EH: CS-315 make sure that the accounting and reporting code is thread safe

Author: jgabler-hpc

source/daemons/qmaster/ocs_JsonAccountingFileWriter.cc

@@ -46,15 +46,19 @@ namespace ocs {
          // get accounting data
          rapidjson::StringBuffer json_buffer;
          rapidjson::Writer<rapidjson::StringBuffer> writer(json_buffer);
-         ret = sge_write_rusage(nullptr, &writer, job_report, job, ja_task, category_string,
-                                &usage_pattern_list, 0, false, false);
+
+         // we need to protect the usage_pattern_list via the config_mutex
+         sge_mutex_lock(config_mutex_name.c_str(), __func__, __LINE__, &config_mutex);
+         ret = sge_write_rusage(nullptr, &writer, job_report, job, ja_task, category_string, &usage_pattern_list, 0,
+                                false, false);
+         sge_mutex_unlock(config_mutex_name.c_str(), __func__, __LINE__, &config_mutex);
 
          if (ret) {
             // append data to buffer
             json_buffer.Put('\n');
-            sge_mutex_lock(typeid(*this).name(), __func__, __LINE__, &mutex);
+            sge_mutex_lock(typeid(*this).name(), __func__, __LINE__, &buffer_mutex);
             buffer += json_buffer.GetString();
-            sge_mutex_unlock(typeid(*this).name(), __func__, __LINE__, &mutex);
+            sge_mutex_unlock(typeid(*this).name(), __func__, __LINE__, &buffer_mutex);
 
             // If immediate flushing is enabled, flush the buffer now
             if (accounting_immediate_flush) {
@@ -66,4 +70,3 @@ namespace ocs {
       DRETURN(ret);
    }
 }
-

source/daemons/qmaster/ocs_JsonReportingFileWriter.cc

@@ -58,14 +58,19 @@ namespace ocs {
       // get accounting data
       rapidjson::StringBuffer json_buffer;
       rapidjson::Writer<rapidjson::StringBuffer> writer(json_buffer);
-      bool ret = sge_write_rusage(nullptr, &writer, job_report, job, ja_task, category_string,
-                                  &usage_pattern_list, 0, false, true);
+
+      // we need to protect the usage_pattern_list via the config_mutex
+      sge_mutex_lock(config_mutex_name.c_str(), __func__, __LINE__, &config_mutex);
+      bool ret = sge_write_rusage(nullptr, &writer, job_report, job, ja_task, category_string, &usage_pattern_list, 0,
+                             false, true);
+      sge_mutex_unlock(config_mutex_name.c_str(), __func__, __LINE__, &config_mutex);
+
       if (ret) {
          // append data to buffer
          json_buffer.Put('\n');
-         sge_mutex_lock(typeid(*this).name(), __func__, __LINE__, &mutex);
+         sge_mutex_lock(typeid(*this).name(), __func__, __LINE__, &buffer_mutex);
          buffer += json_buffer.GetString();
-         sge_mutex_unlock(typeid(*this).name(), __func__, __LINE__, &mutex);
+         sge_mutex_unlock(typeid(*this).name(), __func__, __LINE__, &buffer_mutex);
       }
 
       DRETURN(ret);
@@ -75,9 +80,9 @@ namespace ocs {
    JsonReportingFileWriter::create_record(rapidjson::StringBuffer &stringBuffer) {
       stringBuffer.Put('\n');
 
-      sge_mutex_lock(typeid(*this).name(), __func__, __LINE__, &mutex);
+      sge_mutex_lock(typeid(*this).name(), __func__, __LINE__, &buffer_mutex);
       buffer += stringBuffer.GetString();
-      sge_mutex_unlock(typeid(*this).name(), __func__, __LINE__, &mutex);
+      sge_mutex_unlock(typeid(*this).name(), __func__, __LINE__, &buffer_mutex);
    }
 
    bool

source/daemons/qmaster/ocs_MonitoringFileWriter.cc

@@ -40,10 +40,10 @@ namespace ocs {
       DENTER(TOP_LAYER);
 
       // append data to buffer
-      sge_mutex_lock(typeid(*this).name(), __func__, __LINE__, &mutex);
+      sge_mutex_lock(typeid(*this).name(), __func__, __LINE__, &buffer_mutex);
       this->buffer += json_data;
       this->buffer += "\n";
-      sge_mutex_unlock(typeid(*this).name(), __func__, __LINE__, &mutex);
+      sge_mutex_unlock(typeid(*this).name(), __func__, __LINE__, &buffer_mutex);
 
       DRETURN(true);
    }