be/src/common/config.h

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.

#pragma once

#include <cstddef>
#include <cstdint>
#include <functional>
#include <map>
#include <mutex>
#include <string>
#include <utility>
#include <vector>

#define DECLARE_FIELD(FIELD_TYPE, FIELD_NAME) extern FIELD_TYPE FIELD_NAME

#define DECLARE_Bool(name) DECLARE_FIELD(bool, name)
#define DECLARE_Int16(name) DECLARE_FIELD(int16_t, name)
#define DECLARE_Int32(name) DECLARE_FIELD(int32_t, name)
#define DECLARE_Int64(name) DECLARE_FIELD(int64_t, name)
#define DECLARE_Double(name) DECLARE_FIELD(double, name)
#define DECLARE_String(name) DECLARE_FIELD(std::string, name)
#define DECLARE_Bools(name) DECLARE_FIELD(std::vector<bool>, name)
#define DECLARE_Int16s(name) DECLARE_FIELD(std::vector<int16_t>, name)
#define DECLARE_Int32s(name) DECLARE_FIELD(std::vector<int32_t>, name)
#define DECLARE_Int64s(name) DECLARE_FIELD(std::vector<int64_t>, name)
#define DECLARE_Doubles(name) DECLARE_FIELD(std::vector<double>, name)
#define DECLARE_Strings(name) DECLARE_FIELD(std::vector<std::string>, name)
#define DECLARE_mBool(name) DECLARE_FIELD(bool, name)
#define DECLARE_mInt16(name) DECLARE_FIELD(int16_t, name)
#define DECLARE_mInt32(name) DECLARE_FIELD(int32_t, name)
#define DECLARE_mInt64(name) DECLARE_FIELD(int64_t, name)
#define DECLARE_mDouble(name) DECLARE_FIELD(double, name)
#define DECLARE_mString(name) DECLARE_FIELD(std::string, name)

#define DEFINE_FIELD(FIELD_TYPE, FIELD_NAME, FIELD_DEFAULT, VALMUTABLE)                      \
    FIELD_TYPE FIELD_NAME;                                                                   \
    static Register reg_##FIELD_NAME(#FIELD_TYPE, #FIELD_NAME, &(FIELD_NAME), FIELD_DEFAULT, \
                                     VALMUTABLE);

#define DEFINE_VALIDATOR(FIELD_NAME, VALIDATOR)              \
    static auto validator_##FIELD_NAME = VALIDATOR;          \
    static RegisterConfValidator reg_validator_##FIELD_NAME( \
            #FIELD_NAME, []() -> bool { return validator_##FIELD_NAME(FIELD_NAME); });

// DEFINE_ON_UPDATE macro is used to register a callback function that will be called
// when the config field is updated at runtime.
// The callback function signature is: void callback(T old_value, T new_value)
// where T is the type of the config field.
// Example:
//   DEFINE_ON_UPDATE(my_config, [](int64_t old_val, int64_t new_val) {
//       LOG(INFO) << "my_config changed from " << old_val << " to " << new_val;
//   });
#define DEFINE_ON_UPDATE_IMPL(FIELD_NAME, CALLBACK)                               \
    static auto on_update_callback_##FIELD_NAME = CALLBACK;                       \
    static RegisterConfUpdateCallback reg_update_callback_##FIELD_NAME(           \
            #FIELD_NAME, [](const void* old_ptr, const void* new_ptr) {           \
                on_update_callback_##FIELD_NAME(                                  \
                        *reinterpret_cast<const decltype(FIELD_NAME)*>(old_ptr),  \
                        *reinterpret_cast<const decltype(FIELD_NAME)*>(new_ptr)); \
            });

#define DEFINE_ON_UPDATE(name, callback) DEFINE_ON_UPDATE_IMPL(name, callback)

#define DEFINE_Int16(name, defaultstr) DEFINE_FIELD(int16_t, name, defaultstr, false)
#define DEFINE_Bools(name, defaultstr) DEFINE_FIELD(std::vector<bool>, name, defaultstr, false)
#define DEFINE_Doubles(name, defaultstr) DEFINE_FIELD(std::vector<double>, name, defaultstr, false)
#define DEFINE_Int16s(name, defaultstr) DEFINE_FIELD(std::vector<int16_t>, name, defaultstr, false)
#define DEFINE_Int32s(name, defaultstr) DEFINE_FIELD(std::vector<int32_t>, name, defaultstr, false)
#define DEFINE_Int64s(name, defaultstr) DEFINE_FIELD(std::vector<int64_t>, name, defaultstr, false)
#define DEFINE_Bool(name, defaultstr) DEFINE_FIELD(bool, name, defaultstr, false)
#define DEFINE_Double(name, defaultstr) DEFINE_FIELD(double, name, defaultstr, false)
#define DEFINE_Int32(name, defaultstr) DEFINE_FIELD(int32_t, name, defaultstr, false)
#define DEFINE_Int64(name, defaultstr) DEFINE_FIELD(int64_t, name, defaultstr, false)
#define DEFINE_String(name, defaultstr) DEFINE_FIELD(std::string, name, defaultstr, false)
#define DEFINE_Strings(name, defaultstr) \
    DEFINE_FIELD(std::vector<std::string>, name, defaultstr, false)
#define DEFINE_mBool(name, defaultstr) DEFINE_FIELD(bool, name, defaultstr, true)
#define DEFINE_mInt16(name, defaultstr) DEFINE_FIELD(int16_t, name, defaultstr, true)
#define DEFINE_mInt32(name, defaultstr) DEFINE_FIELD(int32_t, name, defaultstr, true)
#define DEFINE_mInt64(name, defaultstr) DEFINE_FIELD(int64_t, name, defaultstr, true)
#define DEFINE_mDouble(name, defaultstr) DEFINE_FIELD(double, name, defaultstr, true)
#define DEFINE_mString(name, defaultstr) DEFINE_FIELD(std::string, name, defaultstr, true)
#define DEFINE_Validator(name, validator) DEFINE_VALIDATOR(name, validator)

namespace doris {
class Status;

// If you want to modify the value of config, please go to common/config.cpp to modify.
namespace config {
// Dir of custom config file
DECLARE_String(custom_config_dir);

// Dir of jdbc drivers
DECLARE_String(jdbc_drivers_dir);

// cluster id
DECLARE_Int32(cluster_id);
// port on which BackendService is exported
DECLARE_Int32(be_port);

// port for brpc
DECLARE_Int32(brpc_port);

// port for arrow flight sql
// Default -1, do not start arrow flight sql server.
DECLARE_Int32(arrow_flight_sql_port);

// Validate Arrow input buffers in opted-in Arrow readers before converting them to Doris columns.
DECLARE_Bool(enable_arrow_input_validation);

// port for cdc client scan oltp cdc data
DECLARE_Int32(cdc_client_port);

// JVM options passed to cdc_client (whitespace-separated). Inserted before -jar.
DECLARE_String(cdc_client_java_opts);

// If the external client cannot directly access priority_networks, set public_host to be accessible
// to external client.
// There are usually two usage scenarios:
// 1. in production environment, it is often inconvenient to expose Doris BE nodes to the external network.
// However, a reverse proxy (such as Nginx) can be added to all Doris BE nodes, and the external client will be
// randomly routed to a Doris BE node when connecting to Nginx. set public_host to the host of Nginx.
// 2. if priority_networks is an internal network IP, and BE node has its own independent external IP,
// but Doris currently does not support modifying priority_networks, setting public_host to the real external IP.
DECLARE_mString(public_host);

// If the BE node is connected to the external network through a reverse proxy like Nginx
// and need to use Arrow Flight SQL, should add a server in Nginx to reverse proxy
// `Nginx:arrow_flight_sql_proxy_port` to `BE_priority_networks:arrow_flight_sql_port`. For example:
// upstream arrowflight {
//    server 10.16.10.8:8069;
//    server 10.16.10.8:8068;
//}
// server {
//    listen 8167 http2;
//    listen [::]:8167 http2;
//    server_name doris.arrowflight.com;
// }
DECLARE_Int32(arrow_flight_sql_proxy_port);

// the number of bthreads for brpc, the default value is set to -1,
// which means the number of bthreads is #cpu-cores
DECLARE_Int32(brpc_num_threads);
DECLARE_Int32(brpc_idle_timeout_sec);

// Maximum number of multi-dimensional bvar samples exported by /brpc_metrics.
DECLARE_String(bvar_max_dump_multi_dimension_metric_num);

// Declare a selection strategy for those servers have many ips.
// Note that there should at most one ip match this list.
// This is a list in semicolon-delimited format, in CIDR notation, e.g. 10.10.10.0/24
// If no ip match this rule, will choose one randomly.
DECLARE_String(priority_networks);

// performance moderate or compact, only tcmalloc compile
DECLARE_String(memory_mode);

// if true, process memory limit and memory usage based on cgroup memory info.
DECLARE_mBool(enable_use_cgroup_memory_info);

// process memory limit specified as number of bytes
// ('<int>[bB]?'), megabytes ('<float>[mM]'), gigabytes ('<float>[gG]'),
// or percentage of the physical memory ('<int>%').
// defaults to bytes if no unit is given"
// must larger than 0. and if larger than physical memory size,
// it will be set to physical memory size.
DECLARE_String(mem_limit);

// Soft memory limit as a fraction of hard memory limit.
DECLARE_Double(soft_mem_limit_frac);

// Cache capacity reduce mem limit as a fraction of soft mem limit.
DECLARE_mDouble(cache_capacity_reduce_mem_limit_frac);

// Schema change memory limit as a fraction of soft memory limit.
DECLARE_Double(schema_change_mem_limit_frac);

// Many modern allocators (for example) do not do a mremap for
// realloc, even in case of large enough chunks of memory. Although this allows
// you to increase performance and reduce memory consumption during realloc.
// To fix this, we do mremap manually if the chunk of memory is large enough.
//
// The threshold (128 MB, 128 * (1ULL << 20)) is chosen quite large, since changing the address
// space is very slow, especially in the case of a large number of threads. We
// expect that the set of operations mmap/something to do/mremap can only be
// performed about 1000 times per second.
//
// P.S. This is also required, because tcmalloc can not allocate a chunk of
// memory greater than 16 GB.
DECLARE_mInt64(mmap_threshold); // bytes

// When hash table capacity is greater than 2^double_grow_degree(default 2G), grow when 75% of the capacity is satisfied.
// Increase can reduce the number of hash table resize, but may waste more memory.
DECLARE_mInt32(hash_table_double_grow_degree);

// The max fill rate for hash table
DECLARE_mInt32(max_fill_rate);

DECLARE_mInt32(double_resize_threshold);

// The maximum low water mark of the system `/proc/meminfo/MemAvailable`, Unit byte, default -1.
// if it is -1, then low water mark = min(MemTotal - MemLimit, MemTotal * 5%), which is 3.2G on a 64G machine.
// Turn up max. more memory buffers will be reserved for Memory GC.
// Turn down max. will use as much memory as possible.
// note that: `max_` prefix should be removed, but keep it for compatibility.
DECLARE_Int64(max_sys_mem_available_low_water_mark_bytes);

// reserve a small amount of memory so we do not trigger MinorGC
DECLARE_Int64(memtable_limiter_reserved_memory_bytes);

// The size of the memory that gc wants to release each time, as a percentage of the mem limit.
DECLARE_mString(process_minor_gc_size);
DECLARE_mString(process_full_gc_size);

// gc will release cache, cancel task, and task will wait for gc to release memory,
// default gc strategy is conservative, if you want to exclude the interference of gc, let it be true
DECLARE_mBool(disable_memory_gc);

// for the query being canceled,
// if (current time - cancel start time) < revoke_memory_max_tolerance_ms, the query memory is counted in `freed_memory`,
// and the query memory is expected to be released soon.
// if > revoke_memory_max_tolerance_ms, the query memory will not be counted in `freed_memory`,
// and the query may be blocked during the cancel process. skip this query and continue to cancel other queries.
DECLARE_mInt64(revoke_memory_max_tolerance_ms);

// if false, turn off all stacktrace
DECLARE_mBool(enable_stacktrace);

// when alloc memory larger than stacktrace_in_alloc_large_memory_bytes, default 2G - 1,
// if alloc successful, will print a warning with stacktrace, but not prevent memory alloc.
// if alloc failed using Doris Allocator, will print stacktrace in error log.
// if is -1, disable print stacktrace when alloc large memory.
DECLARE_mInt64(stacktrace_in_alloc_large_memory_bytes);

// when alloc memory larger than crash_in_alloc_large_memory_bytes will crash, default -1 means disabled.
// if you need a core dump to analyze large memory allocation,
// modify this parameter to crash when large memory allocation occur will help
DECLARE_mInt64(crash_in_alloc_large_memory_bytes);

// The actual meaning of this parameter is `debug_memory`.
// 1. crash in memory tracker inaccurate, if memory tracker value is inaccurate, BE will crash.
//    usually used in test environments, default value is false.
// 2. print more memory logs.
DECLARE_mBool(crash_in_memory_tracker_inaccurate);

// default is true. if any memory tracking in Orphan mem tracker will report error.
// !! not modify the default value of this conf!! otherwise memory errors cannot be detected in time.
// allocator free memory not need to check, because when the thread memory tracker label is Orphan,
// use the tracker saved in Allocator.
DECLARE_mBool(enable_memory_orphan_check);

// The maximum time a thread waits for a full GC. Currently only query will wait for full gc.
DECLARE_mInt32(thread_wait_gc_max_milliseconds);

// reach mem limit, don't serialize in batch
DECLARE_mInt64(pre_serialize_keys_limit_bytes);

// the port heartbeat service used
DECLARE_Int32(heartbeat_service_port);
// the count of heart beat service
DECLARE_Int32(heartbeat_service_thread_count);
// the count of thread to create table
DECLARE_Int32(create_tablet_worker_count);
// the count of thread to drop table
DECLARE_Int32(drop_tablet_worker_count);
// the count of thread to batch load
DECLARE_Int32(push_worker_count_normal_priority);
// the count of thread to high priority batch load
DECLARE_Int32(push_worker_count_high_priority);
// the count of thread to publish version
DECLARE_Int32(publish_version_worker_count);
// the count of tablet thread to publish version
DECLARE_Int32(tablet_publish_txn_max_thread);
// the timeout of EnginPublishVersionTask
DECLARE_Int32(publish_version_task_timeout_s);
// the count of thread to calc delete bitmap
DECLARE_Int32(calc_delete_bitmap_max_thread);
// the num of threads to calc delete bitmap when building rowset
DECLARE_Int32(calc_delete_bitmap_for_load_max_thread);
// the count of thread to calc delete bitmap worker, only used for cloud
DECLARE_Int32(calc_delete_bitmap_worker_count);
// the count of thread to calc tablet delete bitmap task, only used for cloud
DECLARE_Int32(calc_tablet_delete_bitmap_task_max_thread);
// the count of thread to clear transaction task
DECLARE_Int32(clear_transaction_task_worker_count);
// the count of thread to delete
DECLARE_Int32(delete_worker_count);
// the count of thread to alter table
DECLARE_Int32(alter_tablet_worker_count);
// the count of thread to alter index
DECLARE_Int32(alter_index_worker_count);
// the count of thread to clone
DECLARE_Int32(clone_worker_count);
// the count of thread to clone
DECLARE_Int32(storage_medium_migrate_count);
// the count of thread to check consistency
DECLARE_Int32(check_consistency_worker_count);
// the count of thread to upload
DECLARE_Int32(upload_worker_count);
// the count of thread to download
DECLARE_Int32(download_worker_count);
// the count of thread to make snapshot
DECLARE_Int32(make_snapshot_worker_count);
// the count of thread to release snapshot
DECLARE_Int32(release_snapshot_worker_count);
// the count of thread to make committed rowsets visible in cloud mode
DECLARE_Int32(cloud_make_committed_rs_visible_worker_count);
// report random wait a little time to avoid FE receiving multiple be reports at the same time.
// do not set it to false for production environment
DECLARE_mBool(report_random_wait);
// the interval time(seconds) for agent report tasks signature to FE
DECLARE_mInt32(report_task_interval_seconds);
// the interval time(seconds) for agent report disk state to FE
DECLARE_mInt32(report_disk_state_interval_seconds);
// the interval time(seconds) for agent report olap table to FE
DECLARE_mInt32(report_tablet_interval_seconds);
// the max download speed(KB/s)
DECLARE_mInt32(max_download_speed_kbps);
// download low speed limit(KB/s)
DECLARE_mInt32(download_low_speed_limit_kbps);
// download low speed time(seconds)
DECLARE_mInt32(download_low_speed_time);
// whether to download small files in batch.
DECLARE_mBool(enable_batch_download);
// whether to enable stream load forward endpoint for cloud group commit
DECLARE_mBool(enable_group_commit_streamload_be_forward);
// whether to check md5sum when download
DECLARE_mBool(enable_download_md5sum_check);
// download binlog meta timeout
DECLARE_mInt32(download_binlog_meta_timeout_ms);
// the interval time(seconds) for agent report index policy to FE
DECLARE_mInt32(report_index_policy_interval_seconds);

// DNS cache: log the "Failed to resolve hostname ... use cached ip" warning
// only once per N consecutive failures for the same hostname, to avoid
// flooding be.WARNING. Set <= 1 to log every failure (legacy behavior).
// Should be set <= dns_cache_max_consecutive_failures, otherwise only the
// first-failure log is ever emitted before a host is evicted.
DECLARE_mInt32(dns_cache_log_every_n_failures);

// DNS cache: evict a hostname after this many consecutive resolution failures.
// At the default refresh interval of 60s, the default value of 30 means a
// hostname that was once successfully resolved is evicted after ~30 minutes of
// being un-resolvable.  Hostnames that have never been successfully resolved are
// not tracked and are unaffected by this threshold.
// Eviction additionally requires the most recent failure to be an authoritative
// NXDOMAIN (getaddrinfo returning EAI_NONAME), i.e. the resolver positively
// stating that the name does not exist. Transient failures such as EAI_AGAIN
// (resolver unreachable or timed out) never evict, so a DNS-server outage
// degrades to serving the last known IP instead of emptying the cache for every
// hostname at once and turning a DNS incident into a cluster-wide RPC outage.
// Set <= 0 to disable eviction (legacy behavior, kept for backward compatibility).
DECLARE_mInt32(dns_cache_max_consecutive_failures);

// DNS cache: seconds to suppress re-resolve attempts for a hostname that could
// not be resolved -- either because it was evicted after repeated failures, or
// because it never resolved in the first place.  During this window get()
// returns an error immediately (no blocking getaddrinfo) so request threads are
// not stalled while the backend is being drained or while a bad hostname is
// being retried.
// This also bounds recovery latency: the refresh thread does not retry hostnames
// that are no longer in the cache, so a host comes back only when a caller's
// get() runs after this TTL expires (one such retry per host per TTL).
// Set <= 0 to disable the negative cache.
DECLARE_mInt32(dns_cache_negative_ttl_seconds);

// deprecated, use env var LOG_DIR in be.conf
DECLARE_String(sys_log_dir);
// for udf
DECLARE_String(user_function_dir);
// INFO, WARNING, ERROR, FATAL
DECLARE_String(sys_log_level);
// TIME-DAY, TIME-HOUR, SIZE-MB-nnn
DECLARE_String(sys_log_roll_mode);
// log roll num
DECLARE_Int32(sys_log_roll_num);
// verbose log
DECLARE_Strings(sys_log_verbose_modules);
// verbose log level
DECLARE_Int32(sys_log_verbose_level);
// verbose log FLAGS_v
DECLARE_Int32(sys_log_verbose_flags_v);
// log buffer level
DECLARE_String(log_buffer_level);
// Attach query identity to runtime log events (requires restart).
DECLARE_Bool(sys_log_enable_query_id);
// log enable custom date time format
DECLARE_Bool(sys_log_enable_custom_date_time_format);
// log custom date time format (https://en.cppreference.com/w/cpp/io/manip/put_time)
DECLARE_String(sys_log_custom_date_time_format);
// log custom date time milliseconds format (fmt::format)
DECLARE_String(sys_log_custom_date_time_ms_format);

// number of threads available to serve backend execution requests
DECLARE_Int32(be_service_threads);

// interval between profile reports; in seconds
DECLARE_mInt32(pipeline_status_report_interval);
// Time slice for pipeline task execution (ms)
DECLARE_mInt32(pipeline_task_exec_time_slice);

// task executor min concurrency per task
DECLARE_mInt32(task_executor_min_concurrency_per_task);
// task executor max concurrency per task
DECLARE_mInt32(task_executor_max_concurrency_per_task);
// task task executor inital split max concurrency per task, later concurrency may be adjusted dynamically
DECLARE_mInt32(task_executor_initial_max_concurrency_per_task);
// Enable task executor in internal table scan.
DECLARE_Bool(enable_task_executor_in_internal_table);
// Enable task executor in external table scan.
DECLARE_Bool(enable_task_executor_in_external_table);

// number of scanner thread pool size for olap table
// and the min thread num of remote scanner thread pool
DECLARE_mInt32(doris_scanner_thread_pool_thread_num);
DECLARE_mInt32(doris_scanner_min_thread_pool_thread_num);
// number of batch size to fetch the remote split source
DECLARE_mInt32(remote_split_source_batch_size);
// max number of remote scanner thread pool size
// if equal to -1, value is std::max(512, CpuInfo::num_cores() * 10)
DECLARE_Int32(doris_max_remote_scanner_thread_pool_thread_num);
// number of olap scanner thread pool queue size
DECLARE_Int32(doris_scanner_thread_pool_queue_size);
// default thrift client connect timeout(in seconds)
DECLARE_mInt32(thrift_connect_timeout_seconds);
// default thrift client retry interval (in milliseconds)
DECLARE_mInt64(thrift_client_retry_interval_ms);
// max message size of thrift request
// default: 100 * 1024 * 1024
DECLARE_mInt32(thrift_max_message_size);
// max bytes number for single scan range, used in segmentv2
DECLARE_mInt32(doris_scan_range_max_mb);
// single read execute fragment row number
DECLARE_mInt32(doris_scanner_row_num);
// single read execute fragment row bytes
DECLARE_mInt32(doris_scanner_row_bytes);
// Minimum interval in milliseconds between adaptive scanner concurrency adjustments
DECLARE_mInt32(doris_scanner_dynamic_interval_ms);
// (Advanced) Maximum size of per-query receive-side buffer
DECLARE_mInt32(exchg_node_buffer_size_bytes);
DECLARE_mInt32(exchg_buffer_queue_capacity_factor);

// memory_limitation_per_thread_for_schema_change_bytes unit bytes
DECLARE_mInt64(memory_limitation_per_thread_for_schema_change_bytes);

// all cache prune interval, used by GC and periodic thread.
DECLARE_mInt32(cache_prune_interval_sec);
DECLARE_mInt32(cache_periodic_prune_stale_sweep_sec);
// the clean interval of tablet lookup cache
DECLARE_mInt32(tablet_lookup_cache_stale_sweep_time_sec);
DECLARE_mInt32(point_query_row_cache_stale_sweep_time_sec);
DECLARE_mInt32(disk_stat_monitor_interval);
DECLARE_mInt32(unused_rowset_monitor_interval);
// Legacy name retained for compatibility; controls GLOBAL_ROWID_COL file-map GC.
DECLARE_mInt32(quering_rowsets_evict_interval);
DECLARE_String(storage_root_path);
DECLARE_mString(broken_storage_path);
DECLARE_Int32(min_active_scan_threads);
DECLARE_Int32(min_active_file_scan_threads);

// Config is used to check incompatible old format hdr_ format
// whether doris uses strict way. When config is true, process will log fatal
// and exit. When config is false, process will only log warning.
DECLARE_Bool(storage_strict_check_incompatible_old_format);

// BE process will exit if the percentage of error disk reach this value.
DECLARE_mInt32(max_percentage_of_error_disk);
DECLARE_mInt32(default_num_rows_per_column_file_block);
// pending data policy
DECLARE_mInt32(pending_data_expire_time_sec);
// inc_rowset snapshot rs sweep time interval
DECLARE_mInt32(tablet_rowset_stale_sweep_time_sec);
// tablet stale rowset sweep by threshold size
DECLARE_Bool(tablet_rowset_stale_sweep_by_size);
DECLARE_mInt32(tablet_rowset_stale_sweep_threshold_size);
// garbage sweep policy
DECLARE_Int32(max_garbage_sweep_interval);
DECLARE_Int32(min_garbage_sweep_interval);
// garbage sweep every batch will sleep 1ms
DECLARE_mInt32(garbage_sweep_batch_size);
DECLARE_mInt32(snapshot_expire_time_sec);
// It is only a recommended value. When the disk space is insufficient,
// the file storage period under trash dose not have to comply with this parameter.
DECLARE_mInt32(trash_file_expire_time_sec);
// minimum file descriptor number
// modify them upon necessity
DECLARE_Int32(min_file_descriptor_number);
DECLARE_mBool(disable_segment_cache);
// Enable checking segment rows consistency between rowset meta and segment footer
DECLARE_mBool(enable_segment_rows_consistency_check);
DECLARE_mBool(enable_segment_rows_check_core);
// ATTENTION: For test only. In test environment, there are no historical data,
// so all rowset meta should have segment rows info.
DECLARE_mBool(fail_when_segment_rows_not_in_rowset_meta);
DECLARE_String(row_cache_mem_limit);

// Cache for storage page size
DECLARE_String(storage_page_cache_limit);
// Shard size for page cache, the value must be power of two.
// It's recommended to set it to a value close to the number of BE cores in order to reduce lock contentions.
DECLARE_Int32(storage_page_cache_shard_size);
DECLARE_mInt32(file_cache_mem_storage_shard_num);
// Percentage for index page cache
// all storage page cache will be divided into data_page_cache and index_page_cache
DECLARE_Int32(index_page_cache_percentage);
// whether to disable page cache feature in storage
// TODO delete it. Divided into Data page, Index page, pk index page
DECLARE_Bool(disable_storage_page_cache);
// whether to disable row cache feature in storage
DECLARE_mBool(disable_storage_row_cache);
// Parquet page cache: threshold ratio for caching decompressed vs compressed pages
// If uncompressed_size / compressed_size <= threshold, cache decompressed;
// otherwise cache compressed if enable_parquet_cache_compressed_pages = true
DECLARE_Double(parquet_page_cache_decompress_threshold);
// Parquet page cache: whether to enable caching compressed pages (when ratio exceeds threshold)
DECLARE_Bool(enable_parquet_cache_compressed_pages);
// whether to disable pk page cache feature in storage
DECLARE_Bool(disable_pk_storage_page_cache);

// Cache for mow primary key storage page size, it's seperated from
// storage_page_cache_limit
DECLARE_String(pk_storage_page_cache_limit);
// data page size for primary key index
DECLARE_Int32(primary_key_data_page_size);

// inc_rowset snapshot rs sweep time interval
DECLARE_mInt32(data_page_cache_stale_sweep_time_sec);
DECLARE_mInt32(index_page_cache_stale_sweep_time_sec);
// great impact on the performance of MOW, so it can be longer.
DECLARE_mInt32(pk_index_page_cache_stale_sweep_time_sec);

DECLARE_mBool(enable_low_cardinality_optimize);
DECLARE_Bool(enable_low_cardinality_cache_code);

// Adaptive batch size: dynamically adjust SegmentIterator chunk row count using EWMA
// so that each output block stays close to preferred_block_size_bytes.
// When false, the fixed batch_size row behaviour is preserved.
DECLARE_mBool(enable_adaptive_batch_size);

// be policy
// whether check compaction checksum
DECLARE_mBool(enable_compaction_checksum);
// whether disable automatic compaction task
DECLARE_mBool(disable_auto_compaction);
// whether enable vertical compaction
DECLARE_mBool(enable_vertical_compaction);
// whether enable ordered data compaction
DECLARE_mBool(enable_ordered_data_compaction);
// In vertical compaction, column number for every group
DECLARE_mInt32(vertical_compaction_num_columns_per_group);
// In vertical compaction, max memory usage for row_source_buffer
DECLARE_Int32(vertical_compaction_max_row_source_memory_mb);
// In vertical compaction, max dest segment file size
DECLARE_mInt64(vertical_compaction_max_segment_size);
// Threshold for sparse column compaction optimization (average bytes per row)
// Density threshold for sparse column compaction optimization
// density = (total_cells - null_cells) / total_cells, smaller means more sparse
// When density <= threshold, enable sparse optimization
// 0 = disable optimization, 1 = always enable
// Default 1 means always enable sparse optimization
DECLARE_mDouble(sparse_column_compaction_threshold_percent);
// Enable RLE batch Put optimization for compaction
DECLARE_mBool(enable_rle_batch_put_optimization);
DECLARE_Bool(enable_bmi2_optimizations);

// In ordered data compaction, min segment size for input rowset
DECLARE_mInt32(ordered_data_compaction_min_segment_size);

// This config can be set to limit thread number in compaction thread pool.
DECLARE_mInt32(max_base_compaction_threads);
DECLARE_mInt32(max_cumu_compaction_threads);

// Binlog Compaction
DECLARE_mInt64(binlog_compaction_wait_timesec_after_visible);
DECLARE_mInt64(binlog_compaction_goal_size_mbytes);
DECLARE_mInt32(binlog_compaction_task_num_per_disk);
DECLARE_mInt32(binlog_compaction_file_count_threshold);
DECLARE_mInt32(binlog_level_compaction_max_deltas);
DECLARE_mInt64(binlog_compaction_time_threshold_seconds);
DECLARE_mInt32(max_binlog_compaction_threads);

DECLARE_Bool(enable_base_compaction_idle_sched);
DECLARE_mInt64(base_compaction_min_rowset_num);
DECLARE_mInt64(base_compaction_max_compaction_score);
DECLARE_mInt64(mow_base_compaction_max_compaction_score);
DECLARE_mDouble(base_compaction_min_data_ratio);
DECLARE_mInt64(base_compaction_dup_key_max_file_size_mbytes);

DECLARE_Bool(enable_skip_tablet_compaction);
DECLARE_mInt32(skip_tablet_compaction_second);
// output rowset of cumulative compaction total disk size exceed this config size,
// this rowset will be given to base compaction, unit is m byte.
DECLARE_mInt64(compaction_promotion_size_mbytes);

// output rowset of cumulative compaction total disk size exceed this config ratio of
// base rowset's total disk size, this rowset will be given to base compaction. The value must be between
// 0 and 1.
DECLARE_mDouble(compaction_promotion_ratio);

// the smallest size of rowset promotion. When the rowset is less than this config, this
// rowset will be not given to base compaction. The unit is m byte.
DECLARE_mInt64(compaction_promotion_min_size_mbytes);

// When output rowset of cumulative compaction total version count (end_version - start_version)
// exceed this config count, the rowset will be moved to base compaction
// NOTE: this config will work for unique key merge-on-write table only, to reduce version count
// related cost on delete bitmap more effectively.
DECLARE_mInt64(compaction_promotion_version_count);

// The lower bound size to do cumulative compaction. When total disk size of candidate rowsets is less than
// this size, size_based policy may not do to cumulative compaction. The unit is m byte.
DECLARE_mInt64(compaction_min_size_mbytes);

// cumulative compaction policy: min and max delta file's number
DECLARE_mInt64(cumulative_compaction_min_deltas);
DECLARE_mInt64(cumulative_compaction_max_deltas);
DECLARE_mInt32(cumulative_compaction_max_deltas_factor);

// The upper limit of "permits" held by all compaction tasks. This config can be set to limit memory consumption for compaction.
DECLARE_mInt64(total_permits_for_compaction_score);

// sleep interval in ms after generated compaction tasks
DECLARE_mInt32(generate_compaction_tasks_interval_ms);

// Compaction task number per disk.
// Must be greater than 2, because Base compaction and Cumulative compaction have at least one thread each.
DECLARE_mInt32(compaction_task_num_per_disk);
// compaction thread num for fast disk(typically .SSD), must be greater than 2.
DECLARE_mInt32(compaction_task_num_per_fast_disk);

// How many rounds of cumulative compaction for each round of base compaction when compaction tasks generation.
DECLARE_mInt32(cumulative_compaction_rounds_for_each_base_compaction_round);

// Minimum number of threads required in the thread pool to activate the large cumu compaction delay strategy.
// The delay strategy is only applied when the thread pool has at least this many threads.
DECLARE_mInt32(large_cumu_compaction_task_min_thread_num);
// Maximum size threshold (in bytes) for input rowsets. Compaction tasks with input size
// exceeding this threshold will be delayed when thread pool is near capacity. Default 100MB.
DECLARE_mInt32(large_cumu_compaction_task_bytes_threshold);
// Maximum row count threshold for compaction input. Compaction tasks with row count
// exceeding this threshold will be delayed when thread pool is near capacity. Default 1 million.
DECLARE_mInt32(large_cumu_compaction_task_row_num_threshold);

// Not compact the invisible versions, but with some limitations:
// if not timeout, keep no more than compaction_keep_invisible_version_max_count versions;
// if timeout, keep no more than compaction_keep_invisible_version_min_count versions.
DECLARE_mInt32(compaction_keep_invisible_version_timeout_sec);
DECLARE_mInt32(compaction_keep_invisible_version_min_count);
DECLARE_mInt32(compaction_keep_invisible_version_max_count);

// Threshold to logging compaction trace, in seconds.
DECLARE_mInt32(base_compaction_trace_threshold);
DECLARE_mInt32(cumulative_compaction_trace_threshold);
DECLARE_mBool(disable_compaction_trace_log);

DECLARE_mBool(enable_compaction_task_tracker);
DECLARE_mInt32(compaction_task_tracker_max_records);

// Interval to picking rowset to compact, in seconds
DECLARE_mInt64(pick_rowset_to_compact_interval_sec);

// Compaction priority schedule
DECLARE_mBool(enable_compaction_priority_scheduling);
DECLARE_mInt32(low_priority_compaction_task_num_per_disk);
DECLARE_mInt32(low_priority_compaction_score_threshold);

// Thread count to do tablet meta checkpoint, -1 means use the data directories count.
DECLARE_Int32(max_meta_checkpoint_threads);

// Threshold to logging agent task trace, in seconds.
DECLARE_mInt32(agent_task_trace_threshold_sec);

// This config can be set to limit thread number in tablet migration thread pool.
DECLARE_Int32(min_tablet_migration_threads);
DECLARE_Int32(max_tablet_migration_threads);

DECLARE_mInt32(finished_migration_tasks_size);
// If size less than this, the remaining rowsets will be force to complete
DECLARE_mInt32(migration_remaining_size_threshold_mb);
// If the task runs longer than this time, the task will be terminated, in seconds.
// timeout = std::max(migration_task_timeout_secs,  tablet size / 1MB/s)
DECLARE_mInt32(migration_task_timeout_secs);
// timeout for try_lock migration lock
DECLARE_Int64(migration_lock_timeout_ms);

// Port to start debug webserver on
DECLARE_Int32(webserver_port);
// TLS module enable flag
DECLARE_Bool(enable_tls);
// Path of TLS certificate
DECLARE_String(tls_certificate_path);
// Path of TLS private key
DECLARE_String(tls_private_key_path);
// Password for encrypted TLS private key
DECLARE_String(tls_private_key_password);
// TLS peer verification mode
DECLARE_String(tls_verify_mode);
// Path of TLS CA certificate
DECLARE_String(tls_ca_certificate_path);
// TLS certificate reload interval, in seconds
DECLARE_Int32(tls_cert_refresh_interval_seconds);
// Comma-separated excluded server protocols: brpc,thrift,http,arrowflight
DECLARE_String(tls_excluded_protocols);
// Required peer certificate DNS SAN allowlist for private protocols, syntax: brpc=a.com;thrift=b.com.
// Empty means allow all peers. Once configured, the list acts as an allowlist and only peers whose
// DNS SAN matches at least one configured entry for that protocol are allowed.
DECLARE_String(tls_peer_cert_required_san_dns);
// Https enable flag
DECLARE_Bool(enable_https);
// Path of certificate
DECLARE_String(ssl_certificate_path);
// Path of private key
DECLARE_String(ssl_private_key_path);
// Whether to check authorization
DECLARE_Bool(enable_all_http_auth);
// Number of webserver workers
DECLARE_Int32(webserver_num_workers);

// Async replies: stream load only now
// reply wait timeout only happens if:
// 1. Stream load fragment execution times out
//    HTTP request freed → stream load canceled
// 2. Client disconnects
DECLARE_mInt32(async_reply_timeout_s);

// Used for mini Load. mini load data file will be removed after this time.
DECLARE_Int64(load_data_reserve_hours);
// log error log will be removed after this time
DECLARE_mInt64(load_error_log_reserve_hours);
// error log size limit, default 200MB
DECLARE_mInt64(load_error_log_limit_bytes);

// Dedicated load cancellation workers, default 32. Must be positive; requires a restart.
DECLARE_Int32(brpc_load_light_work_pool_threads);
// Queue capacity for the dedicated load cancellation pool.
// -1 selects max(1024, CPU cores * 32) queued requests. Requires a restart.
DECLARE_Int32(brpc_load_light_work_pool_max_queue_size);

// be brpc interface is classified into two categories: light and heavy
// each category has diffrent thread number
// threads to handle heavy api interface, such as transmit_block etc
DECLARE_Int32(brpc_heavy_work_pool_threads);
DECLARE_Int32(brpc_peer_fetch_pool_threads);
// threads to handle light api interface, such as exec_plan_fragment_prepare/exec_plan_fragment_start
DECLARE_Int32(brpc_light_work_pool_threads);
DECLARE_Int32(brpc_heavy_work_pool_max_queue_size);
DECLARE_Int32(brpc_peer_fetch_pool_max_queue_size);
DECLARE_Int32(brpc_light_work_pool_max_queue_size);
DECLARE_mBool(enable_bthread_transmit_block);
DECLARE_Int32(brpc_arrow_flight_work_pool_threads);
DECLARE_Int32(brpc_arrow_flight_work_pool_max_queue_size);

// The maximum amount of data that can be processed by a stream load
DECLARE_mInt64(streaming_load_max_mb);
// Some data formats, such as JSON, cannot be streamed.
// Therefore, it is necessary to limit the maximum number of
// such data when using stream load to prevent excessive memory consumption.
DECLARE_mInt64(streaming_load_json_max_mb);
// the alive time of a TabletsChannel.
// If the channel does not receive any data till this time,
// the channel will be removed.
DECLARE_mInt32(streaming_load_rpc_max_alive_time_sec);
// the timeout of a rpc to open the tablet writer in remote BE.
// short operation time, can set a short timeout
DECLARE_Int32(tablet_writer_open_rpc_timeout_sec);
// You can ignore brpc error '[E1011]The server is overcrowded' when writing data.
DECLARE_mBool(tablet_writer_ignore_eovercrowded);
// Whether to enable stream load record function, the default is false.
// False: disable stream load record
DECLARE_mBool(enable_stream_load_record);
// Whether to enable stream load record to audit log table, the default is true.
DECLARE_mBool(enable_stream_load_record_to_audit_log_table);
// the maximum bytes of a batch of stream load records to audit log table
DECLARE_mInt64(stream_load_record_batch_bytes);
// the interval to send a batch of stream load records to audit log table
DECLARE_mInt64(stream_load_record_batch_interval_secs);
// batch size of stream load record reported to FE
DECLARE_mInt32(stream_load_record_batch_size);
// expire time of stream load record in rocksdb.
DECLARE_Int32(stream_load_record_expire_time_secs);
// time interval to clean expired stream load records
DECLARE_mInt64(clean_stream_load_record_interval_secs);
// enable stream load commit txn on BE directly, bypassing FE. Only for cloud.
DECLARE_mBool(enable_stream_load_commit_txn_on_be);
// The buffer size to store stream table function schema info
DECLARE_Int64(stream_tvf_buffer_size);

// request cdc client timeout
DECLARE_mInt32(request_cdc_client_timeout_ms);

// OlapTableSink sender's send interval, should be less than the real response time of a tablet writer rpc.
// You may need to lower the speed when the sink receiver bes are too busy.
DECLARE_mInt32(olap_table_sink_send_interval_microseconds);
// For auto partition, the send interval will multiply the factor
DECLARE_mDouble(olap_table_sink_send_interval_auto_partition_factor);

// Fragment thread pool
DECLARE_Int32(fragment_mgr_async_work_pool_thread_num_min);
DECLARE_Int32(fragment_mgr_async_work_pool_thread_num_max);
DECLARE_Int32(fragment_mgr_async_work_pool_queue_size);

// The read size is the size of the reads sent to os.
// There is a trade off of latency and throughout, trying to keep disks busy but
// not introduce seeks.  The literature seems to agree that with 8 MB reads, random
// io and sequential io perform similarly.
DECLARE_Int32(min_buffer_size); // 1024, The minimum read buffer size (in bytes)

// for pprof
DECLARE_String(pprof_profile_dir);
// for jeprofile in jemalloc
DECLARE_mString(jeprofile_dir);
// Purge all unused dirty pages for all arenas.
DECLARE_mBool(enable_je_purge_dirty_pages);
// Jemalloc `arenas.dirty_decay_ms`, equal to `dirty_decay_ms` in JEMALLOC_CONF in be.conf.
DECLARE_mInt32(je_dirty_decay_ms);

// to forward compatibility, will be removed later
DECLARE_mBool(enable_token_check);

// to open/close system metrics
DECLARE_Bool(enable_system_metrics);

// Number of cores Doris will used, this will effect only when it's greater than 0.
// Otherwise, Doris will use all cores returned from "/proc/cpuinfo".
DECLARE_Int32(num_cores);

// When BE start, If there is a broken disk, BE process will exit by default.
// Otherwise, we will ignore the broken disk,
DECLARE_Bool(ignore_broken_disk);
DECLARE_Bool(ignore_file_cache_dir_upgrade_failure);

// Sleep time in milliseconds between memory maintenance iterations
DECLARE_mInt32(memory_maintenance_sleep_time_ms);

// Memory gc are expensive, wait a while to avoid too frequent.
DECLARE_mInt32(memory_gc_sleep_time_ms);

// max write buffer size before flush, default 200MB
DECLARE_mInt64(write_buffer_size);
DECLARE_mBool(enable_adaptive_write_buffer_size);
// max buffer size used in memtable for the aggregated table, default 400MB
DECLARE_mInt64(write_buffer_size_for_agg);

DECLARE_mInt64(min_write_buffer_size_for_partial_update);
// max parallel flush task per memtable writer
DECLARE_mInt32(memtable_flush_running_count_limit);

// maximum sleep time to wait for memory when writing or flushing memtable.
DECLARE_mInt32(memtable_wait_for_memory_sleep_time_s);

DECLARE_Int32(load_process_max_memory_limit_percent); // 50%

// If the memory consumption of load jobs exceed load_process_max_memory_limit,
// all load jobs will hang there to wait for memtable flush. We should have a
// soft limit which can trigger the memtable flush for the load channel who
// consumes lagest memory size before we reach the hard limit. The soft limit
// might avoid all load jobs hang at the same time.
DECLARE_Int32(load_process_soft_mem_limit_percent);

// If load memory consumption is within load_process_safe_mem_permit_percent,
// memtable memory limiter will do nothing.
DECLARE_Int32(load_process_safe_mem_permit_percent);

// result buffer cancelled time (unit: second)
DECLARE_mInt32(result_buffer_cancelled_interval_time);

// arrow flight result sink buffer rows size, default 4096 * 8
DECLARE_mInt32(arrow_flight_result_sink_buffer_size_rows);
// The timeout for ADBC Client to wait for data using arrow flight reader.
// If the query is very complex and no result is generated after this time, consider increasing this timeout.
DECLARE_mInt32(arrow_flight_reader_brpc_controller_timeout_ms);

// the increased frequency of priority for remaining tasks in BlockingPriorityQueue
DECLARE_mInt32(priority_queue_remaining_tasks_increased_frequency);

// sync tablet_meta when modifying meta
DECLARE_mBool(sync_tablet_meta);

// sync a file writer when it is closed
DECLARE_mBool(sync_file_on_close);

// default thrift rpc timeout ms
DECLARE_mInt32(thrift_rpc_timeout_ms);

// txn commit rpc timeout
DECLARE_mInt32(txn_commit_rpc_timeout_ms);

// If set to true, metric calculator will run
DECLARE_Bool(enable_metric_calculator);

// max consumer num in one data consumer group, for routine load
DECLARE_mInt32(max_consumer_num_per_group);

// the max size of thread pool for routine load task.
// this should be larger than FE config 'max_routine_load_task_num_per_be' (default 5)
DECLARE_Int32(max_routine_load_thread_pool_size);

// max external scan cache batch count, means cache max_memory_cache_batch_count * batch_size row
// default is 20, batch_size's default value is 1024 means 20 * 1024 rows will be cached
DECLARE_mInt32(max_memory_sink_batch_count);

// This configuration is used for the context gc thread schedule period
// note: unit is minute, default is 5min
DECLARE_mInt32(scan_context_gc_interval_min);

// es scroll keep-alive
DECLARE_String(es_scroll_keepalive);

// HTTP connection timeout for es
DECLARE_mInt32(es_http_timeout_ms);

// the max client cache number per each host
// There are variety of client cache in BE, but currently we use the
// same cache size configuration.
// TODO(cmy): use different config to set different client cache if necessary.
DECLARE_Int32(max_client_cache_size_per_host);

DECLARE_Int32(max_master_fe_client_cache_size);

// Dir to save files downloaded by SmallFileMgr
DECLARE_String(small_file_dir);
// path gc
DECLARE_Bool(path_gc_check);
DECLARE_mInt32(path_gc_check_interval_second);
DECLARE_mInt32(path_gc_check_step);
DECLARE_mInt32(path_gc_check_step_interval_ms);

// The following 2 configs limit the max usage of disk capacity of a data dir.
// If both of these 2 threshold reached, no more data can be writen into that data dir.
// The percent of max used capacity of a data dir
DECLARE_mInt32(storage_flood_stage_usage_percent); // 90%
// The min bytes that should be left of a data dir
DECLARE_mInt64(storage_flood_stage_left_capacity_bytes); // 1GB
// number of thread for flushing memtable per store
DECLARE_mInt32(flush_thread_num_per_store);
// number of thread for flushing memtable per store, for high priority load task
DECLARE_mInt32(high_priority_flush_thread_num_per_store);
// number of threads = min(flush_thread_num_per_store * num_store,
//                         max_flush_thread_num_per_cpu * num_cpu)
DECLARE_mInt32(max_flush_thread_num_per_cpu);
// minimum flush threads per cpu when adaptive flush is enabled (default 0.5)
DECLARE_mDouble(min_flush_thread_num_per_cpu);

// Whether to enable adaptive flush thread adjustment
DECLARE_mBool(enable_adaptive_flush_threads);

// Whether to block writes when one table has too many pending flush memtables on this BE.
DECLARE_mBool(enable_table_memtable_flush_backpressure);
// Max pending flush memtables for one table on this BE before blocking new writes.
DECLARE_mInt32(table_memtable_flush_pending_count_limit);

// config for tablet meta checkpoint
DECLARE_mInt32(tablet_meta_checkpoint_min_new_rowsets_num);
DECLARE_mInt32(tablet_meta_checkpoint_min_interval_secs);
DECLARE_Int32(generate_tablet_meta_checkpoint_tasks_interval_secs);

// config for default rowset type
// Valid configs: ALPHA, BETA
DECLARE_String(default_rowset_type);

// Maximum size of a single message body in all protocols
DECLARE_Int64(brpc_max_body_size);
// Max unwritten bytes in each socket, if the limit is reached, Socket.Write fails with EOVERCROWDED
// Default, if the physical memory is less than or equal to 64G, the value is 1G
//          if the physical memory is greater than 64G, the value is physical memory * mem_limit(0.8) / 1024 * 20
DECLARE_Int64(brpc_socket_max_unwritten_bytes);
// Whether to set FLAGS_usercode_in_pthread to true in brpc
DECLARE_mBool(brpc_usercode_in_pthread);
// TODO(zxy): expect to be true in v1.3
// Whether to embed the ProtoBuf Request serialized string together with Tuple/Block data into
// Controller Attachment and send it through http brpc when the length of the Tuple/Block data
// is greater than 1.8G. This is to avoid the error of Request length overflow (2G).
DECLARE_mBool(transfer_large_data_by_brpc);

// max number of txns for every txn_partition_map in txn manager
// this is a self protection to avoid too many txns saving in manager
DECLARE_mInt64(max_runnings_transactions_per_txn_map);

// tablet_map_lock shard size, the value is 2^n, n=0,1,2,3,4
// this is a an enhancement for better performance to manage tablet
DECLARE_Int32(tablet_map_shard_size);

// txn_map_lock shard size, the value is 2^n, n=0,1,2,3,4
// this is a an enhancement for better performance to manage txn
DECLARE_Int32(txn_map_shard_size);

// txn_lock shard size, the value is 2^n, n=0,1,2,3,4
// this is a an enhancement for better performance to commit and publish txn
DECLARE_Int32(txn_shard_size);

// Whether to continue to start be when load tablet from header failed.
DECLARE_Bool(ignore_load_tablet_failure);

// Whether to continue to start be when load tablet from header failed.
DECLARE_mBool(ignore_rowset_stale_unconsistent_delete);

// Set max cache's size of query results, the unit is M byte
DECLARE_Int32(query_cache_max_size_mb);

// Cache memory is pruned when reach query_cache_max_size_mb + query_cache_elasticity_size_mb
DECLARE_Int32(query_cache_elasticity_size_mb);

// Maximum number of cache partitions corresponding to a SQL
DECLARE_Int32(query_cache_max_partition_count);

// Maximum number of version of a tablet. If the version num of a tablet exceed limit,
// the load process will reject new incoming load job of this tablet.
// This is to avoid too many version num.
DECLARE_mInt32(max_tablet_version_num);

DECLARE_mInt32(time_series_max_tablet_version_num);

// the max sleep time when meeting high pressure load task
DECLARE_mInt64(max_load_back_pressure_version_wait_time_ms);
// the threshold of rowset number gap that triggers back pressure
DECLARE_mInt64(load_back_pressure_version_threshold);

// Frontend mainly use two thrift sever type: THREAD_POOL, THREADED_SELECTOR. if fe use THREADED_SELECTOR model for thrift server,
// the thrift_server_type_of_fe should be set THREADED_SELECTOR to make be thrift client to fe constructed with TFramedTransport
DECLARE_String(thrift_server_type_of_fe);

// disable zone map index when page row is too few
DECLARE_mInt32(zone_map_row_num_threshold);

// Maximum number of IN values checked exactly against a zone map. For larger sets, only the
// IN-set min/max range is checked.
DECLARE_mInt32(in_zonemap_point_check_threshold);

// aws sdk log level
//    Off = 0,
//    Fatal = 1,
//    Error = 2,
//    Warn = 3,
//    Info = 4,
//    Debug = 5,
//    Trace = 6
DECLARE_Int32(aws_log_level);

// azure sdk log level
//    Verbose = 1,
//    Informational = 2,
//    Warning = 3,
//    Error = 4
DECLARE_Int32(azure_log_level);

// the buffer size when read data from remote storage like s3
DECLARE_mInt32(remote_storage_read_buffer_mb);

// The minimum length when TCMalloc Hook consumes/releases MemTracker, consume size
// smaller than this value will continue to accumulate. specified as number of bytes.
// Decreasing this value will increase the frequency of consume/release.
// Increasing this value will cause MemTracker statistics to be inaccurate.
DECLARE_mInt32(mem_tracker_consume_min_size_bytes);

// The version information of the tablet will be stored in the memory
// in an adjacency graph data structure.
// And as the new version is written and the old version is deleted,
// the data structure will begin to have empty vertex with no edge associations(orphan vertex).
// This config is used to control that when the proportion of orphan vertex is greater than the threshold,
// the adjacency graph will be rebuilt to ensure that the data structure will not expand indefinitely.
// This config usually only needs to be modified during testing.
// In most cases, it does not need to be modified.
DECLARE_mDouble(tablet_version_graph_orphan_vertex_ratio);

// share delta writers when memtable_on_sink_node = true
DECLARE_Bool(share_delta_writers);
// timeout for open load stream rpc in ms
DECLARE_Int64(open_load_stream_timeout_ms);
// enable write background when using brpc stream
DECLARE_mBool(enable_brpc_stream_write_background);

// brpc streaming max_buf_size in bytes
DECLARE_Int64(load_stream_max_buf_size);
// brpc streaming messages_in_batch
DECLARE_Int32(load_stream_messages_in_batch);
// brpc streaming StreamWait seconds on EAGAIN
DECLARE_Int32(load_stream_eagain_wait_seconds);
// max tasks per flush token in load stream
DECLARE_Int32(load_stream_flush_token_max_tasks);
// max wait flush token time in load stream
DECLARE_Int32(load_stream_max_wait_flush_token_time_ms);
// number of send batch thread pool size
DECLARE_Int32(send_batch_thread_pool_thread_num);
// number of send batch thread pool queue size
DECLARE_Int32(send_batch_thread_pool_queue_size);

// Limit the number of segment of a newly created rowset.
// The newly created rowset may to be compacted after loading,
// so if there are too many segment in a rowset, the compaction process
// will run out of memory.
// When doing compaction, each segment may take at least 1MB buffer.
DECLARE_mInt32(max_segment_num_per_rowset);

// Store segment without compression if a segment is smaller than
// segment_compression_threshold_kb.
DECLARE_mInt32(segment_compression_threshold_kb);

// Global bitmap cache capacity for aggregation cache, size in bytes
DECLARE_Int64(delete_bitmap_agg_cache_capacity);
DECLARE_String(delete_bitmap_dynamic_agg_cache_limit);
DECLARE_mInt32(delete_bitmap_agg_cache_stale_sweep_time_sec);

// A common object cache depends on an Sharded LRU Cache.
DECLARE_mInt32(common_obj_lru_cache_stale_sweep_time_sec);

// reference https://github.com/edenhill/librdkafka/blob/master/INTRODUCTION.md#broker-version-compatibility
// If the dependent kafka broker version older than 0.10.0.0,
// the value of kafka_api_version_request should be false, and the
// value set by the fallback version kafka_broker_version_fallback will be used,
// and the valid values are: 0.9.0.x, 0.8.x.y.
DECLARE_String(kafka_api_version_request);
DECLARE_String(kafka_broker_version_fallback);
DECLARE_mString(kafka_debug);

// The number of pool siz of routine load consumer.
// If you meet the error describe in https://github.com/edenhill/librdkafka/issues/3608
// Change this size to 0 to fix it temporarily.
DECLARE_mInt32(routine_load_consumer_pool_size);

// the timeout of condition variable wait in blocking_get and blocking_put
DECLARE_mInt32(blocking_queue_cv_wait_timeout_ms);

// Used in single-stream-multi-table load. When receive a batch of messages from kafka,
// if the size of batch is more than this threshold, we will request plans for all related tables.
DECLARE_Int32(multi_table_batch_plan_threshold);

// Used in single-stream-multi-table load. When receiving a batch of messages from Kafka,
// if the size of the table wait for plan is more than this threshold, we will request plans for all related tables.
// The param is aimed to avoid requesting and executing too many plans at once.
// Performing small batch processing on multiple tables during the loaded process can reduce the pressure of a single RPC
// and improve the real-time processing of data.
DECLARE_Int32(multi_table_max_wait_tables);

// When the timeout of a load task is less than this threshold,
// Doris treats it as a high priority task.
// high priority tasks use a separate thread pool for flush and do not block rpc by memory cleanup logic.
// this threshold is mainly used to identify routine load tasks and should not be modified if not necessary.
DECLARE_mInt32(load_task_high_priority_threshold_second);

// The min timeout of load rpc (add batch, close)
// Because a load rpc may be blocked for a while.
// Increase this config may avoid rpc timeout.
DECLARE_mInt32(min_load_rpc_timeout_ms);

// use which protocol to access function service, candicate is baidu_std/h2:grpc
DECLARE_String(function_service_protocol);

// use which load balancer to select server to connect
DECLARE_String(rpc_load_balancer);

// a soft limit of string type length, the hard limit is 2GB - 4, but if too long will cause very low performance,
// so we set a soft limit, default is 1MB
DECLARE_mInt32(string_type_length_soft_limit_bytes);

// Threshold fo reading a small file into memory
DECLARE_mInt32(in_memory_file_size);

// Max size of parquet page header in bytes
DECLARE_mInt32(parquet_header_max_size_mb);
// Max size of parquet file metadata in bytes
DECLARE_mInt64(parquet_metadata_size_limit);
// Max buffer size for parquet row group
DECLARE_mInt32(parquet_rowgroup_max_buffer_mb);
// Max buffer size for parquet chunk column
DECLARE_mInt32(parquet_column_max_buffer_mb);
// Merge small IO, the max amplified read ratio
DECLARE_mDouble(max_amplified_read_ratio);
// Equivalent min size of each IO that can reach the maximum storage speed limit
// 1MB for oss, 8KB for hdfs
DECLARE_mInt32(merged_oss_min_io_size);
DECLARE_mInt32(merged_hdfs_min_io_size);

// OrcReader
DECLARE_mInt32(orc_natural_read_size_mb);

DECLARE_mInt32(runtime_filter_sampling_frequency);
DECLARE_mInt32(execution_max_rpc_timeout_sec);
DECLARE_mBool(execution_ignore_eovercrowded);

// cooldown task configs
DECLARE_Int32(cooldown_thread_num);
DECLARE_mInt64(generate_cooldown_task_interval_sec);
DECLARE_mInt32(remove_unused_remote_files_interval_sec); // 6h
DECLARE_mInt32(confirm_unused_remote_files_interval_sec);
DECLARE_Int32(cold_data_compaction_thread_num);
DECLARE_mInt32(cold_data_compaction_interval_sec);
DECLARE_mInt32(cold_data_compaction_score_threshold);

DECLARE_Int32(min_s3_file_system_thread_num);
DECLARE_Int32(max_s3_file_system_thread_num);

// Thread pool for S3 reads in cross-CG peer winner race.
// Max should match max_concurrent_peer_races so the pool never fills up under normal operation.
DECLARE_Int32(min_peer_race_s3_thread_num);
DECLARE_Int32(max_peer_race_s3_thread_num);

DECLARE_Bool(enable_time_lut);

DECLARE_mBool(enable_query_like_bloom_filter);
// number of s3 scanner thread pool size
DECLARE_Int32(doris_remote_scanner_thread_pool_thread_num);
// number of s3 scanner thread pool queue size
DECLARE_Int32(doris_remote_scanner_thread_pool_queue_size);

// limit the queue of pending batches which will be sent by a single nodechannel
DECLARE_mInt64(nodechannel_pending_queue_max_bytes);

// The batch size for sending data by brpc streaming client
DECLARE_mInt64(brpc_streaming_client_batch_bytes);
DECLARE_mInt64(block_cache_wait_timeout_ms);

DECLARE_Bool(enable_brpc_builtin_services);

DECLARE_Bool(enable_brpc_connection_check);

DECLARE_mInt64(brpc_connection_check_timeout_ms);

DECLARE_Int32(fragment_mgr_cancel_worker_interval_seconds);

// Node role tag for backend. Mix role is the default role, and computation role have no
// any tablet.
DECLARE_String(be_node_role);

// Hide webserver page for safety.
// Hide the be config page for webserver.
DECLARE_Bool(hide_webserver_config_page);

DECLARE_Bool(enable_segcompaction);

// Max number of segments allowed in a single segcompaction task.
DECLARE_mInt32(segcompaction_batch_size);

// Max row count allowed in a single source segment, bigger segments will be skipped.
DECLARE_Int32(segcompaction_candidate_max_rows);

// Max file size allowed in a single source segment, bigger segments will be skipped.
DECLARE_Int64(segcompaction_candidate_max_bytes);

// Max total row count allowed in a single segcompaction task.
DECLARE_Int32(segcompaction_task_max_rows);

// Max total file size allowed in a single segcompaction task.
DECLARE_Int64(segcompaction_task_max_bytes);

// Global segcompaction thread pool size.
DECLARE_mInt32(segcompaction_num_threads);

// enable java udf and jdbc scannode
DECLARE_Bool(enable_java_support);

// enable python udf
DECLARE_Bool(enable_python_udf_support);
// python env mode, options: conda, venv
DECLARE_String(python_env_mode);
// root path of conda runtime, python_env_mode should be conda
DECLARE_String(python_conda_root_path);
// root path of venv runtime, python_env_mode should be venv
DECLARE_String(python_venv_root_path);
// python interpreter paths used by venv, e.g. /usr/bin/python3.7:/usr/bin/python3.6
DECLARE_String(python_venv_interpreter_paths);
// max python processes in global shared pool, each version can have up to this many processes
DECLARE_mInt32(max_python_process_num);
// Memory limit in bytes for all Python UDF processes; warning is logged when exceeded
DECLARE_mInt64(python_udf_processes_memory_limit_bytes);

// Set config randomly to check more issues in github workflow
DECLARE_Bool(enable_fuzzy_mode);

DECLARE_Bool(enable_graceful_exit_check);

DECLARE_Bool(enable_debug_points);

DECLARE_Int32(pipeline_executor_size);
DECLARE_Int32(blocking_pipeline_executor_size);

// block file cache
DECLARE_Bool(enable_file_cache);
DECLARE_mBool(enable_file_cache_write_from_s3_file_writer);
// format: [{"path":"/path/to/file_cache","total_size":21474836480,"query_limit":10737418240}]
// format: [{"path":"/path/to/file_cache","total_size":21474836480,"query_limit":10737418240},{"path":"/path/to/file_cache2","total_size":21474836480,"query_limit":10737418240}]
// format: [{"path":"/path/to/file_cache","total_size":21474836480,"query_limit":10737418240, "ttl_percent":50, "normal_percent":40, "disposable_percent":5, "index_percent":5}]
// format: [{"path": "xxx", "total_size":53687091200, "storage": "memory"}]
// Note1: storage is "disk" by default
// Note2: when the storage is "memory", the path is ignored. So you can set xxx to anything you like
// and doris will just reset the path to "memory" internally.
// In a very wierd case when your storage is disk, and the directory, by accident, is named
// "memory" for some reason, you should write the path as:
//     {"path": "memory", "total_size":53687091200, "storage": "disk"}
// or use the default storage value:
//     {"path": "memory", "total_size":53687091200}
// Both will use the directory "memory" on the disk instead of the real RAM.
DECLARE_String(file_cache_path);
DECLARE_Int64(file_cache_each_block_size);
DECLARE_Bool(clear_file_cache);
DECLARE_mBool(enable_file_cache_query_limit);
DECLARE_mBool(enable_file_cache_query_limit_segment_meta);
DECLARE_Int32(file_cache_enter_disk_resource_limit_mode_percent);
DECLARE_Int32(file_cache_exit_disk_resource_limit_mode_percent);
DECLARE_mBool(enable_evict_file_cache_in_advance);
DECLARE_mInt32(file_cache_enter_need_evict_cache_in_advance_percent);
DECLARE_mInt32(file_cache_exit_need_evict_cache_in_advance_percent);
DECLARE_mInt32(file_cache_evict_in_advance_interval_ms);
DECLARE_mInt64(file_cache_evict_in_advance_batch_bytes);
DECLARE_mInt64(file_cache_evict_in_advance_recycle_keys_num_threshold);
DECLARE_mBool(enable_read_cache_file_directly);
DECLARE_Bool(file_cache_enable_evict_from_other_queue_by_size);
DECLARE_mBool(enbale_dump_error_file);
// limit the max size of error log on disk
DECLARE_mInt64(file_cache_error_log_limit_bytes);
DECLARE_mInt64(cache_lock_wait_long_tail_threshold_us);
DECLARE_mInt64(cache_lock_held_long_tail_threshold_us);
// Base compaction may retrieve and produce some less frequently accessed data,
// potentially affecting the file cache hit rate.
// This configuration determines whether to retain the output within the file cache.
// Make your choice based on the following considerations:
// If your file cache is ample enough to accommodate all the data in your database,
// enable this option; otherwise, it is recommended to leave it disabled.
DECLARE_mBool(enable_file_cache_keep_base_compaction_output);
DECLARE_mBool(enable_file_cache_adaptive_write);
DECLARE_mDouble(file_cache_keep_base_compaction_output_min_hit_ratio);
DECLARE_mDouble(file_cache_meta_store_vs_file_system_diff_num_threshold);
DECLARE_mDouble(file_cache_keep_schema_change_output_min_hit_ratio);
DECLARE_mDouble(file_cache_leak_fs_to_meta_ratio_threshold);
DECLARE_mInt64(file_cache_leak_scan_interval_seconds);
DECLARE_mInt32(file_cache_leak_scan_batch_files);
DECLARE_mInt32(file_cache_leak_scan_pause_ms);
DECLARE_mInt64(file_cache_leak_grace_seconds);
DECLARE_mInt64(file_cache_remove_block_qps_limit);
DECLARE_mInt64(file_cache_background_gc_interval_ms);
DECLARE_mInt64(file_cache_background_block_lru_update_interval_ms);
DECLARE_mInt64(file_cache_background_block_lru_update_qps_limit);
DECLARE_mInt64(file_cache_background_block_lru_update_queue_max_size);
DECLARE_mBool(enable_file_cache_async_touch_on_get_or_set);
DECLARE_mBool(enable_reader_dryrun_when_download_file_cache);
DECLARE_mInt64(file_cache_background_monitor_interval_ms);
DECLARE_mInt64(file_cache_background_ttl_gc_interval_ms);
DECLARE_mInt64(file_cache_background_ttl_info_update_interval_ms);
DECLARE_mInt64(file_cache_background_tablet_id_flush_interval_ms);
DECLARE_Int32(file_cache_downloader_thread_num_min);
DECLARE_Int32(file_cache_downloader_thread_num_max);
// used to persist lru information before be reboot and load the info back
DECLARE_mInt64(file_cache_background_lru_dump_interval_ms);
// dump queue only if the queue update specific times through several dump intervals
DECLARE_mInt64(file_cache_background_lru_dump_update_cnt_threshold);
DECLARE_mInt64(file_cache_background_lru_dump_tail_record_num);
DECLARE_mInt64(file_cache_background_lru_log_queue_max_size);
DECLARE_mInt64(file_cache_background_lru_log_replay_interval_ms);
DECLARE_mBool(enable_evaluate_shadow_queue_diff);

DECLARE_mBool(file_cache_enable_only_warm_up_idx);

// async file cache write
DECLARE_mBool(enable_async_file_cache_write);
DECLARE_mInt32(async_file_cache_write_workers_per_disk);
DECLARE_mInt64(async_file_cache_write_max_pending_bytes);
DECLARE_mBool(enable_async_file_cache_write_inflight_write_buffer_index);
DECLARE_Int32(async_file_cache_write_inflight_write_buffer_index_shard_count);

// inverted index searcher cache
// cache entry stay time after lookup
DECLARE_mInt32(index_cache_entry_stay_time_after_lookup_s);
// cache entry that have not been visited for a certain period of time can be cleaned up by GC thread
DECLARE_mInt32(inverted_index_cache_stale_sweep_time_sec);
// inverted index searcher cache size
DECLARE_String(inverted_index_searcher_cache_limit);
DECLARE_mBool(enable_write_index_searcher_cache);
DECLARE_Bool(enable_inverted_index_cache_check_timestamp);
DECLARE_mBool(enable_inverted_index_correct_term_write);
DECLARE_Int32(inverted_index_fd_number_limit_percent); // 50%
DECLARE_Int32(inverted_index_query_cache_shards);
// When the candidate row bitmap of a segment scan is smaller than
// num_rows * this ratio, it is pushed down into inverted index queries so
// doc-list intersection and verification run only over the candidates
// (see IndexQueryContext::candidate_rows). <= 0 disables the pushdown.
DECLARE_mDouble(inverted_index_candidate_pushdown_ratio);
double get_inverted_index_candidate_pushdown_ratio();

// inverted index match bitmap cache size
DECLARE_String(inverted_index_query_cache_limit);

// Build-only CommonGrams kill switch. Logical index writers snapshot it at construction; changing
// it affects only writers created after the transition and never changes query/cache semantics.
// Release-calibrated query-planner coefficients. Both remain mutable for controlled recalibration.

// Whether LIKE/REGEXP tries to compile a constant pattern into a gram boolean query pushed down
// to a gram-family inverted index (master switch). Turning it off behaves as if the index did
// not exist -- it only gives up the speedup, it never changes query results.
//
// A pattern is compiled against the gram scheme the segment itself carries, read back from the
// core metadata of the physical index the query is about to read, so the scheme can never
// disagree with the one the writer used -- a policy that was dropped and recreated with different
// properties does not affect segments already written.
DECLARE_mBool(enable_gram_index_regexp);

// Cost gate for the gram boolean query, in basis points (1/10000) of a segment's rows: give up
// pruning once the candidate set is larger than that share. 0 disables the gate, 10000 or more can
// never fire. This is a COST switch, never a semantic one -- a gram index only produces a superset
// of the candidate rows and the expression above re-verifies each of them, so giving up means the
// query returns the whole docid range of the segment and prunes nothing. It can therefore only
// cost speed, never correctness, whatever value it is set to.
//
// The unit is basis points rather than percent because the useful range is far below one percent.
// Pruning only pays when the surviving candidates are sparse enough that whole pages of the column
// can be skipped: with B rows per remote read unit and candidates spread uniformly, the fraction of
// units still touched is 1 - (1 - C/N)^B, which is already ~1 by C/N = 1%. Measured on a 34.46M row
// httplogs table against remote object storage, cold-read speedup by candidate count was 14.1x at
// 2 candidates, 5.5x at 128, 1.5x at 25K, 1.2x at 46K, and turned into a regression from 56K
// upwards, bottoming out at 0.70x. 15 bp (0.15%) is the largest share that admitted no regression
// in that sweep while keeping every double-digit win.
DECLARE_mInt32(gram_index_max_candidate_ratio_bp);

// Row floor below which the candidate ratio gate above is not applied at all. A small segment's
// entire gram index is a few KB and one or two requests, so giving up there saves nothing
// measurable while throwing away the pruning the index really does deliver; the ratio only starts
// to mean something at a size where the skipped index IO can outweigh the rows it stops
// eliminating. The default is one Roaring container's worth of rows (65536). 0 applies the ratio
// at every segment size.
DECLARE_mInt32(gram_index_candidate_ratio_min_rows);

// Adaptive gram density: solve the boundary rate from each segment's own bytes instead of
// taking the rate configured on the tokenizer.
//
// A configured rate cannot mean the same thing on two columns. Measured at a nominal 0.25 the
// realised grams per byte were 0.204 on log text, 0.287 on URL paths and 0.182 on agent
// traces, and the coverage it bought ranged from 91.7% of 12-byte literals to 98.9% -- under
// serving one column while overpaying on another, in every posting list. Solving instead
// makes the promise the constant and the rate the variable.
DECLARE_mBool(enable_gram_index_adaptive_density);
// The promise the solve keeps, not a tuning pair: literals of at least this many bytes are
// findable, for this share of the column's own windows of that length. Both are dimensionless
// and the same on every dataset; what varies is the density they resolve to.
DECLARE_mInt32(gram_index_min_literal_bytes);
DECLARE_mInt32(gram_index_density_coverage_permille);
// How much of a segment is held back to solve on. The sample is buffered rather than
// tokenized, so this is a transient memory cost and a bound on how long the write path waits
// before it can cut anything; the histogram behind the solve is a fixed 256 KB regardless.
DECLARE_mInt64(gram_index_density_sample_bytes);

// condition cache limit
DECLARE_Int16(condition_cache_limit);

// ANN index topn result cache
DECLARE_String(ann_index_result_cache_limit);
DECLARE_Int32(ann_index_result_cache_shards);
DECLARE_Int32(ann_index_result_cache_stale_sweep_time_sec);

// inverted index
DECLARE_mDouble(inverted_index_ram_buffer_size);
DECLARE_mInt32(inverted_index_max_buffered_docs);
// G16-h: zstd levels for SNII dict blocks / prx windows. Default 3 (the
// all-level-3 evaluation showed level 9 buys <=6.3% index size for 17-24%
// import CPU; see the DEFINEs in config.cpp).
DECLARE_mInt32(snii_dict_block_zstd_level);
DECLARE_mInt32(snii_prx_zstd_level);
// Patch C: prx zstd level for DIRECT-LOAD segments only (default 3, cheaper
// import); compaction rewrites at snii_prx_zstd_level so settled segments are
// unaffected. Full contract at the DEFINE in config.cpp.
DECLARE_mInt32(snii_prx_zstd_level_direct_load);
// G16-d: target SNII dict block size in bytes; 0 = format default (64 KiB).
// Bigger blocks -> better per-block zstd on the dict region, larger cold
// fetch+decompress unit per dict-block miss. Write side only.
DECLARE_mInt32(snii_target_dict_block_bytes);
// PROCESS-WIDE share for SNII index-build RAM, as a PERCENT of the process
// memory limit -- the index-build analogue of
// load_process_max_memory_limit_percent. The per-writer
// inverted_index_ram_buffer_size is a reclaimable-buffer spill threshold, not a
// hard cap on persistent vocabulary bytes: a concurrent load keeps (tablets x
// concurrency) writers alive at once, none of which may reach that threshold,
// while their SUM can still be large. Once live SNII index-build memory
// (ingestion plus index-merge compaction) crosses this share, the writers
// holding the largest reclaimable posting arenas are asked to spill early
// (async-safe advisory requests, honored on each writer's own thread; output
// stays byte-identical). Read at every decision, so a change takes effect
// immediately for writers that are already running.
//
// 0 disables SNII's own share trigger; the process-level backstops (system
// available memory below its warning water mark, process usage above the soft
// limit) still apply.
//
// FLOORED AGAINST inverted_index_ram_buffer_size: the share is never less than
// four writers' worth of the per-writer spill threshold. A smaller share would
// put a small BE permanently over it as soon as two writers exist -- unrelievable
// back-pressure rather than a limit -- because the per-writer threshold is what
// one writer may hold before it spills on its own.
DECLARE_mInt32(snii_index_build_max_memory_limit_percent);
// G09 forced-spill floor: minimum reclaimable posting-arena bytes a SNII
// writer must hold before a process-wide forced-spill request is honored, and
// before the global limiter selects it as a spill victim. A forced spill
// reclaims ONLY the posting arena -- the persistent vocab / pair-map
// structures survive it -- so honoring below a real floor degenerates into a
// storm of tiny runs whenever the memory over the share is dominated by
// persistent bytes (each run still adds framing, sorting and merge work for
// near-zero memory relief). THIS FLOOR, not any judgement about whether the
// overage is reachable, is what bounds forced spilling: it caps the cost at one
// >= floor-sized run per floor of arena growth per writer. Forced spilling
// therefore reclaims SPILLABLE memory only, never persistent memory.
// Default 64 MiB.
DECLARE_mInt64(snii_forced_spill_min_arena_bytes);
// Historical run-file cap, now an additional limit on active inputs per merge
// group (minimum two). Ingestion runs share one append-only spool. Zero leaves
// fan-in bounded by the postings workspace and fd limits. Default 64.
DECLARE_mInt32(snii_spill_max_run_files_per_buffer);
// Shared hard posting-workspace budget, captured by new ingestion/compaction
// reporters. Positive bytes; default 32 MiB. High ZSTD levels may need more.
DECLARE_mInt64(snii_postings_workspace_bytes);
// dict path for chinese analyzer
DECLARE_String(inverted_index_dict_path);
// The kuromoji (Japanese) analyzer
DECLARE_mBool(enable_kuromoji_analyzer);
DECLARE_Int32(inverted_index_read_buffer_size);
// tree depth for bkd index
DECLARE_Int32(max_depth_in_bkd_tree);
// index compaction
DECLARE_mBool(inverted_index_compaction_enable);
// Only for debug, do not use in production
DECLARE_mBool(debug_inverted_index_compaction);
// index by RAM directory
DECLARE_mBool(inverted_index_ram_dir_enable);
// wheather index by RAM directory when base compaction
DECLARE_mBool(inverted_index_ram_dir_enable_when_base_compaction);
// Norms cost one byte per segment row, including rows that hold no value for the field. A segment
// holds one index per variant path, so writing norms for them costs rows * paths bytes. Turn this on
// to leave norms out of every index on a variant path, whatever its "norms" property says; BM25
// scoring (score()) on those indexes then fails.
DECLARE_mBool(inverted_index_skip_norms_for_variant);
// use num_broadcast_buffer blocks as buffer to do broadcast
DECLARE_Int32(num_broadcast_buffer);

// max depth of expression tree allowed.
DECLARE_Int32(max_depth_of_expr_tree);

// Report a tablet as bad when io errors occurs more than this value.
DECLARE_mInt64(max_tablet_io_errors);

// Report a tablet as bad when its path not found
DECLARE_Int32(tablet_path_check_interval_seconds);
DECLARE_mInt32(tablet_path_check_batch_size);

// it must be larger than or equal to 5MB
DECLARE_mInt64(s3_write_buffer_size);
// Log interval when doing s3 upload task
DECLARE_mInt32(s3_file_writer_log_interval_second);
// the max number of cached file handle for block segemnt
DECLARE_mInt64(file_cache_max_file_reader_cache_size);
DECLARE_mInt64(hdfs_write_batch_buffer_size_mb);
//enable shrink memory
DECLARE_mBool(enable_shrink_memory);

// max number of segment cache
DECLARE_Int32(segment_cache_capacity);
DECLARE_Int32(segment_cache_fd_percentage);
DECLARE_Int32(segment_cache_memory_percentage);
DECLARE_Bool(enable_segment_cache_prune);

DECLARE_mInt32(estimated_mem_per_column_reader);

// enable binlog
DECLARE_Bool(enable_feature_binlog);

// enable set in BitmapValue
DECLARE_Bool(enable_set_in_bitmap_value);

// Enable compact integer tags in row-store JSONB. Once enabled and compact data is written,
// rollback to code without compact row-store JSONB reader support is not safe.
DECLARE_Bool(enable_row_store_compact_jsonb);

// max number of hdfs file handle in cache
DECLARE_Int64(max_hdfs_file_handle_cache_num);
DECLARE_Int32(max_hdfs_file_handle_cache_time_sec);

// max number of meta info of external files, such as parquet footer
DECLARE_Int64(max_external_file_meta_cache_num);
// Apply delete pred in cumu compaction
DECLARE_mBool(enable_delete_when_cumu_compaction);

// max_write_buffer_number for rocksdb
DECLARE_Int32(rocksdb_max_write_buffer_number);

// Convert date 0000-00-00 to 0000-01-01. It's recommended to set to false.
DECLARE_mBool(allow_zero_date);
// Allow invalid decimalv2 literal for compatible with old version. Recommend set it false strongly.
DECLARE_mBool(allow_invalid_decimalv2_literal);
// Allow to specify kerberos credentials cache path.
DECLARE_mString(kerberos_ccache_path);
// set krb5.conf path, use "/etc/krb5.conf" by default
DECLARE_mString(kerberos_krb5_conf_path);

// JDK-8153057: avoid StackOverflowError thrown from the UncaughtExceptionHandler in thread "process reaper"
DECLARE_mBool(jdk_process_reaper_use_default_stack_size);

DECLARE_mBool(enable_address_sanitizers_with_stack_trace);

// DISABLED: Don't resolve location info.
// FAST: Perform CU lookup using .debug_aranges (might be incomplete).
// FULL: Scan all CU in .debug_info (slow!) on .debug_aranges lookup failure.
// FULL_WITH_INLINE: Scan .debug_info (super slower, use with caution) for inline functions in addition to FULL.
DECLARE_mString(dwarf_location_info_mode);

// the ratio of _prefetch_size/_batch_size in AutoIncIDBuffer
DECLARE_mInt64(auto_inc_prefetch_size_ratio);

// the ratio of _low_level_water_level_mark/_batch_size in AutoIncIDBuffer
DECLARE_mInt64(auto_inc_low_water_level_mark_size_ratio);

// number of threads that fetch auto-inc ranges from FE
DECLARE_mInt64(auto_inc_fetch_thread_num);
// Max connection cache num for point lookup queries
DECLARE_mInt64(lookup_connection_cache_capacity);
// Number of reusable result blocks retained by each point lookup cache entry, in [1, 32].
DECLARE_mInt32(lookup_connection_cache_block_pool_size);
size_t get_lookup_connection_cache_block_pool_size();

// level of compression when using LZ4_HC, whose defalut value is LZ4HC_CLEVEL_DEFAULT
DECLARE_mInt64(LZ4_HC_compression_level);
// Max json key length in bytes when parsing json into variant subcolumns/jsonb.
DECLARE_mInt32(variant_max_json_key_length);
// Treat invalid json format str as string, instead of throwing exception if false
DECLARE_mBool(variant_throw_exeception_on_invalid_json);
// Enable duplicate path check when parsing json into variant subcolumns/jsonb.
DECLARE_mBool(variant_enable_duplicate_json_path_check);
// Controls storage-layer parse target for plain non-doc VARIANT columns:
// 0 = auto, 1 = force parse-time subcolumns, 2 = force doc-value KV staging.
DECLARE_mInt32(variant_storage_parse_mode);
// Enable vertical compact subcolumns of variant column
DECLARE_mBool(enable_vertical_compact_variant_subcolumns);
DECLARE_mBool(enable_variant_doc_sparse_write_subcolumns);
// When true, discard scalar data that conflicts with NestedGroup array<object>
// data at the same path. This simplifies compaction by always prioritizing
// nested structure over scalar. When false, report an error on conflict.
DECLARE_mBool(variant_nested_group_discard_scalar_on_conflict);

DECLARE_mBool(enable_merge_on_write_correctness_check);
// USED FOR DEBUGING
// core directly if the compaction found there's duplicate key on mow table
DECLARE_mBool(enable_mow_compaction_correctness_check_core);
// USED FOR DEBUGING
// let compaction fail if the compaction found there's duplicate key on mow table
DECLARE_mBool(enable_mow_compaction_correctness_check_fail);
// rowid conversion correctness check when compaction for mow table
DECLARE_mBool(enable_rowid_conversion_correctness_check);
// missing rows correctness check when compaction for mow table
DECLARE_mBool(enable_missing_rows_correctness_check);
// When the number of missing versions is more than this value, do not directly
// retry the publish and handle it through async publish.
DECLARE_mInt32(mow_publish_max_discontinuous_version_num);
// When the size of primary keys in memory exceeds this value, finish current segment
// and create a new segment, used in compaction.
DECLARE_mInt64(mow_primary_key_index_max_size_in_memory);
// When the version is not continuous for MOW table in publish phase and the gap between
// current txn's publishing version and the max version of the tablet exceeds this value,
// don't print warning log
DECLARE_mInt32(publish_version_gap_logging_threshold);
// get agg by cache for mow table
DECLARE_mBool(enable_mow_get_agg_by_cache);
// get agg correctness check for mow table
DECLARE_mBool(enable_mow_get_agg_correctness_check_core);
DECLARE_mBool(enable_agg_and_remove_pre_rowsets_delete_bitmap);
DECLARE_mBool(enable_check_agg_and_remove_pre_rowsets_delete_bitmap);
DECLARE_mBool(enable_remove_agg_pre_rowsets_delete_bitmap_by_keys);
DECLARE_mBool(enable_remove_pre_rowsets_delete_bitmap_by_keys);

// The secure path with user files, used in the `local` table function.
DECLARE_String(user_files_secure_path);

// If fe's frontend info has not been updated for more than fe_expire_duration_seconds, it will be regarded
// as an abnormal fe, this will cause be to cancel this fe's related query.
DECLARE_Int32(fe_expire_duration_seconds);

// If use stop_be.sh --grace, then BE has to wait all running queries to stop to avoiding running query failure
// , but if the waiting time exceed the limit, then be will exit directly.
// During this period, FE will not send any queries to BE and waiting for all running queries to stop.
DECLARE_Int32(grace_shutdown_wait_seconds);
// When using the graceful stop feature, after the main process waits for
// all currently running tasks to finish, it will continue to wait for
// an additional period to ensure that queries still running on other nodes have also completed.
// Since a BE node cannot detect the task execution status on other BE nodes,
// you may need to increase this threshold to allow for a longer waiting time.
DECLARE_Int32(grace_shutdown_post_delay_seconds);

// BitmapValue serialize version.
DECLARE_Int16(bitmap_serialize_version);

// group commit config
DECLARE_String(group_commit_wal_path);
DECLARE_Int32(group_commit_replay_wal_retry_num);
DECLARE_Int32(group_commit_replay_wal_retry_interval_seconds);
DECLARE_Int32(group_commit_replay_wal_retry_interval_max_seconds);
DECLARE_mInt32(group_commit_relay_wal_threads);
// This config can be set to limit thread number in group commit request fragment thread pool.
DECLARE_mInt32(group_commit_insert_threads);
DECLARE_mInt32(group_commit_memory_rows_for_max_filter_ratio);
DECLARE_Bool(wait_internal_group_commit_finish);
// Max size(bytes) of group commit queues, used for mem back pressure.
DECLARE_mInt32(group_commit_queue_mem_limit);
// Max size(bytes) or percentage(%) of wal disk usage, used for disk space back pressure, default 10% of the disk available space.
// group_commit_wal_max_disk_limit=1024 or group_commit_wal_max_disk_limit=10% can be automatically identified.
DECLARE_mString(group_commit_wal_max_disk_limit);
DECLARE_Bool(group_commit_wait_replay_wal_finish);
// Max WAL count for one table before rejecting async group commit loads. 0 means no limit.
DECLARE_mInt32(group_commit_max_wal_num_per_table);
// Max time(ms) to wait for creating group commit plan fragment. 0 means no timeout.
DECLARE_mInt32(group_commit_create_plan_timeout_ms);

// The configuration item is used to lower the priority of the scanner thread,
// typically employed to ensure CPU scheduling for write operations.
// Default is 0, which is default value of thread nice value, increase this value
// to lower the priority of scan threads
DECLARE_Int32(scan_thread_nice_value);
// Used to modify the recycle interval of tablet schema cache
DECLARE_mInt32(tablet_schema_cache_recycle_interval);
// Granularity is at the column level
DECLARE_mInt32(tablet_schema_cache_capacity);

// Use `LOG(FATAL)` to replace `throw` when true
DECLARE_mBool(exit_on_exception);

// cgroup
DECLARE_String(doris_cgroup_cpu_path);
DECLARE_mBool(enable_be_proc_monitor);
DECLARE_mInt32(be_proc_monitor_interval_ms);
DECLARE_Int32(workload_group_metrics_interval_ms);
DECLARE_Int32(workload_policy_check_interval_ms);

// This config controls whether the s3 file writer would flush cache asynchronously
DECLARE_Bool(enable_flush_file_cache_async);

// Ingest binlog work pool size
DECLARE_Int32(ingest_binlog_work_pool_size);

// Ingest binlog with persistent connection
DECLARE_Bool(enable_ingest_binlog_with_persistent_connection);

// Log ingest binlog elapsed threshold, -1 means no log
DECLARE_mInt64(ingest_binlog_elapsed_threshold_ms);

// Download binlog rate limit, unit is KB/s
DECLARE_Int32(download_binlog_rate_limit_kbs);

DECLARE_mInt32(buffered_reader_read_timeout_ms);

// whether to enable /api/snapshot api
DECLARE_Bool(enable_snapshot_action);

// The max columns size for a tablet schema
DECLARE_mInt32(variant_max_merged_tablet_schema_size);

DECLARE_mInt64(local_exchange_buffer_mem_limit);

DECLARE_mBool(enable_column_type_check);

// Tolerance for the number of partition id 0 in rowset, default 0
DECLARE_Int32(ignore_invalid_partition_id_rowset_num);

DECLARE_mInt32(report_query_statistics_interval_ms);
DECLARE_mInt32(query_statistics_reserve_timeout_ms);
DECLARE_mInt32(report_exec_status_thread_num);

// consider two high usage disk at the same available level if they do not exceed this diff.
DECLARE_mDouble(high_disk_avail_level_diff_usages);

// create tablet in partition random robin idx lru size, default 10000
DECLARE_Int32(partition_disk_index_lru_size);
DECLARE_String(spill_storage_root_path);
// Spill storage limit specified as number of bytes
// ('<int>[bB]?'), megabytes ('<float>[mM]'), gigabytes ('<float>[gG]'),
// or percentage of capaity ('<int>%').
// Defaults to bytes if no unit is given.
// Must larger than 0.
// If specified as percentage, the final limit value is:
//   disk_capacity_bytes * storage_flood_stage_usage_percent * spill_storage_limit
DECLARE_String(spill_storage_limit);
DECLARE_mInt32(spill_gc_interval_ms);
DECLARE_mInt32(spill_gc_work_time_ms);
// Maximum size of each spill part file before rotation (bytes). Default 1GB.
DECLARE_mInt64(spill_file_part_size_bytes);
DECLARE_Int64(spill_in_paused_queue_timeout_ms);
DECLARE_Int64(wait_cancel_release_memory_ms);

DECLARE_mBool(check_segment_when_build_rowset_meta);

DECLARE_Int32(num_query_ctx_map_partitions);

DECLARE_mBool(enable_s3_rate_limiter);
DECLARE_mInt64(s3_get_bucket_tokens);
DECLARE_mInt64(s3_get_token_per_second);
DECLARE_mInt64(s3_get_token_limit);

DECLARE_mInt64(s3_put_bucket_tokens);
DECLARE_mInt64(s3_put_token_per_second);
DECLARE_mInt64(s3_put_token_limit);
DECLARE_mInt64(s3_rate_limiter_log_interval);

// CPU-aware S3 rate limiter: GET/PUT QPS per CPU core. A negative value means unset and
// falls back to the legacy absolute token configs above; 0 disables QPS limiting.
DECLARE_mInt64(s3_get_requests_per_second_per_core);
DECLARE_mInt64(s3_put_requests_per_second_per_core);
// Hard caps for the CPU-derived GET/PUT QPS. A non-positive value means no cap.
DECLARE_mInt64(s3_get_requests_per_second_max);
DECLARE_mInt64(s3_put_requests_per_second_max);
// GET/PUT bytes per second per CPU core. A non-positive value disables byte-rate limiting.
DECLARE_mInt64(s3_get_bytes_per_second_per_core);
DECLARE_mInt64(s3_put_bytes_per_second_per_core);
// Hard caps for the CPU-derived GET/PUT bytes/s. A non-positive value means no cap.
DECLARE_mInt64(s3_get_bytes_per_second_max);
DECLARE_mInt64(s3_put_bytes_per_second_max);
// Override for cores used to derive effective limits: a non-positive value means auto-detect.
DECLARE_mInt32(s3_rate_limiter_cpu_cores_override);
// max s3 client retry times
DECLARE_mInt32(max_s3_client_retry);
// When meet s3 429 error, the "get" request will
// sleep s3_read_base_wait_time_ms (*1, *2, *3, *4) ms
// get try again.
// The max sleep time is s3_read_max_wait_time_ms
// and the max retry time is max_s3_client_retry
DECLARE_mInt32(s3_read_base_wait_time_ms);
DECLARE_mInt32(s3_read_max_wait_time_ms);
DECLARE_mBool(enable_s3_object_check_after_upload);
DECLARE_mInt32(aws_client_request_timeout_ms);

// write as inverted index tmp directory
DECLARE_String(tmp_file_dir);

// the directory for storing the trino-connector plugins.
DECLARE_String(trino_connector_plugin_dir);

DECLARE_String(jni_plugin_dir);
DECLARE_String(jni_plugin_hadoop_conf_dir);
DECLARE_String(jni_plugin_fs_dir);
DECLARE_Bool(java_plugin_warmup);

// the file paths(one or more) of CA cert, splite using ";" aws s3 lib use it to init s3client
DECLARE_mString(ca_cert_file_paths);

/** Table sink configurations(currently contains only external table types) **/
// Minimum data processed to scale writers in exchange when non partition writing
DECLARE_mInt64(table_sink_non_partition_write_scaling_data_processed_threshold);
// Minimum data processed to trigger skewed partition rebalancing in exchange when partition writing
DECLARE_mInt64(table_sink_partition_write_min_data_processed_rebalance_threshold);
// Minimum partition data processed to rebalance writers in exchange when partition writing
DECLARE_mInt64(table_sink_partition_write_min_partition_data_processed_rebalance_threshold);
// Maximum processed partition nums of per writer when partition writing
DECLARE_mInt32(table_sink_partition_write_max_partition_nums_per_writer);

/** Hive sink configurations **/
DECLARE_mInt64(hive_sink_max_file_size);

/** Iceberg sink configurations **/
DECLARE_mInt64(iceberg_sink_max_file_size);

/** Paimon sink configurations **/
// Hard upper bound for Doris-managed Paimon write-buffer memory per JNI writer.
DECLARE_mInt64(paimon_jni_writer_memory_pool_limit_bytes);

// Number of open tries, default 1 means only try to open once.
// Retry the Open num_retries time waiting 100 milliseconds between retries.
DECLARE_mInt32(thrift_client_open_num_tries);

// Default HTTP scheme used by S3Client when the endpoint has no scheme.
DECLARE_String(s3_client_http_scheme);

DECLARE_mBool(ignore_schema_change_check);

// The min thread num for BufferedReaderPrefetchThreadPool
DECLARE_Int64(num_buffered_reader_prefetch_thread_pool_min_thread);
// The max thread num for BufferedReaderPrefetchThreadPool
DECLARE_Int64(num_buffered_reader_prefetch_thread_pool_max_thread);

DECLARE_mBool(enable_segment_prefetch_verbose_log);
// The thread num for SegmentPrefetchThreadPool
DECLARE_Int64(segment_prefetch_thread_pool_thread_num_min);
DECLARE_Int64(segment_prefetch_thread_pool_thread_num_max);

DECLARE_mInt32(segment_file_cache_consume_rowids_batch_size);
// Enable segment file cache block prefetch for query
DECLARE_mBool(enable_query_segment_file_cache_prefetch);
// Number of blocks to prefetch ahead in segment iterator for query
DECLARE_mInt32(query_segment_file_cache_prefetch_block_size);
// Enable segment file cache block prefetch for compaction
DECLARE_mBool(enable_compaction_segment_file_cache_prefetch);
// Number of blocks to prefetch ahead in segment iterator for compaction
DECLARE_mInt32(compaction_segment_file_cache_prefetch_block_size);
// The min thread num for S3FileUploadThreadPool
DECLARE_Int64(num_s3_file_upload_thread_pool_min_thread);
// The max thread num for S3FileUploadThreadPool
DECLARE_Int64(num_s3_file_upload_thread_pool_max_thread);
// The maximum jvm heap usage ratio for hdfs write workload
DECLARE_mDouble(max_hdfs_wirter_jni_heap_usage_ratio);
// The sleep milliseconds duration when hdfs write exceeds the maximum usage
DECLARE_mInt64(hdfs_jni_write_sleep_milliseconds);
// The max retry times when hdfs write failed
DECLARE_mInt64(hdfs_jni_write_max_retry_time);

// The min thread num for NonBlockCloseThreadPool
DECLARE_Int64(min_nonblock_close_thread_num);
// The max thread num for NonBlockCloseThreadPool
DECLARE_Int64(max_nonblock_close_thread_num);
// The possibility that mem allocator throws an exception during memory allocation
// This config is for test usage, be careful when changing it.
DECLARE_mDouble(mem_alloc_fault_probability);
// The time out milliseconds for remote fetch schema RPC
DECLARE_mInt64(fetch_remote_schema_rpc_timeout_ms);
// The size of the local buffer for S3FileSytem's upload function

DECLARE_Int64(s3_file_system_local_upload_buffer_size);

//JVM monitoring enable. To prevent be from crashing due to jvm compatibility issues.
DECLARE_Bool(enable_jvm_monitor);

// Num threads to load data dirs, default value -1 indicates the same number of threads as the number of data dirs
DECLARE_Int32(load_data_dirs_threads);

// Skip loading stale rowset meta when initializing `TabletMeta` from protobuf
DECLARE_mBool(skip_loading_stale_rowset_meta);
// Whether to use file to record log. When starting BE with --console,
// all logs will be written to both standard output and file.
// Disable this option will no longer use file to record log.
// Only works when starting BE with --console.
DECLARE_Bool(enable_file_logger);

// Enable partition column fallback when partition columns are missing from file
DECLARE_Bool(enable_iceberg_partition_column_fallback);

// The minimum row group size when exporting Parquet files.
DECLARE_Int64(min_row_group_size);

DECLARE_mInt64(compaction_memory_bytes_limit);

DECLARE_mInt64(compaction_batch_size);

DECLARE_mBool(enable_parquet_page_index);

// Wheather to ignore not found file in external teble(eg, hive)
// Default is true, if set to false, the not found file will result in query failure.
DECLARE_mBool(ignore_not_found_file_in_external_table);

DECLARE_mBool(enable_hdfs_mem_limiter);

// Define how many percent data in hashtable bigger than limit
// we should do agg limit opt
DECLARE_mInt16(topn_agg_limit_multiplier);

DECLARE_mInt64(tablet_meta_serialize_size_limit);

DECLARE_mInt64(pipeline_task_leakage_detect_period_secs);
// To be compatible with hadoop's block compression
DECLARE_mInt32(snappy_compression_block_size);
DECLARE_mInt32(lz4_compression_block_size);

DECLARE_mBool(enable_pipeline_task_leakage_detect);

DECLARE_mInt32(check_score_rounds_num);

// MB
DECLARE_Int32(query_cache_size);
// Max incremental merges on one query cache entry before forcing a full
// recompute to compact the entry (see query_cache.h QueryCacheRuntime).
DECLARE_mInt32(query_cache_max_incremental_merge_count);
DECLARE_Bool(force_regenerate_rowsetid_on_start_error);

// Enable validation to check the correctness of table size.
DECLARE_Bool(enable_table_size_correctness_check);
// Enable sleep 5s between delete cumulative compaction.
DECLARE_mBool(enable_sleep_between_delete_cumu_compaction);

DECLARE_mInt32(compaction_num_per_round);
DECLARE_mInt32(max_automatic_compaction_num_per_round);

DECLARE_mInt32(check_tablet_delete_bitmap_interval_seconds);
DECLARE_mInt32(check_tablet_delete_bitmap_score_top_n);
DECLARE_mBool(enable_check_tablet_delete_bitmap_score);

// whether to prune rows with delete sign = 1 in base compaction
// ATTN: this config is only for test
DECLARE_mBool(enable_prune_delete_sign_when_base_compaction);

DECLARE_mBool(enable_mow_verbose_log);

DECLARE_mInt32(tablet_sched_delay_time_ms);
DECLARE_mInt32(load_trigger_compaction_version_percent);
DECLARE_mInt64(base_compaction_interval_seconds_since_last_operation);
DECLARE_mBool(enable_compaction_pause_on_high_memory);

DECLARE_mBool(enable_quorum_success_write);
DECLARE_mDouble(quorum_success_max_wait_multiplier);
DECLARE_mInt64(quorum_success_min_wait_seconds);
DECLARE_mInt32(quorum_success_remaining_timeout_seconds);

DECLARE_mBool(enable_calc_delete_bitmap_between_segments_concurrently);

DECLARE_mBool(enable_update_delete_bitmap_kv_check_core);

DECLARE_mBool(enable_fetch_rowsets_from_peer_replicas);
// the max length of segments key bounds, in bytes
// ATTENTION: as long as this conf has ever been enabled, cluster downgrade and backup recovery will no longer be supported.
DECLARE_mInt32(segments_key_bounds_truncation_threshold);
// ATTENTION: for test only, use random segments key bounds truncation threshold every time
DECLARE_mBool(random_segments_key_bounds_truncation);

// If true, non-MOW rowsets store a single aggregated [rowset_min, rowset_max]
// key-bounds entry instead of per-segment bounds, to reduce meta size on cloud FDB.
DECLARE_mBool(enable_aggregate_non_mow_key_bounds);

DECLARE_mBool(enable_auto_clone_on_compaction_missing_version);

DECLARE_mBool(enable_auto_clone_on_mow_publish_missing_version);

// p0, daily, rqg, external
DECLARE_String(fuzzy_test_type);

// The maximum csv line reader output buffer size
DECLARE_mInt64(max_csv_line_reader_output_buffer_size);

// The maximum bytes of a single block returned by load file readers (CsvReader, NewJsonReader,
// ParquetReader, OrcReader). Default is 200MB. Set to 0 to disable the limit.
DECLARE_mInt64(load_reader_max_block_bytes);

// Maximum number of OpenMP threads available for concurrent index builds.
// -1 means auto: use 80% of detected CPU cores.
DECLARE_Int32(omp_threads_limit);
// The capacity of segment partial column cache, used to cache column readers for each segment.
DECLARE_mInt32(max_segment_partial_column_cache_size);
// Cache for ANN index IVF on-disk list data.
// Default "70%" means 70% of total physical memory.
DECLARE_String(ann_index_ivf_list_cache_limit);
// Stale sweep time for ANN index IVF list cache in seconds.
DECLARE_mInt32(ann_index_ivf_list_cache_stale_sweep_time_sec);
// Minimum segment rows required to persist an ANN index.
DECLARE_mInt64(ann_index_build_min_segment_rows);

DECLARE_mBool(enable_prefill_output_dbm_agg_cache_after_compaction);
DECLARE_mBool(enable_prefill_all_dbm_agg_cache_after_compaction);

DECLARE_mBool(enable_wal_tde);

DECLARE_mBool(print_stack_when_cache_miss);

DECLARE_mBool(read_cluster_cache_opt_verbose_log);

DECLARE_mString(aws_credentials_provider_version);

// Concurrency stats dump configuration
DECLARE_mBool(enable_concurrency_stats_dump);
DECLARE_mInt32(concurrency_stats_dump_interval_ms);

DECLARE_mBool(cloud_mow_sync_rowsets_when_load_txn_begin);

DECLARE_mBool(enable_cloud_make_rs_visible_on_be);
DECLARE_mBool(enable_cloud_random_segment_id);
DECLARE_mInt32(file_handles_deplenish_frequency_times);

#ifdef BE_TEST
DECLARE_String(test_s3_ak);
DECLARE_String(test_s3_sk);
DECLARE_String(test_s3_endpoint);
DECLARE_String(test_s3_region);
DECLARE_String(test_s3_bucket);
DECLARE_String(test_s3_prefix);
#endif

class Register {
public:
    struct Field {
        const char* type = nullptr;
        const char* name = nullptr;
        void* storage = nullptr;
        const char* defval = nullptr;
        bool valmutable = false;
        Field(const char* ftype, const char* fname, void* fstorage, const char* fdefval,
              bool fvalmutable)
                : type(ftype),
                  name(fname),
                  storage(fstorage),
                  defval(fdefval),
                  valmutable(fvalmutable) {}
    };

public:
    static std::map<std::string, Field>* _s_field_map;

public:
    Register(const char* ftype, const char* fname, void* fstorage, const char* fdefval,
             bool fvalmutable) {
        if (_s_field_map == nullptr) {
            _s_field_map = new std::map<std::string, Field>();
        }
        Field field(ftype, fname, fstorage, fdefval, fvalmutable);
        _s_field_map->insert(std::make_pair(std::string(fname), field));
    }
};

// RegisterConfValidator class is used to store validator function of registered config fields in
// Register::_s_field_map.
// If any validator return false when BE bootstart, the bootstart will be terminated.
// If validator return false when use http API to update some config, the config will not
// be modified and the API will return failure.
class RegisterConfValidator {
public:
    // Validator for each config name.
    static std::map<std::string, std::function<bool()>>* _s_field_validator;

public:
    RegisterConfValidator(const char* fname, const std::function<bool()>& validator) {
        if (_s_field_validator == nullptr) {
            _s_field_validator = new std::map<std::string, std::function<bool()>>();
        }
        // register validator to _s_field_validator
        _s_field_validator->insert(std::make_pair(std::string(fname), validator));
    }
};

// RegisterConfUpdateCallback class is used to store callback functions that will be called
// when a config field is updated at runtime.
// The callback function takes two void pointers: old_value and new_value.
// The actual type casting is done in the DEFINE_ON_UPDATE macro.
class RegisterConfUpdateCallback {
public:
    using CallbackFunc = std::function<void(const void* old_ptr, const void* new_ptr)>;
    // Callback map for each config name.
    static std::map<std::string, CallbackFunc>* _s_field_update_callback;

public:
    RegisterConfUpdateCallback(const char* fname, const CallbackFunc& callback) {
        if (_s_field_update_callback == nullptr) {
            _s_field_update_callback = new std::map<std::string, CallbackFunc>();
        }
        // register callback to _s_field_update_callback
        _s_field_update_callback->insert(std::make_pair(std::string(fname), callback));
    }
};

// configuration properties load from config file.
class Properties {
public:
    // load conf from file, if must_exist is true and file does not exist, return false
    bool load(const char* conf_file, bool must_exist = true);

    // Find the config value by key from `file_conf_map`.
    // If found, set `retval` to the config value,
    // or set `retval` to `defstr`
    // if retval is not set(in case defstr is nullptr), set is_retval_set to false
    template <typename T>
    bool get_or_default(const char* key, const char* defstr, T& retval, bool* is_retval_set,
                        std::string& rawval) const;

    void set(const std::string& key, const std::string& val);

    void set_force(const std::string& key, const std::string& val);

    // dump props to conf file
    Status dump(const std::string& conffile);

    const std::map<std::string, std::string>& conf_map() const { return file_conf_map; }

private:
    std::map<std::string, std::string> file_conf_map;
};

// full configurations.
extern std::map<std::string, std::string>* full_conf_map;

extern std::mutex custom_conf_lock;

// Init the config from `conf_file`.
// If fill_conf_map is true, the updated config will also update the `full_conf_map`.
// If must_exist is true and `conf_file` does not exist, this function will return false.
// If set_to_default is true, the config value will be set to default value if not found in `conf_file`.
bool init(const char* conf_file, bool fill_conf_map = false, bool must_exist = true,
          bool set_to_default = true);

Status set_config(const std::string& field, const std::string& value, bool need_persist = false,
                  bool force = false);

Status persist_config(const std::string& field, const std::string& value);

std::mutex* get_mutable_string_config_lock();

std::vector<std::vector<std::string>> get_config_info();

Status set_fuzzy_configs();

void update_config(const std::string& field, const std::string& value);

} // namespace config
} // namespace doris

Coverage Report

Created: 2026-10-15 04:57