << "/migrate." << seq;
+
+    rmigrate_file = os.str();
 }
 
 /* -------------------------------------------------------------------------- */

src/vmm/LibVirtDriverKVM.cc

@@ -2298,5 +2298,7 @@ int LibVirtDriver::deployment_description_kvm(
 
     file << "</domain>" << endl;
 
+    file.close();
+
     return 0;
 }

src/vmm/VirtualMachineManager.cc

@@ -221,7 +221,9 @@ string VirtualMachineManager::format_message(
         const string& tmpl,
         int ds_id,
         int sgid,
-        int nicid)
+        int nicid,
+        const string& lmfile,
+        const string& rmfile)
 {
     ostringstream oss;
 
@@ -266,6 +268,17 @@ string VirtualMachineManager::format_message(
         oss << "<REMOTE_DEPLOYMENT_FILE/>";
     }
 
+    if (!lmfile.empty())
+    {
+        oss << "<LOCAL_MIGRATE_FILE>" << lmfile << "</LOCAL_MIGRATE_FILE>";
+        oss << "<REMOTE_MIGRATE_FILE>" << rmfile << "</REMOTE_MIGRATE_FILE>";
+    }
+    else
+    {
+        oss << "<LOCAL_MIGRATE_FILE/>";
+        oss << "<REMOTE_MIGRATE_FILE/>";
+    }
+
     if (!cfile.empty())
     {
         oss << "<CHECKPOINT_FILE>" << cfile << "</CHECKPOINT_FILE>";
@@ -495,8 +508,9 @@ void VirtualMachineManager::trigger_save(int vid)
     trigger([this, vid]
     {
         const VirtualMachineManagerDriver * vmd;
+        int rc;
 
-        string   hostname, checkpoint_file;
+        string   hostname, checkpoint_file, migrate_file, rmigrate_file;
         string   vm_tmpl;
         string   drv_msg;
         int      ds_id;
@@ -535,12 +549,31 @@ void VirtualMachineManager::trigger_save(int vid)
             hostname        = vm->get_previous_hostname();
             checkpoint_file = vm->get_previous_checkpoint_file();
             ds_id           = vm->get_previous_ds_id();
+
+            //Generate VM description file
+            os << "Generating migrate file: " << vm->get_migrate_file();
+
+            vm->log("VMM", Log::INFO, os);
+
+            os.str("");
+
+            rc = vmd->deployment_description(vm.get(), vm->get_migrate_file());
+
+            if (rc != 0)
+            {
+                goto error_file;
+            }
+
+            migrate_file = vm->get_migrate_file();
+            rmigrate_file = vm->get_rmigrate_file();
         }
         else
         {
             hostname        = vm->get_hostname();
             checkpoint_file = vm->get_checkpoint_file();
             ds_id           = vm->get_ds_id();
+            migrate_file    = "";
+            rmigrate_file   = "";
         }
 
         // Invoke driver method
@@ -556,7 +589,10 @@ void VirtualMachineManager::trigger_save(int vid)
                           "",
                           vm->to_xml(vm_tmpl),
                           ds_id,
-                          -1);
+                         -1,
+                         -1,
+                         migrate_file,
+                         rmigrate_file);
 
         vmd->save(vid, drv_msg);
 
@@ -570,6 +606,11 @@ void VirtualMachineManager::trigger_save(int vid)
         os << "save_action, error getting driver " << vm->get_vmm_mad();
         goto error_common;
 
+error_file:
+        os << "save_action, error generating migrate file: "
+        << vm->get_migrate_file();
+        goto error_common;
+
 error_previous_history:
         os << "save_action, VM has no previous history";
 
@@ -1154,10 +1195,12 @@ void VirtualMachineManager::trigger_migrate(int vid)
     trigger([this, vid]
     {
         const VirtualMachineManagerDriver * vmd;
+        int rc;
 
         ostringstream os;
         string   vm_tmpl;
         string   drv_msg;
+        string   tm_command = "";
 
         // Get the VM from the pool
         auto vm = vmpool->get(vid);
@@ -1187,6 +1230,24 @@ void VirtualMachineManager::trigger_migrate(int vid)
 
         Nebula::instance().get_tm()->migrate_transfer_command(vm.get(), os);
 
+        tm_command = os.str();
+
+        os.str("");
+
+        //Generate VM description file
+        os << "Generating migrate file: " << vm->get_migrate_file();
+
+        vm->log("VMM", Log::INFO, os);
+
+        os.str("");
+
+        rc = vmd->deployment_description(vm.get(), vm->get_migrate_file());
+
+        if (rc != 0)
+        {
+            goto error_file;
+        }
+
         // Invoke driver method
         drv_msg = format_message(
                 vm->get_previous_hostname(),
@@ -1195,12 +1256,15 @@ void VirtualMachineManager::trigger_migrate(int vid)
                 "",
                 "",
                 "",
-                os.str(),
+                tm_command,
                 "",
                 vm->get_system_dir(),
                 vm->to_xml(vm_tmpl),
                 vm->get_previous_ds_id(),
-                -1);
+                -1,
+                -1,
+                vm->get_migrate_file(),
+                vm->get_rmigrate_file());
 
         vmd->migrate(vid, drv_msg);
 
@@ -1214,6 +1278,11 @@ void VirtualMachineManager::trigger_migrate(int vid)
         os << "migrate_action, error getting driver " << vm->get_vmm_mad();
         goto error_common;
 
+error_file:
+        os << "migrate_action, error generating migrate file: "
+        << vm->get_migrate_file();
+        goto error_common;
+
 error_previous_history:
         os << "migrate_action, error VM has no previous history";