crushtool -c crushmap.txt -o crushmap-new.bin

#方式1： 指定使用rule 1(即ruleset 5),映射[0,10]这11个PG
crushtool --test -i crushmap-new.bin --show-mappings --rule 1 --num-rep=3 --min_x=0 --max_x=10

#方式2： 指定使用ruleset 5,映射[0,10]这11个PG
crushtool --test -i crushmap-new.bin --show-mappings --ruleset 5 --num-rep=3 --min_x=0 --max_x=10

#方式3： 单独指定映射某个PG
crushtool --test -i crushmap-new.bin --show-mappings --ruleset 5 --num-rep=3 --x=100

int CrushTester::test()
{
  if (min_rule < 0 || max_rule < 0) {
    min_rule = 0;
    max_rule = crush.get_max_rules() - 1;
  }
  if (min_x < 0 || max_x < 0) {
    min_x = 0;
    max_x = 1023;
  }

  // initial osd weights
  vector<__u32> weight;

  /*
   * note device weight is set by crushtool
   * (likely due to a given a command line option)
   */
  for (int o = 0; o < crush.get_max_devices(); o++) {
    if (device_weight.count(o)) {
      weight.push_back(device_weight[o]);
    } else if (crush.check_item_present(o)) {
      weight.push_back(0x10000);
    } else {
      weight.push_back(0);
    }
  }

  if (output_utilization_all)
    err << "devices weights (hex): " << hex << weight << dec << std::endl;

  // make adjustments
  adjust_weights(weight);


  int num_devices_active = 0;
  for (vector<__u32>::iterator p = weight.begin(); p != weight.end(); ++p)
    if (*p > 0)
      num_devices_active++;

  if (output_choose_tries)
    crush.start_choose_profile();
  
  for (int r = min_rule; r < crush.get_max_rules() && r <= max_rule; r++) {
    if (!crush.rule_exists(r)) {
      if (output_statistics)
        err << "rule " << r << " dne" << std::endl;
      continue;
    }
    if (ruleset >= 0 &&
	crush.get_rule_mask_ruleset(r) != ruleset) {
      continue;
    }
    int minr = min_rep, maxr = max_rep;
    if (min_rep < 0 || max_rep < 0) {
      minr = crush.get_rule_mask_min_size(r);
      maxr = crush.get_rule_mask_max_size(r);
    }
    
    if (output_statistics)
      err << "rule " << r << " (" << crush.get_rule_name(r)
      << "), x = " << min_x << ".." << max_x
      << ", numrep = " << minr << ".." << maxr
      << std::endl;

    for (int nr = minr; nr <= maxr; nr++) {
      vector<int> per(crush.get_max_devices());
      map<int,int> sizes;

      int num_objects = ((max_x - min_x) + 1);
      float num_devices = (float) per.size(); // get the total number of devices, better to cast as a float here 

      // create a structure to hold data for post-processing
      tester_data_set tester_data;
      vector<int> vector_data_buffer;
      vector<float> vector_data_buffer_f;

      // create a map to hold batch-level placement information
      map<int, vector<int> > batch_per;
      int objects_per_batch = num_objects / num_batches;
      int batch_min = min_x;
      int batch_max = min_x + objects_per_batch - 1;

      // get the total weight of the system
      int total_weight = 0;
      for (unsigned i = 0; i < per.size(); i++)
        total_weight += weight[i];

      if (total_weight == 0)
	continue;

      // compute the expected number of objects stored per device in the absence of weighting
      float expected_objects = min(nr, get_maximum_affected_by_rule(r)) * num_objects;

      // compute each device's proportional weight
      vector<float> proportional_weights( per.size() );

      for (unsigned i = 0; i < per.size(); i++)
        proportional_weights[i] = (float) weight[i] / (float) total_weight;

      if (output_data_file) {
        // stage the absolute weight information for post-processing
        for (unsigned i = 0; i < per.size(); i++) {
          tester_data.absolute_weights[i] = (float) weight[i] / (float)0x10000;
        }

        // stage the proportional weight information for post-processing
        for (unsigned i = 0; i < per.size(); i++) {
          if (proportional_weights[i] > 0 )
            tester_data.proportional_weights[i] = proportional_weights[i];

          tester_data.proportional_weights_all[i] = proportional_weights[i];
        }

      }
      // compute the expected number of objects stored per device when a device's weight is considered
      vector<float> num_objects_expected(num_devices);

      for (unsigned i = 0; i < num_devices; i++)
        num_objects_expected[i] = (proportional_weights[i]*expected_objects);

      for (int current_batch = 0; current_batch < num_batches; current_batch++) {
        if (current_batch == (num_batches - 1)) {
          batch_max = max_x;
          objects_per_batch = (batch_max - batch_min + 1);
        }

        float batch_expected_objects = min(nr, get_maximum_affected_by_rule(r)) * objects_per_batch;
        vector<float> batch_num_objects_expected( per.size() );

        for (unsigned i = 0; i < per.size() ; i++)
          batch_num_objects_expected[i] = (proportional_weights[i]*batch_expected_objects);

        // create a vector to hold placement results temporarily 
        vector<int> temporary_per ( per.size() );

        for (int x = batch_min; x <= batch_max; x++) {
          // create a vector to hold the results of a CRUSH placement or RNG simulation
          vector<int> out;

          if (use_crush) {
            if (output_mappings)
	      err << "CRUSH"; // prepend CRUSH to placement output
            crush.do_rule(r, x, out, nr, weight);
          } else {
            if (output_mappings)
	      err << "RNG"; // prepend RNG to placement output to denote simulation
            // test our new monte carlo placement generator
            random_placement(r, out, nr, weight);
          }

	  if (output_mappings)
	    err << " rule " << r << " x " << x << " " << out << std::endl;

          if (output_data_file)
            write_integer_indexed_vector_data_string(tester_data.placement_information, x, out);

          bool has_item_none = false;
          for (unsigned i = 0; i < out.size(); i++) {
            if (out[i] != CRUSH_ITEM_NONE) {
              per[out[i]]++;
              temporary_per[out[i]]++;
            } else {
              has_item_none = true;
            }
          }

          batch_per[current_batch] = temporary_per;
          sizes[out.size()]++;
          if (output_bad_mappings && 
              (out.size() != (unsigned)nr ||
               has_item_none)) {
            err << "bad mapping rule " << r << " x " << x << " num_rep " << nr << " result " << out << std::endl;
          }
        }

        batch_min = batch_max + 1;
        batch_max = batch_min + objects_per_batch - 1;
      }

      for (unsigned i = 0; i < per.size(); i++)
        if (output_utilization && !output_statistics)
          err << "  device " << i
          << ":\t" << per[i] << std::endl;

      for (map<int,int>::iterator p = sizes.begin(); p != sizes.end(); ++p)
        if (output_statistics)
          err << "rule " << r << " (" << crush.get_rule_name(r) << ") num_rep " << nr
          << " result size == " << p->first << ":\t"
          << p->second << "/" << (max_x-min_x+1) << std::endl;

      if (output_statistics)
        for (unsigned i = 0; i < per.size(); i++) {
          if (output_utilization) {
            if (num_objects_expected[i] > 0 && per[i] > 0) {
              err << "  device " << i << ":\t"
                  << "\t" << " stored " << ": " << per[i]
                  << "\t" << " expected " << ": " << num_objects_expected[i]
                  << std::endl;
            }
          } else if (output_utilization_all) {
            err << "  device " << i << ":\t"
                << "\t" << " stored " << ": " << per[i]
                << "\t" << " expected " << ": " << num_objects_expected[i]
                << std::endl;
          }
        }

      if (output_data_file)
        for (unsigned i = 0; i < per.size(); i++) {
          vector_data_buffer_f.clear();
          vector_data_buffer_f.push_back( (float) per[i]);
          vector_data_buffer_f.push_back( (float) num_objects_expected[i]);

          write_integer_indexed_vector_data_string(tester_data.device_utilization_all, i, vector_data_buffer_f);

          if (num_objects_expected[i] > 0 && per[i] > 0)
            write_integer_indexed_vector_data_string(tester_data.device_utilization, i, vector_data_buffer_f);
        }

      if (output_data_file && num_batches > 1) {
        // stage batch utilization information for post-processing
        for (int i = 0; i < num_batches; i++) {
          write_integer_indexed_vector_data_string(tester_data.batch_device_utilization_all, i, batch_per[i]);
          write_integer_indexed_vector_data_string(tester_data.batch_device_expected_utilization_all, i, batch_per[i]);
        }
      }

      string rule_tag = crush.get_rule_name(r);

      if (output_csv)
        write_data_set_to_csv(output_data_file_name+rule_tag,tester_data);
    }
  }

  if (output_choose_tries) {
    __u32 *v = 0;
    int n = crush.get_choose_profile(&v);
    for (int i=0; i<n; i++) {
      cout.setf(std::ios::right);
      cout << std::setw(2)
      << i << ": " << std::setw(9) << v[i];
      cout.unsetf(std::ios::right);
      cout << std::endl;
    }

    crush.stop_choose_profile();
  }

  return 0;
}

// initial osd weights
  vector<__u32> weight;

  /*
   * note device weight is set by crushtool
   * (likely due to a given a command line option)
   */
  for (int o = 0; o < crush.get_max_devices(); o++) {
    if (device_weight.count(o)) {
      weight.push_back(device_weight[o]);
    } else if (crush.check_item_present(o)) {
      weight.push_back(0x10000);
    } else {
      weight.push_back(0);
    }
  }

  if (output_utilization_all)
    err << "devices weights (hex): " << hex << weight << dec << std::endl;

  // make adjustments
  adjust_weights(weight);

  int num_devices_active = 0;
  for (vector<__u32>::iterator p = weight.begin(); p != weight.end(); ++p)
    if (*p > 0)
      num_devices_active++;

for (int r = min_rule; r < crush.get_max_rules() && r <= max_rule; r++) {

    //1: 检测指定的rule是否存在，以及是否是我们使用crushtool时指定的ruleset 5
    
    //2: 获得表示副本数的minr,maxr。 由于我们在参数中传入了--num-rep=3，因此这里minr=maxr=3

    for (int nr = minr; nr <= maxr; nr++) {
        
         per.size() = 9;
         num_objects = 11;
         num_devices = 9;
         num_batches = 1;
         objects_per_batch = 11;
         batch_min = min_x;
         batch_max = min_x + objects_per_batch - 1;
        

         //3: 计算total_weight(9*0x10000)

         //4: get_maximum_affected_by_rule(r)，这里采用ruleset 5时，结果为1。因此expected_objects = 11

         //5: 计算每一个device的权重占比，这里为均为1/9

         //6: 根据权重占比计算平均每个device所存储的对象数
         for (unsigned i = 0; i < num_devices; i++)
         num_objects_expected[i] = (proportional_weights[i]*expected_objects);

        
         //7: 遍历当前所有的batches
         for (int current_batch = 0; current_batch < num_batches; current_batch++) {

                //nr为当前的副本数，这里取值为3；object_per_batch取值为11，因为我们只有一个batch
                float batch_expected_objects = min(nr, get_maximum_affected_by_rule(r)) * objects_per_batch;

                
                //对当前batch内的所有PG做映射,这里batch_min为0,batch_max为10，及映射PG0~PG10
                for (int x = batch_min; x <= batch_max; x++) {
                      if (use_crush) {
                          if (output_mappings)
	                          err << "CRUSH"; // prepend CRUSH to placement output


                      //8: do_rule是关键的PG映射函数。 第一个参数当前所采用的rule；第二个参数为所映射的PG；第三个参数为映射结果；
                      // 第四个参数为当前的副本数； 第五个参数为各device的权重列表
                      crush.do_rule(r, x, out, nr, weight);
                      } else {
                       if (output_mappings)
	                       err << "RNG"; // prepend RNG to placement output to denote simulation

                           // test our new monte carlo placement generator
                           random_placement(r, out, nr, weight);
                      }
                }

         }

       
       //9: 
      

	}
}