int check_all_block_in_disk(const char* ns_addr, const char* ns_slave_addr, const uint64_t ds_id,
const VUINT64& blocks, VUINT64& need_sync_block_list, VUINT64& lost_block_list)
{
int ret = TFS_SUCCESS;
uint64_t ns_id = Func::get_host_ip(ns_addr);
uint64_t slave_ns_id = Func::get_host_ip(ns_slave_addr);
need_sync_block_list.clear();
lost_block_list.clear();
// check current cluster's block in blocks array
ret = get_and_check_all_blocks_copy(ns_id, blocks, need_sync_block_list, ds_id);
if (TFS_SUCCESS == ret && need_sync_block_list.size() > 0)
{
// check slave cluster's block in need_sync_block_list array
VUINT64 raw_lost_blocks;
ret = get_and_check_all_blocks_copy(slave_ns_id, need_sync_block_list, raw_lost_blocks);
// raw_lost_blocks need unique, then remove all blocks which not in need_sync_block_list
set<uint64_t> need_sync_block_set, lost_block_set;
need_sync_block_set.insert(need_sync_block_list.begin(), need_sync_block_list.end()); // to sort
lost_block_set.insert(raw_lost_blocks.begin(), raw_lost_blocks.end()); // to sort & unique
lost_block_list.resize(raw_lost_blocks.size());
VUINT64::iterator it = set_intersection(raw_lost_blocks.begin(), raw_lost_blocks.end(),
need_sync_block_list.begin(), need_sync_block_list.end(), lost_block_list.begin());
lost_block_list.resize(it - lost_block_list.begin());// set real size
}
return ret;
}
void print_result(const VUINT64& need_sync_block_list, const VUINT64& lost_block_list)
{
if (lost_block_list.size() > 0)
{
fprintf(stdout, "BOTH LOST BLOCK COUNT: %zd\n", lost_block_list.size());
for (uint32_t i = 0; i < lost_block_list.size(); i++)
{
fprintf(stdout, "%" PRI64_PREFIX "u\n", lost_block_list.at(i));
}
}
int32_t real_need_sync_block_size = need_sync_block_list.size() - lost_block_list.size();
if (real_need_sync_block_size > 0)
{
fprintf(stdout, "NEED SYNC BLOCK COUNT: %d\n", real_need_sync_block_size);
set<uint64_t> lost_block_set;
lost_block_set.insert(lost_block_list.begin(), lost_block_list.end());
for (uint32_t i = 0; i < need_sync_block_list.size(); i++)
{
uint64_t block_id = need_sync_block_list.at(i);
if (lost_block_set.find(block_id) == lost_block_set.end())
{
fprintf(stdout, "%" PRI64_PREFIX "u\n", block_id);// can sync from slave cluster
}
}
}
}
/*
* get file info from every replica
* return value don't needed, we just need finfos
*/
int read_file_info(const VUINT64& replicas,
const uint64_t block_id, const uint64_t file_id, map<uint64_t, FileInfo>& finfos)
{
finfos.clear();
int ret = TFS_SUCCESS;
VUINT64::const_iterator iter = replicas.begin();
for ( ; iter != replicas.end(); iter++)
{
int32_t retry = 2;
while (retry--)
{
FileInfo info;
ret = ToolUtil::read_file_info(*iter, block_id, file_id, FORCE_STAT, info);
if (TFS_SUCCESS == ret)
{
finfos.insert(make_pair(*iter, info));
break;
}
else if (EXIT_META_NOT_FOUND_ERROR == ret) // file not exist, just ignore
{
break;
}
}
}
return (finfos.size() > 0) ? TFS_SUCCESS : TFS_ERROR;
}
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);
}
}
}
// 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;
}