void recover_block_from_slave_cluster(const char* ns_addr, const char* ns_slave_addr,
    const VUINT64& tmp_fail_block_list, VUINT64& success_block_list,
    VUINT64& fail_block_list, multimap<uint64_t, uint64_t>& fail_block_file_list)
{
  int ret = TFS_SUCCESS;
  VUINT64::const_iterator vit = tmp_fail_block_list.begin();
  multimap<uint64_t, uint64_t> tmp_fail_block_file_list;
  vector<FileInfoV2> finfos;
  for (; vit != tmp_fail_block_list.end(); vit++)
  {
    uint64_t block_id = (*vit);
    finfos.clear();
    ret = ToolUtil::read_file_infos_v2(Func::get_host_ip(ns_slave_addr), block_id, finfos);
    if (ret == TFS_SUCCESS)
    {
      bool all_success = true;
      int32_t copy_file_succ_count = 0;
      tmp_fail_block_file_list.clear();
      vector<FileInfoV2>::const_iterator v_file_iter = finfos.begin();
      for (; v_file_iter != finfos.end(); v_file_iter++)
      {
        if ((v_file_iter->status_ & FILE_STATUS_DELETE) != 0)
          continue;

        uint64_t file_id = v_file_iter->id_;
        ret = copy_file_from_slave_cluster(ns_slave_addr, ns_addr, block_id, (*v_file_iter));
        if (TFS_SUCCESS == ret)
        {
          ++copy_file_succ_count;
          TBSYS_LOG(DEBUG, "recover block_id: %" PRI64_PREFIX "u, file_id: %" PRI64_PREFIX "u successful from slave cluster!", block_id, file_id);
        }
        else
        {
          TBSYS_LOG(WARN, "recover block_id: %" PRI64_PREFIX "u, file_id: %" PRI64_PREFIX "u failed from slave cluster, ret: %d", block_id, file_id, ret);
          all_success = false;
          tmp_fail_block_file_list.insert(pair<uint64_t, uint64_t>(block_id, file_id));
        }
      }

      if (all_success)
      {
        success_block_list.push_back(block_id);
      }
      else if (0 == copy_file_succ_count)
      {
        fail_block_list.push_back(block_id);
      }
      else
      {
        fail_block_file_list.insert(tmp_fail_block_file_list.begin(), tmp_fail_block_file_list.end());
      }
    }
    else
    {
      fail_block_list.push_back(block_id);
    }
  }
}
Esempio n. 2
0
    int32_t elect_ds_normal(const DS_WEIGHT& weights, const int32_t elect_count, int64_t& elect_seq,
        VUINT64& elect_ds_list)
    {
      if (elect_count == 0)
        return 0;

      int32_t need_elect_count = elect_count;

      DS_WEIGHT::const_iterator iter = weights.begin();
      while (iter != weights.end() && need_elect_count > 0)
      {
        elect_ds_list.push_back(iter->second->get_ds()->id_);
        if (elect_seq > 0)
          iter->second->elect(++elect_seq);
        --need_elect_count;
        ++iter;
      }
      return elect_count - need_elect_count;
    }
Esempio n. 3
0
    int32_t elect_ds_exclude_group(const DS_WEIGHT& weights, const int32_t elect_count, int64_t& elect_seq,
        VUINT64& elect_ds_list)
    {
      if (elect_count == 0)
      {
        TBSYS_LOG(DEBUG, "current elect count(%d) <= 0, must be return", elect_count);
        return 0;
      }

      std::set < uint32_t > existlan;
      for (uint32_t i = 0; i < elect_ds_list.size(); ++i)
      {
        uint32_t lan = Func::get_lan(elect_ds_list[i], SYSPARAM_NAMESERVER.group_mask_);
        existlan.insert(lan);
      }

      //dump_weigths(weights);

      DS_WEIGHT::const_iterator iter = weights.begin();
      int32_t need_elect_count = elect_count;
      TBSYS_LOG(DEBUG, "weights.size(%u), need_elect_count(%d)", weights.size(), need_elect_count);
      DataServerStatInfo* ds_stat_info = NULL;
      while (iter != weights.end() && need_elect_count > 0)
      {
        ds_stat_info = iter->second->get_ds();
        uint32_t dlan = Func::get_lan(ds_stat_info->id_, SYSPARAM_NAMESERVER.group_mask_);
        if (existlan.find(dlan) == existlan.end())
        {
          existlan.insert(dlan);
          elect_ds_list.push_back(ds_stat_info->id_);
          if (elect_seq > 0)
            iter->second->elect(++elect_seq);
          --need_elect_count;
        }
        ++iter;
      }
      TBSYS_LOG(DEBUG, "current elect_count(%d)", elect_count - need_elect_count);
      return elect_count - need_elect_count;
    }
Esempio n. 4
0
// if parameter: ds_id != INVALID_SERVER_ID, check from current cluster, need exclude repliates in this ds;
// if parameter: ds_id == INVALID_SERVER_ID, check from slave cluster
int get_and_check_all_blocks_copy(const uint64_t ns_id, const VUINT64& blocks, VUINT64& no_copy_blocks, const uint64_t ds_id = INVALID_SERVER_ID)
{
  no_copy_blocks.clear();
  vector<BlockMeta> blocks_meta;
  int ret = ToolUtil::get_all_blocks_meta(ns_id, blocks, blocks_meta, true);
  if (TFS_SUCCESS == ret)
  {
    VUINT64 lost_in_family;
    for (uint32_t block_index = 0; block_index < blocks_meta.size(); ++block_index)
    {
      // blk_meta.result_ just be used in trunk(stable2.7)
      BlockMeta blk_meta = blocks_meta.at(block_index);
      uint64_t block_id = blk_meta.block_id_;
      FamilyInfoExt family_info = blk_meta.family_info_;
      if (INVALID_FAMILY_ID == family_info.family_id_)
      {
        // if family can't reinstall, family_id here is also invalid
        if (INVALID_BLOCK_ID == block_id || IS_VERFIFY_BLOCK(block_id))
          continue;
        // else family data block regard as normal block

        if (0 == blk_meta.size_)
        {
          // maybe not exist the block
          no_copy_blocks.push_back(block_id);
        }
        else if (INVALID_SERVER_ID != ds_id && 1 == blk_meta.size_ && ds_id == blk_meta.ds_[0])
        {
          // only exist one copy and be positioned in this ds
          no_copy_blocks.push_back(block_id);
        }
      }
      else // block is marshalling
      {
        int data_num = GET_DATA_MEMBER_NUM(family_info.family_aid_info_);
        int member_num = data_num + GET_CHECK_MEMBER_NUM(family_info.family_aid_info_);
        int32_t alive_num = 0;
        lost_in_family.clear();
        for (int i = 0; i < member_num; ++i)
        {
          // pair is <blockid, ds_id>
          pair<uint64_t, uint64_t>& item = family_info.members_[i];
          if (INVALID_SERVER_ID != item.second && item.second != ds_id) // exclude current ds
          {
            ++alive_num;
          }
          else
          {
            // keep all lost(and will lost) data blockid
            if (!IS_VERFIFY_BLOCK(item.first) && INVALID_BLOCK_ID != item.first)
            {
              lost_in_family.push_back(item.first);
            }
          }
        }
        if (alive_num < data_num) // can't reinstate
        {
          no_copy_blocks.insert(no_copy_blocks.end(), lost_in_family.begin(), lost_in_family.end());
        }
      }
    }
  }
  else
  {
    // this error need script-calller care
    TBSYS_LOG(ERROR, "get blockis ds_list fail from ns: %s, ret:%d",
        tbsys::CNetUtil::addrToString(ns_id).c_str(), ret);
  }
  return ret;
}
int recover_block_from_disk_data(const char* ns_addr, const char* ns_slave_addr, BlockManager& block_manager,
    VUINT64& tmp_fail_block_list, VUINT64& no_need_recover_block_list, VUINT64& success_block_list,
    multimap<uint64_t, uint64_t>& fail_block_file_list, VUINT64& fail_block_list)
{
  int ret = TFS_SUCCESS;
  VUINT64 blocks;
  vector<FileInfoV2> finfos;
  multimap<uint64_t, uint64_t> tmp_fail_block_file_list;
  uint64_t ns_id = Func::get_host_ip(ns_addr);

  //get all blocks from ds disk
  block_manager.get_all_block_ids(blocks);//肯定返回成功

  vector<BlockMeta> blocks_meta;
  ret = ToolUtil::get_all_blocks_meta(ns_id, blocks, blocks_meta, false);// no need get check block
  TBSYS_LOG(DEBUG , "all logic blocks count: %zd, data block count: %zd", blocks.size(), blocks_meta.size());
  if (TFS_SUCCESS == ret)
  {
    int32_t bret = TFS_SUCCESS;
    for (uint32_t block_index = 0; block_index < blocks_meta.size(); ++block_index)
    {
      BlockMeta blk_meta = blocks_meta.at(block_index);
      uint64_t block_id = blk_meta.block_id_;
      if (blk_meta.size_ > 0 || INVALID_FAMILY_ID != blk_meta.family_info_.family_id_)
      {
        TBSYS_LOG(DEBUG , "blockid: %" PRI64_PREFIX "u  no need recover, ds_size:%d, family_id: %" PRI64_PREFIX "u",
            block_id, blk_meta.size_, blk_meta.family_info_.family_id_);
        no_need_recover_block_list.push_back(block_id);//只要还有副本或者副本丢失但有编组(不考虑退化读恢复)都不用本工具恢复
      }
      else
      {
        bret = rm_no_replicate_block_from_ns(ns_addr, block_id);//now T_NEWBLK will can not remove empty ds_list's block
        if (TFS_SUCCESS != bret)
        {
          fail_block_list.push_back(block_id);
          TBSYS_LOG(WARN , "remove block %" PRI64_PREFIX "u from ns: %s failed, ret: %d", block_id, ns_addr, bret);
        }
        else
        {
          IndexHeaderV2 header;
          finfos.clear();
          bret = block_manager.traverse(header, finfos, block_id, block_id);
          if (TFS_SUCCESS != bret)
          {
            TBSYS_LOG(WARN , "block %" PRI64_PREFIX "u get local file infos failed, ret: %d", block_id, bret);
            tmp_fail_block_list.push_back(block_id);//只有本磁盘读取block的文件index错误,才需要整个block尝试从辅集群恢复
          }
          else
          {
            bool all_success = true;
            int32_t copy_file_succ_count = 0;
            tmp_fail_block_file_list.clear();
            for (uint32_t file_index = 0; file_index < finfos.size(); ++file_index)
            {
              // skip deleted file
              if ((finfos.at(file_index).status_ & FILE_STATUS_DELETE) != 0)
                continue;

              uint64_t file_id = finfos.at(file_index).id_;
              bret = copy_file(block_manager, block_id, finfos.at(file_index));
              if (TFS_SUCCESS == bret)
              {
                TBSYS_LOG(DEBUG, "recover block_id: %" PRI64_PREFIX "u, file_id: %" PRI64_PREFIX "u successful!", block_id, file_id);
              }
              else
              {// 如果磁盘中该文件数据已经损坏(如crc出错),则从对等集群(ns_slave_addr)拷贝数据复制
                bret = copy_file_from_slave_cluster(ns_slave_addr, ns_addr, block_id, finfos.at(file_index));
                if (TFS_SUCCESS == bret)
                {
                  TBSYS_LOG(DEBUG, "recover block_id: %" PRI64_PREFIX "u, file_id: %" PRI64_PREFIX "u successful from slave cluster!", block_id, file_id);
                }
                else
                {
                  TBSYS_LOG(WARN, "recover block_id: %" PRI64_PREFIX "u, file_id: %" PRI64_PREFIX "u failed from slave cluster, ret: %d!", block_id, file_id, bret);
                  all_success = false;
                  tmp_fail_block_file_list.insert(pair<uint64_t, uint64_t>(block_id, file_id));
                }
              }
              if (TFS_SUCCESS == bret)
              {
                ++copy_file_succ_count;
              }
            }

            if (all_success)
            {
              success_block_list.push_back(block_id);
              if (0 == copy_file_succ_count)
              {
                TBSYS_LOG(DEBUG, "recover block_id: %" PRI64_PREFIX "u need to do nothing,"
                    " because the count of files who need to recover is ZERO except DELETE status files!", block_id);
              }
            }
            else if (0 == copy_file_succ_count)// all file(exclude DELETE) fail
            {
              TBSYS_LOG(WARN, "recover block_id: %" PRI64_PREFIX "u's files failed, copy_file_succ_count is Zero !", block_id);
              fail_block_list.push_back(block_id);// for print fail block to out log file at end
            }
            else
            {
              fail_block_file_list.insert(tmp_fail_block_file_list.begin(), tmp_fail_block_file_list.end());
            }
          }
        }
      }
    }//end for blocks loop
  }
  else
  {
    TBSYS_LOG(WARN, "get blockis ds_list error, ret:%d", ret);
  }
  TBSYS_LOG(INFO, "success_block_list size: %zd, tmp_fail_block_list size: %zd, fail_block_list size: %zd",
      success_block_list.size(), tmp_fail_block_list.size(), fail_block_list.size());
  return ret;
}