be/src/storage/storage_engine.cpp

Source
// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements.  See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership.  The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License.  You may obtain a copy of the License at
//
//   http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, either express or implied.  See the License for the
// specific language governing permissions and limitations
// under the License.

#include "storage/storage_engine.h"

// IWYU pragma: no_include <bthread/errno.h>
#include <fmt/format.h>
#include <gen_cpp/AgentService_types.h>
#include <gen_cpp/FrontendService.h>
#include <gen_cpp/Types_types.h>
#include <glog/logging.h>
#include <rapidjson/document.h>
#include <rapidjson/encodings.h>
#include <rapidjson/prettywriter.h>
#include <rapidjson/stringbuffer.h>
#include <sys/resource.h>
#include <thrift/protocol/TDebugProtocol.h>

#include <algorithm>
#include <boost/algorithm/string/case_conv.hpp>
#include <boost/container/detail/std_fwd.hpp>
#include <cassert>
#include <cerrno> // IWYU pragma: keep
#include <chrono>
#include <cstdlib>
#include <cstring>
#include <filesystem>
#include <iterator>
#include <memory>
#include <mutex>
#include <ostream>
#include <set>
#include <thread>
#include <unordered_set>
#include <utility>

#include "agent/task_worker_pool.h"
#include "cloud/cloud_storage_engine.h"
#include "common/config.h"
#include "common/logging.h"
#include "common/metrics/doris_metrics.h"
#include "common/metrics/metrics.h"
#include "common/status.h"
#include "core/assert_cast.h"
#include "io/fs/local_file_system.h"
#include "load/memtable/memtable_flush_executor.h"
#include "load/stream_load/stream_load_recorder.h"
#include "runtime/cluster_info.h"
#include "runtime/exec_env.h"
#include "storage/binlog.h"
#include "storage/data_dir.h"
#include "storage/id_manager.h"
#include "storage/olap_common.h"
#include "storage/olap_define.h"
#include "storage/rowset/rowset_fwd.h"
#include "storage/rowset/rowset_meta.h"
#include "storage/rowset/rowset_meta_manager.h"
#include "storage/rowset/unique_rowset_id_generator.h"
#include "storage/snapshot/snapshot_manager.h"
#include "storage/tablet/tablet_manager.h"
#include "storage/tablet/tablet_meta.h"
#include "storage/tablet/tablet_meta_manager.h"
#include "storage/txn/txn_manager.h"
#include "util/client_cache.h"
#include "util/mem_info.h"
#include "util/stopwatch.hpp"
#include "util/thread.h"
#include "util/threadpool.h"
#include "util/thrift_rpc_helper.h"
#include "util/uid_util.h"
#include "util/work_thread_pool.hpp"

using std::filesystem::directory_iterator;
using std::filesystem::path;
using std::map;
using std::set;
using std::string;
using std::stringstream;
using std::vector;

namespace doris {
using namespace ErrorCode;
extern void get_round_robin_stores(int64_t curr_index, const std::vector<DirInfo>& dir_infos,
                                   std::vector<DataDir*>& stores);
DEFINE_GAUGE_METRIC_PROTOTYPE_2ARG(unused_rowsets_count, MetricUnit::ROWSETS);
bvar::Status<int64_t> g_max_rowsets_with_useless_delete_bitmap(
        "max_rowsets_with_useless_delete_bitmap", 0);
bvar::Status<int64_t> g_max_rowsets_with_useless_delete_bitmap_version(
        "max_rowsets_with_useless_delete_bitmap_version", 0);

namespace {
bvar::Adder<uint64_t> unused_rowsets_counter("ununsed_rowsets_counter");
};

BaseStorageEngine::BaseStorageEngine(Type type, const UniqueId& backend_uid)
        : _type(type),
          _rowset_id_generator(std::make_unique<UniqueRowsetIdGenerator>(backend_uid)),
          _stop_background_threads_latch(1) {
    _memory_limitation_bytes_for_schema_change = static_cast<int64_t>(
            static_cast<double>(MemInfo::soft_mem_limit()) * config::schema_change_mem_limit_frac);
    _tablet_max_delete_bitmap_score_metrics =
            std::make_shared<bvar::Status<size_t>>("tablet_max", "delete_bitmap_score", 0);
    _tablet_max_base_rowset_delete_bitmap_score_metrics = std::make_shared<bvar::Status<size_t>>(
            "tablet_max_base_rowset", "delete_bitmap_score", 0);
}

BaseStorageEngine::~BaseStorageEngine() = default;

RowsetId BaseStorageEngine::next_rowset_id() {
    return _rowset_id_generator->next_id();
}

StorageEngine& BaseStorageEngine::to_local() {
    CHECK_EQ(_type, Type::LOCAL);
    return *static_cast<StorageEngine*>(this);
}

CloudStorageEngine& BaseStorageEngine::to_cloud() {
    CHECK_EQ(_type, Type::CLOUD);
    return *static_cast<CloudStorageEngine*>(this);
}

int64_t BaseStorageEngine::memory_limitation_bytes_per_thread_for_schema_change() const {
    return std::max(_memory_limitation_bytes_for_schema_change / config::alter_tablet_worker_count,
                    config::memory_limitation_per_thread_for_schema_change_bytes);
}

void BaseStorageEngine::_start_adaptive_thread_controller() {
    if (!config::enable_adaptive_flush_threads) {
        return;
    }

    auto* system_metrics = DorisMetrics::instance()->system_metrics();
    auto* s3_upload_pool = ExecEnv::GetInstance()->s3_file_upload_thread_pool();

    _adaptive_thread_controller.init(system_metrics, s3_upload_pool);

    if (_memtable_flush_executor) {
        auto* flush_pool = _memtable_flush_executor->flush_pool();
        auto* high_prio_pool = _memtable_flush_executor->high_prio_flush_pool();
        _adaptive_thread_controller.add("flush", {flush_pool, high_prio_pool},
                                        AdaptiveThreadPoolController::make_flush_adjust_func(
                                                &_adaptive_thread_controller, flush_pool),
                                        config::max_flush_thread_num_per_cpu,
                                        config::min_flush_thread_num_per_cpu);
    }
}

Status BaseStorageEngine::init_stream_load_recorder(const std::string& stream_load_record_path) {
    LOG(INFO) << "stream load record path: " << stream_load_record_path;
    // init stream load record rocksdb
    _stream_load_recorder = StreamLoadRecorder::create_shared(stream_load_record_path);
    if (_stream_load_recorder == nullptr) {
        RETURN_NOT_OK_STATUS_WITH_WARN(
                Status::MemoryAllocFailed("allocate memory for StreamLoadRecorder failed"),
                "new StreamLoadRecorder failed");
    }
    auto st = _stream_load_recorder->init();
    if (!st.ok()) {
        RETURN_NOT_OK_STATUS_WITH_WARN(
                Status::IOError("open StreamLoadRecorder rocksdb failed, path={}",
                                stream_load_record_path),
                "init StreamLoadRecorder failed");
    }
    return Status::OK();
}

void CompactionSubmitRegistry::jsonfy_compaction_status(std::string* result) {
    rapidjson::Document root;
    root.SetObject();

    auto add_node = [&root](const std::string& name, const Registry& registry) {
        rapidjson::Value compaction_name;
        compaction_name.SetString(name.c_str(), cast_set<uint32_t>(name.length()),
                                  root.GetAllocator());
        rapidjson::Document path_obj;
        path_obj.SetObject();
        for (const auto& it : registry) {
            const auto& dir = it.first->path();
            rapidjson::Value path_key;
            path_key.SetString(dir.c_str(), cast_set<uint32_t>(dir.length()), root.GetAllocator());

            rapidjson::Document arr;
            arr.SetArray();

            for (const auto& tablet : it.second) {
                rapidjson::Value tablet_id;
                auto tablet_id_str = std::to_string(tablet->tablet_id());
                tablet_id.SetString(tablet_id_str.c_str(),
                                    cast_set<uint32_t>(tablet_id_str.length()),
                                    root.GetAllocator());
                arr.PushBack(tablet_id, root.GetAllocator());
            }
            path_obj.AddMember(path_key, arr, root.GetAllocator());
        }
        root.AddMember(compaction_name, path_obj, root.GetAllocator());
    };

    std::unique_lock<std::mutex> l(_tablet_submitted_compaction_mutex);
    add_node("BaseCompaction", _tablet_submitted_base_compaction);
    add_node("CumulativeCompaction", _tablet_submitted_cumu_compaction);
    add_node("FullCompaction", _tablet_submitted_full_compaction);

    rapidjson::StringBuffer str_buf;
    rapidjson::PrettyWriter<rapidjson::StringBuffer> writer(str_buf);
    root.Accept(writer);
    *result = std::string(str_buf.GetString());
}

static Status _validate_options(const EngineOptions& options) {
    if (options.store_paths.empty()) {
        return Status::InternalError("store paths is empty");
    }
    return Status::OK();
}

Status StorageEngine::open() {
    RETURN_IF_ERROR(_validate_options(_options));
    LOG(INFO) << "starting backend using uid:" << _options.backend_uid.to_string();
    RETURN_NOT_OK_STATUS_WITH_WARN(_open(), "open engine failed");
    LOG(INFO) << "success to init storage engine.";
    return Status::OK();
}

StorageEngine::StorageEngine(const EngineOptions& options)
        : BaseStorageEngine(Type::LOCAL, options.backend_uid),
          _options(options),
          _available_storage_medium_type_count(0),
          _is_all_cluster_id_exist(true),
          _stopped(false),
          _tablet_manager(new TabletManager(*this, config::tablet_map_shard_size)),
          _txn_manager(new TxnManager(*this, config::txn_map_shard_size, config::txn_shard_size)),
          _default_rowset_type(BETA_ROWSET),
          _create_tablet_idx_lru_cache(
                  new CreateTabletRRIdxCache(config::partition_disk_index_lru_size)),
          _snapshot_mgr(std::make_unique<SnapshotManager>(*this)) {
    REGISTER_HOOK_METRIC(unused_rowsets_count, [this]() {
        // std::lock_guard<std::mutex> lock(_gc_mutex);
        return _unused_rowsets.size();
    });

    _broken_paths = options.broken_paths;
}

StorageEngine::~StorageEngine() {
    DEREGISTER_HOOK_METRIC(unused_rowsets_count);
    stop();
}

static Status load_data_dirs(const std::vector<DataDir*>& data_dirs) {
    std::unique_ptr<ThreadPool> pool;

    int num_threads = config::load_data_dirs_threads;
    if (num_threads <= 0) {
        num_threads = cast_set<int>(data_dirs.size());
    }

    auto st = ThreadPoolBuilder("load_data_dir")
                      .set_min_threads(num_threads)
                      .set_max_threads(num_threads)
                      .build(&pool);
    CHECK(st.ok()) << st;

    std::mutex result_mtx;
    Status result;

    for (auto* data_dir : data_dirs) {
        st = pool->submit_func([&, data_dir] {
            SCOPED_INIT_THREAD_CONTEXT();
            {
                std::lock_guard lock(result_mtx);
                if (!result.ok()) { // Some data dir has failed
                    return;
                }
            }

            auto st = data_dir->load();
            if (!st.ok()) {
                LOG(WARNING) << "error occurred when init load tables. res=" << st
                             << ", data dir=" << data_dir->path();
                std::lock_guard lock(result_mtx);
                result = std::move(st);
            }
        });

        if (!st.ok()) {
            return st;
        }
    }

    pool->wait();

    return result;
}

Status StorageEngine::_open() {
    // init store_map
    RETURN_NOT_OK_STATUS_WITH_WARN(_init_store_map(), "_init_store_map failed");

    _effective_cluster_id = config::cluster_id;
    RETURN_NOT_OK_STATUS_WITH_WARN(_check_all_root_path_cluster_id(), "fail to check cluster id");

    _update_storage_medium_type_count();

    RETURN_NOT_OK_STATUS_WITH_WARN(_check_file_descriptor_number(), "check fd number failed");

    auto dirs = get_stores();
    RETURN_IF_ERROR(load_data_dirs(dirs));

    _disk_num = cast_set<int>(dirs.size());
    _memtable_flush_executor = std::make_unique<MemTableFlushExecutor>();
    _memtable_flush_executor->init(_disk_num);

    _calc_delete_bitmap_executor = std::make_unique<CalcDeleteBitmapExecutor>();
    _calc_delete_bitmap_executor->init("TabletCalcDeleteBitmapThreadPool",
                                       config::calc_delete_bitmap_max_thread);

    _calc_delete_bitmap_executor_for_load = std::make_unique<CalcDeleteBitmapExecutor>();
    _calc_delete_bitmap_executor_for_load->init(
            "LoadCalcDeleteBitmapThreadPool",
            config::calc_delete_bitmap_for_load_max_thread > 0
                    ? config::calc_delete_bitmap_for_load_max_thread
                    : std::max(1, CpuInfo::num_cores() / 2));

    _parse_default_rowset_type();

    return Status::OK();
}

Status StorageEngine::_init_store_map() {
    std::vector<std::thread> threads;
    std::mutex error_msg_lock;
    std::string error_msg;
    for (auto& path : _options.store_paths) {
        auto store = std::make_unique<DataDir>(*this, path.path, path.capacity_bytes,
                                               path.storage_medium);
        threads.emplace_back([store = store.get(), &error_msg_lock, &error_msg]() {
            SCOPED_INIT_THREAD_CONTEXT();
            auto st = store->init();
            if (!st.ok()) {
                {
                    std::lock_guard<std::mutex> l(error_msg_lock);
                    error_msg.append(st.to_string() + ";");
                }
                LOG(WARNING) << "Store load failed, status=" << st.to_string()
                             << ", path=" << store->path();
            }
        });
        _store_map.emplace(store->path(), std::move(store));
    }
    for (auto& thread : threads) {
        thread.join();
    }

    // All store paths MUST init successfully
    if (!error_msg.empty()) {
        return Status::InternalError("init path failed, error={}", error_msg);
    }

    RETURN_NOT_OK_STATUS_WITH_WARN(init_stream_load_recorder(_options.store_paths[0].path),
                                   "init StreamLoadRecorder failed");

    return Status::OK();
}

void StorageEngine::_update_storage_medium_type_count() {
    set<TStorageMedium::type> available_storage_medium_types;

    std::lock_guard<std::mutex> l(_store_lock);
    for (auto& it : _store_map) {
        if (it.second->is_used()) {
            available_storage_medium_types.insert(it.second->storage_medium());
        }
    }

    _available_storage_medium_type_count =
            cast_set<uint32_t>(available_storage_medium_types.size());
}

Status StorageEngine::_judge_and_update_effective_cluster_id(int32_t cluster_id) {
    if (cluster_id == -1 && _effective_cluster_id == -1) {
        // maybe this is a new cluster, cluster id will get from heartbeat message
        return Status::OK();
    } else if (cluster_id != -1 && _effective_cluster_id == -1) {
        _effective_cluster_id = cluster_id;
        return Status::OK();
    } else if (cluster_id == -1 && _effective_cluster_id != -1) {
        // _effective_cluster_id is the right effective cluster id
        return Status::OK();
    } else {
        if (cluster_id != _effective_cluster_id) {
            RETURN_NOT_OK_STATUS_WITH_WARN(
                    Status::Corruption("multiple cluster ids is not equal. one={}, other={}",
                                       _effective_cluster_id, cluster_id),
                    "cluster id not equal");
        }
    }

    return Status::OK();
}

std::vector<DataDir*> StorageEngine::get_stores(bool include_unused) {
    std::vector<DataDir*> stores;
    stores.reserve(_store_map.size());

    std::lock_guard<std::mutex> l(_store_lock);
    if (include_unused) {
        for (auto&& [_, store] : _store_map) {
            stores.push_back(store.get());
        }
    } else {
        for (auto&& [_, store] : _store_map) {
            if (store->is_used()) {
                stores.push_back(store.get());
            }
        }
    }
    return stores;
}

Status StorageEngine::get_all_data_dir_info(std::vector<DataDirInfo>* data_dir_infos,
                                            bool need_update) {
    Status res = Status::OK();
    data_dir_infos->clear();

    MonotonicStopWatch timer;
    timer.start();

    // 1. update available capacity of each data dir
    // get all root path info and construct a path map.
    // path -> DataDirInfo
    std::map<std::string, DataDirInfo> path_map;
    {
        std::lock_guard<std::mutex> l(_store_lock);
        for (auto& it : _store_map) {
            if (need_update) {
                RETURN_IF_ERROR(it.second->update_capacity());
            }
            path_map.emplace(it.first, it.second->get_dir_info());
        }
    }

    // 2. get total tablets' size of each data dir
    size_t tablet_count = 0;
    _tablet_manager->update_root_path_info(&path_map, &tablet_count);

    // 3. update metrics in DataDir
    for (auto& path : path_map) {
        std::lock_guard<std::mutex> l(_store_lock);
        auto data_dir = _store_map.find(path.first);
        DCHECK(data_dir != _store_map.end());
        data_dir->second->update_local_data_size(path.second.local_used_capacity);
        data_dir->second->update_remote_data_size(path.second.remote_used_capacity);
    }

    // add path info to data_dir_infos
    for (auto& entry : path_map) {
        data_dir_infos->emplace_back(entry.second);
    }

    timer.stop();
    LOG(INFO) << "get root path info cost: " << timer.elapsed_time() / 1000000
              << " ms. tablet counter: " << tablet_count;

    return res;
}

int64_t StorageEngine::get_file_or_directory_size(const std::string& file_path) {
    if (!std::filesystem::exists(file_path)) {
        return 0;
    }
    if (!std::filesystem::is_directory(file_path)) {
        return std::filesystem::file_size(file_path);
    }
    int64_t sum_size = 0;
    for (const auto& it : std::filesystem::directory_iterator(file_path)) {
        sum_size += get_file_or_directory_size(it.path());
    }
    return sum_size;
}

void StorageEngine::_start_disk_stat_monitor() {
    for (auto& it : _store_map) {
        it.second->health_check();
    }

    _update_storage_medium_type_count();

    _exit_if_too_many_disks_are_failed();
}

// TODO(lingbin): Should be in EnvPosix?
Status StorageEngine::_check_file_descriptor_number() {
    struct rlimit l;
    int ret = getrlimit(RLIMIT_NOFILE, &l);
    if (ret != 0) {
        LOG(WARNING) << "call getrlimit() failed. errno=" << strerror(errno)
                     << ", use default configuration instead.";
        return Status::OK();
    }
    if (getenv("SKIP_CHECK_ULIMIT") == nullptr) {
        LOG(INFO) << "will check 'ulimit' value.";
    } else if (std::string(getenv("SKIP_CHECK_ULIMIT")) == "true") {
        LOG(INFO) << "the 'ulimit' value check is skipped"
                  << ", the SKIP_CHECK_ULIMIT env value is " << getenv("SKIP_CHECK_ULIMIT");
        return Status::OK();
    } else {
        LOG(INFO) << "the SKIP_CHECK_ULIMIT env value is " << getenv("SKIP_CHECK_ULIMIT")
                  << ", will check ulimit value.";
    }
    if (l.rlim_cur < config::min_file_descriptor_number) {
        LOG(ERROR) << "File descriptor number is less than " << config::min_file_descriptor_number
                   << ". Please use (ulimit -n) to set a value equal or greater than "
                   << config::min_file_descriptor_number;
        return Status::Error<ErrorCode::EXCEEDED_LIMIT>(
                "file descriptors limit {} is small than {}", l.rlim_cur,
                config::min_file_descriptor_number);
    }
    return Status::OK();
}

Status StorageEngine::_check_all_root_path_cluster_id() {
    int32_t cluster_id = -1;
    for (auto& it : _store_map) {
        int32_t tmp_cluster_id = it.second->cluster_id();
        if (it.second->cluster_id_incomplete()) {
            _is_all_cluster_id_exist = false;
        } else if (tmp_cluster_id == cluster_id) {
            // both have right cluster id, do nothing
        } else if (cluster_id == -1) {
            cluster_id = tmp_cluster_id;
        } else {
            RETURN_NOT_OK_STATUS_WITH_WARN(
                    Status::Corruption("multiple cluster ids is not equal. one={}, other={}",
                                       cluster_id, tmp_cluster_id),
                    "cluster id not equal");
        }
    }

    // judge and get effective cluster id
    RETURN_IF_ERROR(_judge_and_update_effective_cluster_id(cluster_id));

    // write cluster id into cluster_id_path if get effective cluster id success
    if (_effective_cluster_id != -1 && !_is_all_cluster_id_exist) {
        RETURN_IF_ERROR(set_cluster_id(_effective_cluster_id));
    }

    return Status::OK();
}

Status StorageEngine::set_cluster_id(int32_t cluster_id) {
    std::lock_guard<std::mutex> l(_store_lock);
    for (auto& it : _store_map) {
        RETURN_IF_ERROR(it.second->set_cluster_id(cluster_id));
    }
    _effective_cluster_id = cluster_id;
    _is_all_cluster_id_exist = true;
    return Status::OK();
}

int StorageEngine::_get_and_set_next_disk_index(int64_t partition_id,
                                                TStorageMedium::type storage_medium) {
    auto key = CreateTabletRRIdxCache::get_key(partition_id, storage_medium);
    int curr_index = _create_tablet_idx_lru_cache->get_index(key);
    // -1, lru can't find key
    if (curr_index == -1) {
        curr_index = std::max(0, _last_use_index[storage_medium] + 1);
    }
    _last_use_index[storage_medium] = curr_index;
    _create_tablet_idx_lru_cache->set_index(key, std::max(0, curr_index + 1));
    return curr_index;
}

void StorageEngine::_get_candidate_stores(TStorageMedium::type storage_medium,
                                          std::vector<DirInfo>& dir_infos) {
    std::vector<double> usages;
    for (auto& it : _store_map) {
        DataDir* data_dir = it.second.get();
        if (data_dir->is_used()) {
            if ((_available_storage_medium_type_count == 1 ||
                 data_dir->storage_medium() == storage_medium) &&
                !data_dir->reach_capacity_limit(0)) {
                double usage = data_dir->get_usage(0);
                DirInfo dir_info;
                dir_info.data_dir = data_dir;
                dir_info.usage = usage;
                dir_info.available_level = 0;
                usages.push_back(usage);
                dir_infos.push_back(dir_info);
            }
        }
    }

    if (dir_infos.size() <= 1) {
        return;
    }

    std::sort(usages.begin(), usages.end());
    if (usages.back() < 0.7) {
        return;
    }

    std::vector<double> level_min_usages;
    level_min_usages.push_back(usages[0]);
    for (auto usage : usages) {
        // usage < 0.7 consider as one level, give a small skew
        if (usage < 0.7 - (config::high_disk_avail_level_diff_usages / 2.0)) {
            continue;
        }

        // at high usages,  default 15% is one level
        // for example: there disk usages are:   0.66,  0.72,  0.83
        // then level_min_usages = [0.66, 0.83], divide disks into 2 levels:  [0.66, 0.72], [0.83]
        if (usage >= level_min_usages.back() + config::high_disk_avail_level_diff_usages) {
            level_min_usages.push_back(usage);
        }
    }
    for (auto& dir_info : dir_infos) {
        double usage = dir_info.usage;
        for (size_t i = 1; i < level_min_usages.size() && usage >= level_min_usages[i]; i++) {
            dir_info.available_level++;
        }

        // when usage is too high, no matter consider balance now,
        // make it a higher level.
        // for example, two disks and usages are: 0.85 and 0.92, then let tablets fall on the first disk.
        // by default, storage_flood_stage_usage_percent = 90
        if (usage > config::storage_flood_stage_usage_percent / 100.0) {
            dir_info.available_level++;
        }
    }
}

std::vector<DataDir*> StorageEngine::get_stores_for_create_tablet(
        int64_t partition_id, TStorageMedium::type storage_medium) {
    std::vector<DirInfo> dir_infos;
    int curr_index = 0;
    std::vector<DataDir*> stores;
    {
        std::lock_guard<std::mutex> l(_store_lock);
        curr_index = _get_and_set_next_disk_index(partition_id, storage_medium);
        _get_candidate_stores(storage_medium, dir_infos);
    }

    std::sort(dir_infos.begin(), dir_infos.end());
    get_round_robin_stores(curr_index, dir_infos, stores);

    return stores;
}

// maintain in stores LOW,MID,HIGH level round robin
void get_round_robin_stores(int64_t curr_index, const std::vector<DirInfo>& dir_infos,
                            std::vector<DataDir*>& stores) {
    for (size_t i = 0; i < dir_infos.size();) {
        size_t end = i + 1;
        while (end < dir_infos.size() &&
               dir_infos[i].available_level == dir_infos[end].available_level) {
            end++;
        }
        // data dirs [i, end) have the same tablet size, round robin range [i, end)
        size_t count = end - i;
        for (size_t k = 0; k < count; k++) {
            size_t index = i + ((k + curr_index) % count);
            stores.push_back(dir_infos[index].data_dir);
        }
        i = end;
    }
}

DataDir* StorageEngine::get_store(const std::string& path) {
    // _store_map is unchanged, no need to lock
    auto it = _store_map.find(path);
    if (it == _store_map.end()) {
        return nullptr;
    }
    return it->second.get();
}

static bool too_many_disks_are_failed(uint32_t unused_num, uint32_t total_num) {
    return ((total_num == 0) ||
            (unused_num * 100 / total_num > config::max_percentage_of_error_disk));
}

void StorageEngine::_exit_if_too_many_disks_are_failed() {
    uint32_t unused_root_path_num = 0;
    uint32_t total_root_path_num = 0;

    {
        // TODO(yingchun): _store_map is only updated in main and ~StorageEngine, maybe we can remove it?
        std::lock_guard<std::mutex> l(_store_lock);
        if (_store_map.empty()) {
            return;
        }

        for (auto& it : _store_map) {
            ++total_root_path_num;
            if (it.second->is_used()) {
                continue;
            }
            ++unused_root_path_num;
        }
    }

    if (too_many_disks_are_failed(unused_root_path_num, total_root_path_num)) {
        LOG(FATAL) << "meet too many error disks, process exit. "
                   << "max_ratio_allowed=" << config::max_percentage_of_error_disk << "%"
                   << ", error_disk_count=" << unused_root_path_num
                   << ", total_disk_count=" << total_root_path_num;
        exit(0);
    }
}

void StorageEngine::stop() {
    if (_stopped) {
        LOG(WARNING) << "Storage engine is stopped twice.";
        return;
    }
    // trigger the waiting threads
    notify_listeners();

    {
        std::lock_guard<std::mutex> l(_store_lock);
        for (auto& store_pair : _store_map) {
            store_pair.second->stop_bg_worker();
        }
    }

    _stop_background_threads_latch.count_down();
#define THREAD_JOIN(thread) \
    if (thread) {           \
        thread->join();     \
    }

    THREAD_JOIN(_compaction_tasks_producer_thread);
    THREAD_JOIN(_binlog_compaction_tasks_producer_thread);
    THREAD_JOIN(_unused_rowset_monitor_thread);
    THREAD_JOIN(_garbage_sweeper_thread);
    THREAD_JOIN(_disk_stat_monitor_thread);
    THREAD_JOIN(_cache_clean_thread);
    THREAD_JOIN(_tablet_checkpoint_tasks_producer_thread);
    THREAD_JOIN(_async_publish_thread);
    THREAD_JOIN(_cold_data_compaction_producer_thread);
    THREAD_JOIN(_cooldown_tasks_producer_thread);
    THREAD_JOIN(_check_delete_bitmap_score_thread);
#undef THREAD_JOIN

#define THREADS_JOIN(threads)            \
    for (const auto& thread : threads) { \
        if (thread) {                    \
            thread->join();              \
        }                                \
    }

    THREADS_JOIN(_path_gc_threads);
#undef THREADS_JOIN

    if (_base_compaction_thread_pool) {
        _base_compaction_thread_pool->shutdown();
    }
    if (_cumu_compaction_thread_pool) {
        _cumu_compaction_thread_pool->shutdown();
    }
    if (_binlog_compaction_thread_pool) {
        _binlog_compaction_thread_pool->shutdown();
    }

    if (_seg_compaction_thread_pool) {
        _seg_compaction_thread_pool->shutdown();
    }
    if (_tablet_meta_checkpoint_thread_pool) {
        _tablet_meta_checkpoint_thread_pool->shutdown();
    }
    if (_cold_data_compaction_thread_pool) {
        _cold_data_compaction_thread_pool->shutdown();
    }

    if (_cooldown_thread_pool) {
        _cooldown_thread_pool->shutdown();
    }

    _adaptive_thread_controller.stop();
    _memtable_flush_executor.reset(nullptr);
    _calc_delete_bitmap_executor.reset(nullptr);
    _calc_delete_bitmap_executor_for_load.reset();

    _stopped = true;
    LOG(INFO) << "Storage engine is stopped.";
}

void StorageEngine::clear_transaction_task(const TTransactionId transaction_id) {
    // clear transaction task may not contains partitions ids, we should get partition id from txn manager.
    std::vector<int64_t> partition_ids;
    _txn_manager->get_partition_ids(transaction_id, &partition_ids);
    clear_transaction_task(transaction_id, partition_ids);
}

void StorageEngine::clear_transaction_task(const TTransactionId transaction_id,
                                           const std::vector<TPartitionId>& partition_ids) {
    LOG(INFO) << "begin to clear transaction task. transaction_id=" << transaction_id;

    for (const TPartitionId& partition_id : partition_ids) {
        std::map<TabletInfo, RowsetSharedPtr> tablet_infos;
        _txn_manager->get_txn_related_tablets(transaction_id, partition_id, &tablet_infos);

        // each tablet
        for (auto& tablet_info : tablet_infos) {
            // should use tablet uid to ensure clean txn correctly
            TabletSharedPtr tablet = _tablet_manager->get_tablet(tablet_info.first.tablet_id,
                                                                 tablet_info.first.tablet_uid);
            // The tablet may be dropped or altered, leave a INFO log and go on process other tablet
            if (tablet == nullptr) {
                LOG(INFO) << "tablet is no longer exist. tablet_id=" << tablet_info.first.tablet_id
                          << ", tablet_uid=" << tablet_info.first.tablet_uid;
                continue;
            }
            Status s = _txn_manager->delete_txn(partition_id, tablet, transaction_id);
            if (!s.ok()) {
                LOG(WARNING) << "failed to clear transaction. txn_id=" << transaction_id
                             << ", partition_id=" << partition_id
                             << ", tablet_id=" << tablet_info.first.tablet_id
                             << ", status=" << s.to_string();
            }
        }
    }
    LOG(INFO) << "finish to clear transaction task. transaction_id=" << transaction_id;
}

Status StorageEngine::start_trash_sweep(double* usage, bool ignore_guard) {
    Status res = Status::OK();

    std::unique_lock<std::mutex> l(_trash_sweep_lock, std::defer_lock);
    if (!l.try_lock()) {
        LOG(INFO) << "trash and snapshot sweep is running.";
        if (ignore_guard) {
            _need_clean_trash.store(true, std::memory_order_relaxed);
        }
        return res;
    }

    LOG(INFO) << "start trash and snapshot sweep. is_clean=" << ignore_guard;

    const int32_t snapshot_expire = config::snapshot_expire_time_sec;
    const int32_t trash_expire = config::trash_file_expire_time_sec;
    // the guard space should be lower than storage_flood_stage_usage_percent,
    // so here we multiply 0.9
    // if ignore_guard is true, set guard_space to 0.
    const double guard_space =
            ignore_guard ? 0 : config::storage_flood_stage_usage_percent / 100.0 * 0.9;
    std::vector<DataDirInfo> data_dir_infos;
    RETURN_NOT_OK_STATUS_WITH_WARN(get_all_data_dir_info(&data_dir_infos, false),
                                   "failed to get root path stat info when sweep trash.")
    std::sort(data_dir_infos.begin(), data_dir_infos.end(), DataDirInfoLessAvailability());

    time_t now = time(nullptr); //获取UTC时间
    tm local_tm_now;
    local_tm_now.tm_isdst = 0;
    if (localtime_r(&now, &local_tm_now) == nullptr) {
        return Status::Error<OS_ERROR>("fail to localtime_r time. time={}", now);
    }
    const time_t local_now = mktime(&local_tm_now); //得到当地日历时间

    double tmp_usage = 0.0;
    for (DataDirInfo& info : data_dir_infos) {
        LOG(INFO) << "Start to sweep path " << info.path;
        if (!info.is_used) {
            continue;
        }

        double curr_usage =
                (double)(info.disk_capacity - info.available) / (double)info.disk_capacity;
        tmp_usage = std::max(tmp_usage, curr_usage);

        Status curr_res = Status::OK();
        auto snapshot_path = fmt::format("{}/{}", info.path, SNAPSHOT_PREFIX);
        curr_res = _do_sweep(snapshot_path, local_now, snapshot_expire);
        if (!curr_res.ok()) {
            LOG(WARNING) << "failed to sweep snapshot. path=" << snapshot_path
                         << ", err_code=" << curr_res;
            res = curr_res;
        }

        auto trash_path = fmt::format("{}/{}", info.path, TRASH_PREFIX);
        curr_res = _do_sweep(trash_path, local_now, curr_usage > guard_space ? 0 : trash_expire);
        if (!curr_res.ok()) {
            LOG(WARNING) << "failed to sweep trash. path=" << trash_path
                         << ", err_code=" << curr_res;
            res = curr_res;
        }
    }

    if (usage != nullptr) {
        *usage = tmp_usage; // update usage
    }

    // clear expire incremental rowset, move deleted tablet to trash
    RETURN_IF_ERROR(_tablet_manager->start_trash_sweep());

    // clean rubbish transactions
    _clean_unused_txns();

    // clean unused rowset metas in OlapMeta
    _clean_unused_rowset_metas();

    // clean unused binlog metas in OlapMeta
    _clean_unused_binlog_metas();

    // cleand unused delete bitmap for deleted tablet
    _clean_unused_delete_bitmap();

    // cleand unused pending publish info for deleted tablet
    _clean_unused_pending_publish_info();

    // clean unused partial update info for finished txns
    _clean_unused_partial_update_info();

    // clean unused rowsets in remote storage backends
    for (auto data_dir : get_stores()) {
        data_dir->perform_remote_rowset_gc();
        data_dir->perform_remote_tablet_gc();
        data_dir->update_trash_capacity();
    }

    return res;
}

void StorageEngine::_clean_unused_rowset_metas() {
    std::vector<RowsetMetaSharedPtr> invalid_rowset_metas;
    auto clean_rowset_func = [this, &invalid_rowset_metas](TabletUid tablet_uid, RowsetId rowset_id,
                                                           std::string_view meta_str) -> bool {
        // return false will break meta iterator, return true to skip this error
        RowsetMetaSharedPtr rowset_meta(new RowsetMeta());
        bool parsed = rowset_meta->init(meta_str);
        if (!parsed) {
            LOG(WARNING) << "parse rowset meta string failed for rowset_id:" << rowset_id;
            rowset_meta->set_rowset_id(rowset_id);
            rowset_meta->set_tablet_uid(tablet_uid);
            invalid_rowset_metas.push_back(rowset_meta);
            return true;
        }
        if (rowset_meta->tablet_uid() != tablet_uid) {
            LOG(WARNING) << "tablet uid is not equal, skip the rowset"
                         << ", rowset_id=" << rowset_meta->rowset_id()
                         << ", in_put_tablet_uid=" << tablet_uid
                         << ", tablet_uid in rowset meta=" << rowset_meta->tablet_uid();
            invalid_rowset_metas.push_back(rowset_meta);
            return true;
        }

        TabletSharedPtr tablet = _tablet_manager->get_tablet(rowset_meta->tablet_id());
        if (tablet == nullptr) {
            // tablet may be dropped
            // TODO(cmy): this is better to be a VLOG, because drop table is a very common case.
            // leave it as INFO log for observation. Maybe change it in future.
            LOG(INFO) << "failed to find tablet " << rowset_meta->tablet_id()
                      << " for rowset: " << rowset_meta->rowset_id() << ", tablet may be dropped";
            invalid_rowset_metas.push_back(rowset_meta);
            return true;
        }
        if (tablet->tablet_uid() != rowset_meta->tablet_uid()) {
            // In this case, we get the tablet using the tablet id recorded in the rowset meta.
            // but the uid in the tablet is different from the one recorded in the rowset meta.
            // How this happened:
            // Replica1 of Tablet A exists on BE1. Because of the clone task, a new replica2 is createed on BE2,
            // and then replica1 deleted from BE1. After some time, we created replica again on BE1,
            // which will creates a new tablet with the same id but a different uid.
            // And in the historical version, when we deleted the replica, we did not delete the corresponding rowset meta,
            // thus causing the original rowset meta to remain(with same tablet id but different uid).
            LOG(WARNING) << "rowset's tablet uid " << rowset_meta->tablet_uid()
                         << " does not equal to tablet uid: " << tablet->tablet_uid();
            invalid_rowset_metas.push_back(rowset_meta);
            return true;
        }
        if (rowset_meta->rowset_state() == RowsetStatePB::VISIBLE &&
            (!tablet->rowset_meta_is_useful(rowset_meta)) &&
            !check_rowset_id_in_unused_rowsets(rowset_id)) {
            LOG(INFO) << "rowset meta is not used any more, remove it. rowset_id="
                      << rowset_meta->rowset_id();
            invalid_rowset_metas.push_back(rowset_meta);
        }
        return true;
    };
    std::vector<std::pair<RowsetId, RowsetMetaSharedPtr>> invalid_row_binlog_metas;
    auto clean_row_binlog_rowsets = [this, &invalid_row_binlog_metas](
                                            const TabletUid& tablet_uid, RowsetId rowset_id,
                                            RowsetId row_binlog_rowset_id,
                                            const std::string& meta_str) -> bool {
        // return false will break meta iterator, return true to skip this error
        RowsetMetaSharedPtr row_binlog_rowset_meta(new RowsetMeta());
        bool parsed = row_binlog_rowset_meta->init(meta_str);
        if (!parsed) {
            LOG(WARNING) << "parse binlog<row> meta string failed for rowset_id:"
                         << row_binlog_rowset_id;
            row_binlog_rowset_meta->set_rowset_id(row_binlog_rowset_id);
            row_binlog_rowset_meta->set_tablet_uid(tablet_uid);
            invalid_row_binlog_metas.emplace_back(rowset_id, row_binlog_rowset_meta);
            return true;
        }
        TabletSharedPtr tablet = _tablet_manager->get_tablet(row_binlog_rowset_meta->tablet_id());
        if (tablet == nullptr) {
            LOG(INFO) << "failed to find tablet " << row_binlog_rowset_meta->tablet_id()
                      << " for binlog<row>: " << row_binlog_rowset_meta->rowset_id()
                      << ", tablet may be dropped";
            invalid_row_binlog_metas.emplace_back(rowset_id, row_binlog_rowset_meta);
            return true;
        }
        if (tablet->tablet_uid() != row_binlog_rowset_meta->tablet_uid()) {
            LOG(WARNING) << "binlog<row> meta's tablet uid " << row_binlog_rowset_meta->tablet_uid()
                         << " does not equal to tablet uid: " << tablet->tablet_uid();
            invalid_row_binlog_metas.emplace_back(rowset_id, row_binlog_rowset_meta);
            return true;
        }
        if (row_binlog_rowset_meta->rowset_state() == RowsetStatePB::VISIBLE &&
            !tablet->rowset_meta_is_useful(row_binlog_rowset_meta) &&
            !check_rowset_id_in_unused_rowsets(rowset_id)) {
            LOG(INFO) << "binlog<row> meta is not used any more, remove it. rowset_id="
                      << row_binlog_rowset_meta->rowset_id();
            invalid_row_binlog_metas.emplace_back(rowset_id, row_binlog_rowset_meta);
        }
        return true;
    };
    auto data_dirs = get_stores();
    for (auto data_dir : data_dirs) {
        static_cast<void>(
                RowsetMetaManager::traverse_rowset_metas(data_dir->get_meta(), clean_rowset_func));
        // 1. delete delete_bitmap
        std::set<int64_t> tablets_to_save_meta;
        for (auto& rowset_meta : invalid_rowset_metas) {
            TabletSharedPtr tablet = _tablet_manager->get_tablet(rowset_meta->tablet_id());
            if (tablet && tablet->tablet_meta()->enable_unique_key_merge_on_write()) {
                tablet->tablet_meta()->remove_rowset_delete_bitmap(rowset_meta->rowset_id(),
                                                                   rowset_meta->version());
                tablets_to_save_meta.emplace(tablet->tablet_id());
            }
        }
        for (const auto& tablet_id : tablets_to_save_meta) {
            auto tablet = _tablet_manager->get_tablet(tablet_id);
            if (tablet) {
                std::shared_lock rlock(tablet->get_header_lock());
                tablet->save_meta();
            }
        }
        // 2. delete rowset meta
        for (auto& rowset_meta : invalid_rowset_metas) {
            static_cast<void>(RowsetMetaManager::remove(
                    data_dir->get_meta(), rowset_meta->tablet_uid(), rowset_meta->rowset_id()));
        }
        LOG(INFO) << "remove " << invalid_rowset_metas.size()
                  << " invalid rowset meta from dir: " << data_dir->path();

        static_cast<void>(RowsetMetaManager::traverse_row_binlog_metas(data_dir->get_meta(),
                                                                       clean_row_binlog_rowsets));
        for (auto& rs_id_to_meta : invalid_row_binlog_metas) {
            static_cast<void>(RowsetMetaManager::remove_row_binlog(
                    data_dir->get_meta(), rs_id_to_meta.second->tablet_uid(), rs_id_to_meta.first,
                    rs_id_to_meta.second->rowset_id()));
        }
        LOG(INFO) << "remove " << invalid_row_binlog_metas.size()
                  << " invalid binlog<row> meta from dir: " << data_dir->path();
        invalid_row_binlog_metas.clear();
        invalid_rowset_metas.clear();
    }
}

void StorageEngine::_clean_unused_binlog_metas() {
    std::vector<std::string> unused_binlog_key_suffixes;
    auto unused_binlog_collector = [this, &unused_binlog_key_suffixes](std::string_view key,
                                                                       std::string_view value,
                                                                       bool need_check) -> bool {
        if (need_check) {
            BinlogMetaEntryPB binlog_meta_pb;
            if (UNLIKELY(!binlog_meta_pb.ParseFromArray(value.data(),
                                                        cast_set<int>(value.size())))) {
                LOG(WARNING) << "parse rowset meta string failed for binlog meta key: " << key;
            } else if (_tablet_manager->get_tablet(binlog_meta_pb.tablet_id()) == nullptr) {
                LOG(INFO) << "failed to find tablet " << binlog_meta_pb.tablet_id()
                          << " for binlog rowset: " << binlog_meta_pb.rowset_id()
                          << ", tablet may be dropped";
            } else {
                return false;
            }
        }

        unused_binlog_key_suffixes.emplace_back(key.substr(kBinlogMetaPrefix.size()));
        return true;
    };
    auto data_dirs = get_stores();
    for (auto data_dir : data_dirs) {
        static_cast<void>(RowsetMetaManager::traverse_binlog_metas(data_dir->get_meta(),
                                                                   unused_binlog_collector));
        for (const auto& suffix : unused_binlog_key_suffixes) {
            static_cast<void>(RowsetMetaManager::remove_binlog(data_dir->get_meta(), suffix));
        }
        LOG(INFO) << "remove " << unused_binlog_key_suffixes.size()
                  << " invalid binlog meta from dir: " << data_dir->path();
        unused_binlog_key_suffixes.clear();
    }
}

void StorageEngine::_clean_unused_delete_bitmap() {
    std::unordered_set<int64_t> removed_tablets;
    auto clean_delete_bitmap_func = [this, &removed_tablets](int64_t tablet_id, int64_t version,
                                                             std::string_view val) -> bool {
        TabletSharedPtr tablet = _tablet_manager->get_tablet(tablet_id);
        if (tablet == nullptr) {
            if (removed_tablets.insert(tablet_id).second) {
                LOG(INFO) << "clean ununsed delete bitmap for deleted tablet, tablet_id: "
                          << tablet_id;
            }
        }
        return true;
    };
    auto data_dirs = get_stores();
    for (auto data_dir : data_dirs) {
        static_cast<void>(TabletMetaManager::traverse_delete_bitmap(data_dir->get_meta(),
                                                                    clean_delete_bitmap_func));
        for (auto id : removed_tablets) {
            static_cast<void>(
                    TabletMetaManager::remove_old_version_delete_bitmap(data_dir, id, INT64_MAX));
        }
        LOG(INFO) << "removed invalid delete bitmap from dir: " << data_dir->path()
                  << ", deleted tablets size: " << removed_tablets.size();
        removed_tablets.clear();
    }
}

void StorageEngine::_clean_unused_pending_publish_info() {
    std::vector<std::pair<int64_t, int64_t>> removed_infos;
    auto clean_pending_publish_info_func = [this, &removed_infos](int64_t tablet_id,
                                                                  int64_t publish_version,
                                                                  std::string_view info) -> bool {
        TabletSharedPtr tablet = _tablet_manager->get_tablet(tablet_id);
        if (tablet == nullptr) {
            removed_infos.emplace_back(tablet_id, publish_version);
        }
        return true;
    };
    auto data_dirs = get_stores();
    for (auto data_dir : data_dirs) {
        static_cast<void>(TabletMetaManager::traverse_pending_publish(
                data_dir->get_meta(), clean_pending_publish_info_func));
        for (auto& [tablet_id, publish_version] : removed_infos) {
            static_cast<void>(TabletMetaManager::remove_pending_publish_info(data_dir, tablet_id,
                                                                             publish_version));
        }
        LOG(INFO) << "removed invalid pending publish info from dir: " << data_dir->path()
                  << ", deleted pending publish info size: " << removed_infos.size();
        removed_infos.clear();
    }
}

void StorageEngine::_clean_unused_partial_update_info() {
    std::vector<std::tuple<int64_t, int64_t, int64_t>> remove_infos;
    auto unused_partial_update_info_collector =
            [this, &remove_infos](int64_t tablet_id, int64_t partition_id, int64_t txn_id,
                                  std::string_view value) -> bool {
        TabletSharedPtr tablet = _tablet_manager->get_tablet(tablet_id);
        if (tablet == nullptr) {
            remove_infos.emplace_back(tablet_id, partition_id, txn_id);
            return true;
        }
        TxnState txn_state =
                _txn_manager->get_txn_state(partition_id, txn_id, tablet_id, tablet->tablet_uid());
        if (txn_state == TxnState::NOT_FOUND || txn_state == TxnState::ABORTED ||
            txn_state == TxnState::DELETED) {
            remove_infos.emplace_back(tablet_id, partition_id, txn_id);
            return true;
        }
        return true;
    };
    auto data_dirs = get_stores();
    for (auto* data_dir : data_dirs) {
        static_cast<void>(RowsetMetaManager::traverse_partial_update_info(
                data_dir->get_meta(), unused_partial_update_info_collector));
        static_cast<void>(
                RowsetMetaManager::remove_partial_update_infos(data_dir->get_meta(), remove_infos));
    }
}

void StorageEngine::gc_binlogs(const std::unordered_map<int64_t, int64_t>& gc_tablet_infos) {
    for (auto [tablet_id, version] : gc_tablet_infos) {
        LOG(INFO) << fmt::format("start to gc binlogs for tablet_id: {}, version: {}", tablet_id,
                                 version);

        TabletSharedPtr tablet = _tablet_manager->get_tablet(tablet_id);
        if (tablet == nullptr) {
            LOG(WARNING) << fmt::format("tablet_id: {} not found", tablet_id);
            continue;
        }
        tablet->gc_binlogs(version);
    }
}

void StorageEngine::_clean_unused_txns() {
    std::set<TabletInfo> tablet_infos;
    _txn_manager->get_all_related_tablets(&tablet_infos);
    for (auto& tablet_info : tablet_infos) {
        TabletSharedPtr tablet =
                _tablet_manager->get_tablet(tablet_info.tablet_id, tablet_info.tablet_uid, true);
        if (tablet == nullptr) {
            // TODO(ygl) :  should check if tablet still in meta, it's a improvement
            // case 1: tablet still in meta, just remove from memory
            // case 2: tablet not in meta store, remove rowset from meta
            // currently just remove them from memory
            // nullptr to indicate not remove them from meta store
            _txn_manager->force_rollback_tablet_related_txns(nullptr, tablet_info.tablet_id,
                                                             tablet_info.tablet_uid);
        }
    }
}

Status StorageEngine::_do_sweep(const std::string& scan_root, const time_t& local_now,
                                const int32_t expire) {
    Status res = Status::OK();
    bool exists = true;
    RETURN_IF_ERROR(io::global_local_filesystem()->exists(scan_root, &exists));
    if (!exists) {
        // dir not existed. no need to sweep trash.
        return res;
    }

    int curr_sweep_batch_size = 0;
    try {
        // Sort pathes by name, that is by delete time.
        std::vector<path> sorted_pathes;
        std::copy(directory_iterator(scan_root), directory_iterator(),
                  std::back_inserter(sorted_pathes));
        std::sort(sorted_pathes.begin(), sorted_pathes.end());
        for (const auto& sorted_path : sorted_pathes) {
            string dir_name = sorted_path.filename().string();
            string str_time = dir_name.substr(0, dir_name.find('.'));
            tm local_tm_create;
            local_tm_create.tm_isdst = 0;
            if (strptime(str_time.c_str(), "%Y%m%d%H%M%S", &local_tm_create) == nullptr) {
                res = Status::Error<OS_ERROR>("fail to strptime time. time={}", str_time);
                continue;
            }

            int32_t actual_expire = expire;
            // try get timeout in dir name, the old snapshot dir does not contain timeout
            // eg: 20190818221123.3.86400, the 86400 is timeout, in second
            size_t pos = dir_name.find('.', str_time.size() + 1);
            if (pos != string::npos) {
                actual_expire = std::stoi(dir_name.substr(pos + 1));
            }
            VLOG_TRACE << "get actual expire time " << actual_expire << " of dir: " << dir_name;

            string path_name = sorted_path.string();
            if (difftime(local_now, mktime(&local_tm_create)) >= actual_expire) {
                res = io::global_local_filesystem()->delete_directory(path_name);
                LOG(INFO) << "do sweep delete directory " << path_name << " local_now " << local_now
                          << "actual_expire " << actual_expire << " res " << res;
                if (!res.ok()) {
                    continue;
                }

                curr_sweep_batch_size++;
                if (config::garbage_sweep_batch_size > 0 &&
                    curr_sweep_batch_size >= config::garbage_sweep_batch_size) {
                    curr_sweep_batch_size = 0;
                    std::this_thread::sleep_for(std::chrono::milliseconds(1));
                }
            } else {
                // Because files are ordered by filename, i.e. by create time, so all the left files are not expired.
                break;
            }
        }
    } catch (...) {
        res = Status::Error<IO_ERROR>("Exception occur when scan directory. path_desc={}",
                                      scan_root);
    }

    return res;
}

// invalid rowset type config will return ALPHA_ROWSET for system to run smoothly
void StorageEngine::_parse_default_rowset_type() {
    std::string default_rowset_type_config = config::default_rowset_type;
    boost::to_upper(default_rowset_type_config);
    if (default_rowset_type_config == "BETA") {
        _default_rowset_type = BETA_ROWSET;
    } else if (default_rowset_type_config == "ALPHA") {
        _default_rowset_type = ALPHA_ROWSET;
        LOG(WARNING) << "default_rowset_type in be.conf should be set to beta, alpha is not "
                        "supported any more";
    } else {
        LOG(FATAL) << "unknown value " << default_rowset_type_config
                   << " in default_rowset_type in be.conf";
    }
}

void StorageEngine::start_delete_unused_rowset() {
    DBUG_EXECUTE_IF("StorageEngine::start_delete_unused_rowset.block", DBUG_BLOCK);
    LOG(INFO) << "start to delete unused rowset, size: " << _unused_rowsets.size()
              << ", unused delete bitmap size: " << _unused_delete_bitmap.size();
    std::vector<RowsetSharedPtr> unused_rowsets_copy;
    unused_rowsets_copy.reserve(_unused_rowsets.size());
    auto due_to_use_count = 0;
    auto due_to_not_delete_file = 0;
    auto due_to_delayed_expired_ts = 0;
    std::set<int64_t> tablets_to_save_meta;
    {
        std::lock_guard<std::mutex> lock(_gc_mutex);
        for (auto it = _unused_rowsets.begin(); it != _unused_rowsets.end();) {
            auto&& rs = it->second;
            if (rs.use_count() == 1 && rs->need_delete_file()) {
                // remote rowset data will be reclaimed by `remove_unused_remote_files`
                if (rs->is_local()) {
                    unused_rowsets_copy.push_back(std::move(rs));
                }
                it = _unused_rowsets.erase(it);
            } else {
                if (rs.use_count() != 1) {
                    ++due_to_use_count;
                } else if (!rs->need_delete_file()) {
                    ++due_to_not_delete_file;
                } else {
                    ++due_to_delayed_expired_ts;
                }
                ++it;
            }
        }
        // check remove delete bitmaps
        for (auto it = _unused_delete_bitmap.begin(); it != _unused_delete_bitmap.end();) {
            auto tablet_id = std::get<0>(*it);
            auto tablet = _tablet_manager->get_tablet(tablet_id);
            if (tablet == nullptr) {
                it = _unused_delete_bitmap.erase(it);
                continue;
            }
            auto& rowset_ids = std::get<1>(*it);
            auto& key_ranges = std::get<2>(*it);
            bool find_unused_rowset = false;
            for (const auto& rowset_id : rowset_ids) {
                if (_unused_rowsets.find(rowset_id) != _unused_rowsets.end()) {
                    VLOG_DEBUG << "can not remove pre rowset delete bitmap because rowset is in use"
                               << ", tablet_id=" << tablet_id
                               << ", rowset_id=" << rowset_id.to_string();
                    find_unused_rowset = true;
                    break;
                }
            }
            if (find_unused_rowset) {
                ++it;
                continue;
            }
            tablet->tablet_meta()->delete_bitmap().remove(key_ranges);
            tablets_to_save_meta.emplace(tablet_id);
            it = _unused_delete_bitmap.erase(it);
        }
    }
    LOG(INFO) << "collected " << unused_rowsets_copy.size() << " unused rowsets to remove, skipped "
              << due_to_use_count << " rowsets due to use count > 1, skipped "
              << due_to_not_delete_file << " rowsets due to don't need to delete file, skipped "
              << due_to_delayed_expired_ts << " rowsets due to delayed expired timestamp. left "
              << _unused_delete_bitmap.size() << " unused delete bitmap.";
    for (auto&& rs : unused_rowsets_copy) {
        VLOG_NOTICE << "start to remove rowset:" << rs->rowset_id()
                    << ", version:" << rs->version();
        // delete delete_bitmap of unused rowsets
        if (auto tablet = _tablet_manager->get_tablet(rs->rowset_meta()->tablet_id());
            tablet && tablet->enable_unique_key_merge_on_write()) {
            tablet->tablet_meta()->remove_rowset_delete_bitmap(rs->rowset_id(), rs->version());
            tablets_to_save_meta.emplace(tablet->tablet_id());
        }
        Status status = rs->remove();
        unused_rowsets_counter << -1;
        VLOG_NOTICE << "remove rowset:" << rs->rowset_id() << " finished. status:" << status;
    }
    for (const auto& tablet_id : tablets_to_save_meta) {
        auto tablet = _tablet_manager->get_tablet(tablet_id);
        if (tablet) {
            std::shared_lock rlock(tablet->get_header_lock());
            tablet->save_meta();
        }
    }
    LOG(INFO) << "removed all collected unused rowsets";
}

void StorageEngine::add_unused_rowset(RowsetSharedPtr rowset) {
    if (rowset == nullptr) {
        return;
    }
    VLOG_NOTICE << "add unused rowset, rowset id:" << rowset->rowset_id()
                << ", version:" << rowset->version();
    std::lock_guard<std::mutex> lock(_gc_mutex);
    auto it = _unused_rowsets.find(rowset->rowset_id());
    if (it == _unused_rowsets.end()) {
        rowset->set_need_delete_file();
        rowset->close();
        _unused_rowsets[rowset->rowset_id()] = std::move(rowset);
        unused_rowsets_counter << 1;
    }
}

void StorageEngine::add_unused_delete_bitmap_key_ranges(int64_t tablet_id,
                                                        const std::vector<RowsetId>& rowsets,
                                                        const DeleteBitmapKeyRanges& key_ranges) {
    VLOG_NOTICE << "add unused delete bitmap key ranges, tablet id:" << tablet_id;
    std::lock_guard<std::mutex> lock(_gc_mutex);
    _unused_delete_bitmap.push_back(std::make_tuple(tablet_id, rowsets, key_ranges));
}

// TODO(zc): refactor this funciton
Status StorageEngine::create_tablet(const TCreateTabletReq& request, RuntimeProfile* profile) {
    // Get all available stores, use ref_root_path if the caller specified
    std::vector<DataDir*> stores;
    {
        SCOPED_TIMER(ADD_TIMER(profile, "GetStores"));
        stores = get_stores_for_create_tablet(request.partition_id, request.storage_medium);
    }
    if (stores.empty()) {
        return Status::Error<CE_CMD_PARAMS_ERROR>(
                "there is no available disk that can be used to create tablet.");
    }
    return _tablet_manager->create_tablet(request, stores, profile);
}

Result<BaseTabletSPtr> StorageEngine::get_tablet(int64_t tablet_id, SyncRowsetStats* sync_stats,
                                                 bool force_use_only_cached, bool cache_on_miss) {
    BaseTabletSPtr tablet;
    std::string err;
    tablet = _tablet_manager->get_tablet(tablet_id, true, &err);
    if (tablet == nullptr) {
        return unexpected(
                Status::InternalError("failed to get tablet: {}, reason: {}", tablet_id, err));
    }
    return tablet;
}

Status StorageEngine::get_tablet_meta(int64_t tablet_id, TabletMetaSharedPtr* tablet_meta,
                                      bool force_use_only_cached) {
    if (tablet_meta == nullptr) {
        return Status::InvalidArgument("tablet_meta output is null");
    }

    auto res = get_tablet(tablet_id, nullptr, force_use_only_cached, true);
    if (!res.has_value()) {
        return res.error();
    }

    *tablet_meta = res.value()->tablet_meta();
    return Status::OK();
}

Status StorageEngine::obtain_shard_path(TStorageMedium::type storage_medium, int64_t path_hash,
                                        std::string* shard_path, DataDir** store,
                                        int64_t partition_id) {
    LOG(INFO) << "begin to process obtain root path. storage_medium=" << storage_medium;

    if (shard_path == nullptr) {
        return Status::Error<CE_CMD_PARAMS_ERROR>(
                "invalid output parameter which is null pointer.");
    }

    auto stores = get_stores_for_create_tablet(partition_id, storage_medium);
    if (stores.empty()) {
        return Status::Error<NO_AVAILABLE_ROOT_PATH>(
                "no available disk can be used to create tablet.");
    }

    *store = nullptr;
    if (path_hash != -1) {
        for (auto data_dir : stores) {
            if (data_dir->path_hash() == path_hash) {
                *store = data_dir;
                break;
            }
        }
    }
    if (*store == nullptr) {
        *store = stores[0];
    }

    uint64_t shard = (*store)->get_shard();

    std::stringstream root_path_stream;
    root_path_stream << (*store)->path() << "/" << DATA_PREFIX << "/" << shard;
    *shard_path = root_path_stream.str();

    LOG(INFO) << "success to process obtain root path. path=" << *shard_path;
    return Status::OK();
}

Status StorageEngine::load_header(const string& shard_path, const TCloneReq& request,
                                  bool restore) {
    LOG(INFO) << "begin to process load headers."
              << "tablet_id=" << request.tablet_id << ", schema_hash=" << request.schema_hash;
    Status res = Status::OK();

    DataDir* store = nullptr;
    {
        // TODO(zc)
        try {
            auto store_path =
                    std::filesystem::path(shard_path).parent_path().parent_path().string();
            store = get_store(store_path);
            if (store == nullptr) {
                return Status::Error<INVALID_ROOT_PATH>("invalid shard path, path={}", shard_path);
            }
        } catch (...) {
            return Status::Error<INVALID_ROOT_PATH>("invalid shard path, path={}", shard_path);
        }
    }

    std::stringstream schema_hash_path_stream;
    schema_hash_path_stream << shard_path << "/" << request.tablet_id << "/" << request.schema_hash;
    // not surely, reload and restore tablet action call this api
    // reset tablet uid here

    string header_path = TabletMeta::construct_header_file_path(schema_hash_path_stream.str(),
                                                                request.tablet_id);
    res = _tablet_manager->load_tablet_from_dir(store, request.tablet_id, request.schema_hash,
                                                schema_hash_path_stream.str(), false, restore);
    if (!res.ok()) {
        LOG(WARNING) << "fail to process load headers. res=" << res;
        return res;
    }

    LOG(INFO) << "success to process load headers.";
    return res;
}

void BaseStorageEngine::register_report_listener(ReportWorker* listener) {
    std::lock_guard<std::mutex> l(_report_mtx);
    if (std::find(_report_listeners.begin(), _report_listeners.end(), listener) !=
        _report_listeners.end()) [[unlikely]] {
        return;
    }
    _report_listeners.push_back(listener);
}

void BaseStorageEngine::deregister_report_listener(ReportWorker* listener) {
    std::lock_guard<std::mutex> l(_report_mtx);
    if (auto it = std::find(_report_listeners.begin(), _report_listeners.end(), listener);
        it != _report_listeners.end()) {
        _report_listeners.erase(it);
    }
}

void BaseStorageEngine::notify_listeners() {
    std::lock_guard<std::mutex> l(_report_mtx);
    for (auto& listener : _report_listeners) {
        listener->notify();
    }
}

bool BaseStorageEngine::notify_listener(std::string_view name) {
    bool found = false;
    std::lock_guard<std::mutex> l(_report_mtx);
    for (auto& listener : _report_listeners) {
        if (listener->name() == name) {
            listener->notify();
            found = true;
        }
    }
    return found;
}

void BaseStorageEngine::_evict_quring_rowset_thread_callback() {
    int32_t interval = config::quering_rowsets_evict_interval;
    do {
        _evict_querying_rowset();
        interval = config::quering_rowsets_evict_interval;
        if (interval <= 0) {
            LOG(WARNING) << "quering_rowsets_evict_interval config is illegal: " << interval
                         << ", force set to 1";
            interval = 1;
        }
    } while (!_stop_background_threads_latch.wait_for(std::chrono::seconds(interval)));
}

// check whether any unused rowsets's id equal to rowset_id
bool StorageEngine::check_rowset_id_in_unused_rowsets(const RowsetId& rowset_id) {
    std::lock_guard<std::mutex> lock(_gc_mutex);
    return _unused_rowsets.contains(rowset_id);
}

PendingRowsetGuard StorageEngine::add_pending_rowset(const RowsetWriterContext& ctx) {
    if (ctx.is_local_rowset()) {
        return _pending_local_rowsets.add(ctx.rowset_id);
    }
    return _pending_remote_rowsets.add(ctx.rowset_id);
}

bool StorageEngine::get_peers_replica_backends(int64_t tablet_id, std::vector<TBackend>* backends) {
    TabletSharedPtr tablet = _tablet_manager->get_tablet(tablet_id);
    if (tablet == nullptr) {
        LOG(WARNING) << "tablet is no longer exist: tablet_id=" << tablet_id;
        return false;
    }
    int64_t cur_time = UnixMillis();
    if (cur_time - _last_get_peers_replica_backends_time_ms < 10000) {
        LOG_WARNING("failed to get peers replica backens.")
                .tag("tablet_id", tablet_id)
                .tag("last time", _last_get_peers_replica_backends_time_ms)
                .tag("cur time", cur_time);
        return false;
    }
    LOG_INFO("start get peers replica backends info.").tag("tablet id", tablet_id);
    ClusterInfo* cluster_info = ExecEnv::GetInstance()->cluster_info();
    if (cluster_info == nullptr) {
        LOG(WARNING) << "Have not get FE Master heartbeat yet";
        return false;
    }
    TNetworkAddress master_addr = cluster_info->master_fe_addr;
    if (master_addr.hostname.empty() || master_addr.port == 0) {
        LOG(WARNING) << "Have not get FE Master heartbeat yet";
        return false;
    }
    TGetTabletReplicaInfosRequest request;
    TGetTabletReplicaInfosResult result;
    request.tablet_ids.emplace_back(tablet_id);
    Status rpc_st = ThriftRpcHelper::rpc<FrontendServiceClient>(
            master_addr.hostname, master_addr.port,
            [&request, &result](FrontendServiceConnection& client) {
                client->getTabletReplicaInfos(result, request);
            });

    if (!rpc_st.ok()) {
        LOG(WARNING) << "Failed to get tablet replica infos, encounter rpc failure, "
                        "tablet id: "
                     << tablet_id;
        return false;
    }
    if (result.tablet_replica_infos.contains(tablet_id)) {
        std::vector<TReplicaInfo> reps = result.tablet_replica_infos[tablet_id];
        if (reps.empty()) [[unlikely]] {
            VLOG_DEBUG << "get_peers_replica_backends reps is empty, maybe this tablet is in "
                          "schema change. Go to FE to see more info. Tablet id: "
                       << tablet_id;
        }
        for (const auto& rep : reps) {
            if (rep.replica_id != tablet->replica_id()) {
                TBackend backend;
                backend.__set_host(rep.host);
                backend.__set_be_port(rep.be_port);
                backend.__set_http_port(rep.http_port);
                backend.__set_brpc_port(rep.brpc_port);
                if (rep.__isset.is_alive) {
                    backend.__set_is_alive(rep.is_alive);
                }
                if (rep.__isset.backend_id) {
                    backend.__set_id(rep.backend_id);
                }
                backends->emplace_back(backend);
                std::stringstream backend_string;
                backend.printTo(backend_string);
                LOG_INFO("get 1 peer replica backend info.")
                        .tag("tablet id", tablet_id)
                        .tag("backend info", backend_string.str());
            }
        }
        _last_get_peers_replica_backends_time_ms = UnixMillis();
        LOG_INFO("succeed get peers replica backends info.")
                .tag("tablet id", tablet_id)
                .tag("replica num", backends->size());
        return true;
    }
    return false;
}

// Return json:
// {
//   "CumulativeCompaction": {
//          "/home/disk1" : [10001, 10002],
//          "/home/disk2" : [10003]
//   },
//   "BaseCompaction": {
//          "/home/disk1" : [10001, 10002],
//          "/home/disk2" : [10003]
//   }
// }
void StorageEngine::get_compaction_status_json(std::string* result) {
    _compaction_submit_registry.jsonfy_compaction_status(result);
}

void BaseStorageEngine::add_quering_rowset(RowsetSharedPtr rs) {
    std::lock_guard<std::mutex> lock(_quering_rowsets_mutex);
    _querying_rowsets.emplace(rs->rowset_id(), rs);
}

RowsetSharedPtr BaseStorageEngine::get_quering_rowset(RowsetId rs_id) {
    std::lock_guard<std::mutex> lock(_quering_rowsets_mutex);
    auto it = _querying_rowsets.find(rs_id);
    if (it != _querying_rowsets.end()) {
        return it->second;
    }
    return nullptr;
}

void BaseStorageEngine::_evict_querying_rowset() {
    {
        std::lock_guard<std::mutex> lock(_quering_rowsets_mutex);
        for (auto it = _querying_rowsets.begin(); it != _querying_rowsets.end();) {
            uint64_t now = UnixSeconds();
            // We delay the GC time of this rowset since it's maybe still needed, see #20732
            if (now > it->second->delayed_expired_timestamp()) {
                it = _querying_rowsets.erase(it);
            } else {
                ++it;
            }
        }
    }

    uint64_t now = UnixSeconds();
    ExecEnv::GetInstance()->get_id_manager()->gc_expired_id_file_map(now);
}

bool BaseStorageEngine::_should_delay_large_task() {
    DCHECK_GE(_cumu_compaction_thread_pool->max_threads(),
              _cumu_compaction_thread_pool_used_threads);
    DCHECK_GE(_cumu_compaction_thread_pool_small_tasks_running, 0);
    // Case 1: Multiple threads available => accept large task
    if (_cumu_compaction_thread_pool->max_threads() - _cumu_compaction_thread_pool_used_threads >
        0) {
        return false; // No delay needed
    }
    // Case 2: Only one thread left => accept large task only if another small task is already running
    if (_cumu_compaction_thread_pool_small_tasks_running > 0) {
        return false; // No delay needed
    }
    // Case 3: Only one thread left, this is a large task, and no small tasks are running
    // Delay this task to reserve capacity for potential small tasks
    return true; // Delay this large task
}

bool StorageEngine::add_broken_path(std::string path) {
    std::lock_guard<std::mutex> lock(_broken_paths_mutex);
    auto success = _broken_paths.emplace(path).second;
    if (success) {
        static_cast<void>(_persist_broken_paths());
    }
    return success;
}

bool StorageEngine::remove_broken_path(std::string path) {
    std::lock_guard<std::mutex> lock(_broken_paths_mutex);
    auto count = _broken_paths.erase(path);
    if (count > 0) {
        static_cast<void>(_persist_broken_paths());
    }
    return count > 0;
}

Status StorageEngine::_persist_broken_paths() {
    std::string config_value;
    for (const std::string& path : _broken_paths) {
        config_value += path + ";";
    }

    if (config_value.length() > 0) {
        auto st = config::set_config("broken_storage_path", config_value, true);
        LOG(INFO) << "persist broken_storage_path " << config_value << st;
        return st;
    }

    return Status::OK();
}

Status StorageEngine::submit_clone_task(Tablet* tablet, int64_t version) {
    std::vector<TBackend> backends;
    if (!get_peers_replica_backends(tablet->tablet_id(), &backends)) {
        return Status::Error<ErrorCode::INTERNAL_ERROR, false>(
                "get_peers_replica_backends failed.");
    }
    TAgentTaskRequest task;
    TCloneReq req;
    req.__set_tablet_id(tablet->tablet_id());
    req.__set_schema_hash(tablet->schema_hash());
    req.__set_src_backends(backends);
    req.__set_version(version);
    req.__set_replica_id(tablet->replica_id());
    req.__set_partition_id(tablet->partition_id());
    req.__set_table_id(tablet->table_id());
    task.__set_task_type(TTaskType::CLONE);
    task.__set_clone_req(req);
    task.__set_priority(TPriority::HIGH);
    task.__set_signature(tablet->tablet_id());
    LOG_INFO("BE start to submit missing rowset clone task.")
            .tag("tablet_id", tablet->tablet_id())
            .tag("version", version)
            .tag("replica_id", tablet->replica_id())
            .tag("partition_id", tablet->partition_id())
            .tag("table_id", tablet->table_id());
    RETURN_IF_ERROR(assert_cast<PriorTaskWorkerPool*>(workers->at(TTaskType::CLONE).get())
                            ->submit_high_prior_and_cancel_low(task));
    return Status::OK();
}

int CreateTabletRRIdxCache::get_index(const std::string& key) {
    auto* lru_handle = lookup(key);
    if (lru_handle) {
        Defer release([cache = this, lru_handle] { cache->release(lru_handle); });
        auto* value = (CacheValue*)LRUCachePolicy::value(lru_handle);
        VLOG_DEBUG << "use create tablet idx cache key=" << key << " value=" << value->idx;
        return value->idx;
    }
    return -1;
}

void CreateTabletRRIdxCache::set_index(const std::string& key, int next_idx) {
    assert(next_idx >= 0);
    auto* value = new CacheValue;
    value->idx = next_idx;
    auto* lru_handle = insert(key, value, 1, sizeof(int), CachePriority::NORMAL);
    release(lru_handle);
}
} // namespace doris

Coverage Report

Created: 2026-08-04 11:20