E-MapReduce:Parameters of the FE node

max_dynamic_partition_num

The maximum number of partitions that can be created in a single operation for a dynamic partition table. Set this limit to prevent a large number of partitions from being created at once. The number of partitions created is determined by the dynamic_partition.start and dynamic_partition.end values.

dynamic_partition_enable

Enables or disables dynamic partition scheduling globally.

dynamic_partition_check_interval_seconds

The interval at which the FE node checks for dynamic partition updates.

min_replication_num_per_tablet

The minimum number of replicas per tablet.

max_replication_num_per_tablet

The maximum number of replicas per tablet.

enable_batch_delete_by_default

When set to true, adds the _DORIS_DELETE_SIGN_ column automatically when a Unique Key table is created.

recover_with_empty_tablet

When set to true, Apache Doris replaces damaged tablets (or tablets whose replicas are all lost) with empty tablets so that queries can continue. Query results may be incomplete due to data loss.

max_allowed_in_element_num_of_delete

The maximum number of elements that the predicate of a DELETE statement can filter.

max_running_rollup_job_num_per_table

The maximum number of rollup jobs that can run concurrently per table.

disable_cluster_feature

Disables the multi-cluster feature, which is deprecated in Apache Doris V0.12 and later. When set to true, the following cluster operations are disabled:

• Create and delete clusters

• Add and remove BE nodes, add BE nodes to a cluster, and disable cluster-level balancing

• Change the number of BE nodes in a cluster

• Connect and migrate databases

default_max_filter_ratio

The maximum percentage of rows that can be filtered during a load job. The default value of 0 means strict mode: the job fails if any row is filtered. Set this to a value between 0 and 1 to allow a certain percentage of rows to be skipped.

default_db_data_quota_bytes

The default data size quota for each database. To change the quota for a specific database:

-- Set the data quota (unit: KB, MB, GB, TB, or PB)
ALTER DATABASE db_name SET DATA QUOTA quota;

-- View the current database configuration
SHOW DATA;
-- For more information: HELP SHOW DATA

default_db_replica_quota_size

The default replica count quota for each database. To change the quota:

ALTER DATABASE db_name SET REPLICA QUOTA quota;
SHOW DATA;

max_running_txn_num_per_db

The maximum number of load jobs that can run concurrently in a single database. If this limit is exceeded, new jobs return the following error:

current running txns on db xxx is xx, larger than limit xx

Wait for running jobs to complete before retrying. Avoid increasing this value unnecessarily — high concurrency increases system load.

max_unfinished_load_job

The maximum number of active load jobs across all statuses: PENDING, ETL, LOADING, and QUORUM_FINISHED. New jobs cannot be submitted when this limit is reached.

async_loading_load_task_pool_size

The thread pool size for Broker Load jobs in the Loading state. This limits the maximum number of concurrent Broker Load jobs in the Loading state.

async_pending_load_task_pool_size

The thread pool size for load jobs in the Pending state. This limits Broker Load and Spark Load jobs in the Pending state. This value must be less than max_running_txn_num_per_db.

async_load_task_pool_size

This parameter is deprecated. It exists only for backward compatibility and will be removed in a future version. Use async_loading_load_task_pool_size instead.

desired_max_waiting_jobs

The target maximum number of pending Routine Load V2.0 jobs. This value may be exceeded temporarily during events such as a master FE node role change.

max_bytes_per_broker_scanner

The maximum amount of data that a single broker scanner processes in a load job. In most deployments, each BE node has one broker scanner.

default_load_parallelism

The default number of parallel broker scanners for Broker Load jobs on the FE node. The effective parallelism is determined by this parameter together with max_broker_concurrency and min_bytes_per_broker_scanner. If you specify parallelism when submitting a Broker Load job, that value overrides this default.

max_broker_concurrency

The maximum number of broker scanners that can run concurrently.

min_bytes_per_broker_scanner

The minimum amount of data assigned to a single broker scanner.

load_running_job_num_limit

The maximum number of running load jobs. The default value of 0 means no limit.

load_input_size_limit_gb

The maximum data size per load job. The default value of 0 means no limit.

load_straggler_wait_second

The maximum time the FE node waits for a lagging node before marking it as failed. For example, if a load job runs on nodes A, B, and C, and A and B finish at time T1 but C has not, C is marked as failed after 300 seconds. This also sets the timeout for publishing load jobs. Database administrators can override this on a per-job basis.

load_etl_thread_num_normal_priority

Concurrent threads for NORMAL-priority load jobs in the ETL state.

load_etl_thread_num_high_priority

Concurrent threads for HIGH-priority load jobs in the ETL state.

load_pending_thread_num_normal_priority

Concurrent threads for NORMAL-priority load jobs in the Pending state.

load_pending_thread_num_high_priority

Concurrent threads for HIGH-priority load jobs in the Pending state. Small batch load jobs use HIGH priority; all other load jobs use NORMAL priority. This scheduling prevents slow jobs from blocking threads for extended periods.

load_checker_interval_second

The interval at which the scheduler runs a load check thread.

disable_load_job

When set to true, disables all load job submissions.

disable_hadoop_load

When set to true, disables the feature for loading data from Hadoop. This feature is enabled by default but is not recommended for production use.

enable_spark_load

Enables Spark Load. Disabled by default.

enable_odbc_table

Enables Open Database Connectivity (ODBC) external tables. To enable at runtime:

ADMIN SET FRONTEND CONFIG ("enable_odbc_table" = "true");

using_old_load_usage_pattern

When set to true, a failed INSERT statement returns a label that you can use to check the load job status. By default, a failed INSERT statement throws an exception without a label.

broker_load_default_timeout_second

The default timeout for Broker Load jobs.

stream_load_default_timeout_second

The default timeout for Stream Load and mini load jobs.

insert_load_default_timeout_second

The default timeout for INSERT INTO jobs.

mini_load_default_timeout_second

The default timeout for mini load jobs that are not Stream Load.

spark_load_default_timeout_second

The default timeout for Spark Load jobs.

hadoop_load_default_timeout_second

The default timeout for Hadoop load jobs.

min_load_timeout_second

The minimum allowed timeout for any load job type.

max_load_timeout_second

The maximum allowed timeout for load jobs, excluding Stream Load.

max_stream_load_timeout_second

The maximum allowed timeout for Stream Load and mini load jobs.

broker_timeout_ms

The default timeout for broker RPC connections.

db_used_data_quota_update_interval_secs

The interval at which a daemon thread updates the data usage quota for each database. Load jobs check this cached value (not a real-time calculation) when verifying quota compliance.

agent_task_resend_wait_time_ms

The minimum time that must pass since an agent task was created before ReportHandler can resend it. This prevents repeated sending of PUBLISH_VERSION agent tasks. Increase this value if tasks arrive at the queue or BE nodes with noticeable latency, but be aware that increasing it also delays retry after a task failure.

disable_show_stream_load

When set to true, disables the display and record clearing for Stream Load jobs.

max_stream_load_record_size

The maximum number of Stream Load job records stored in memory.

fetch_stream_load_record_interval_second

The interval at which Stream Load job records are fetched.

max_routine_load_task_num_per_be

The maximum number of Routine Load tasks that can run concurrently on a single BE node. This value must be less than routine_load_thread_pool_size on the BE node (default: 10).

max_routine_load_task_concurrent_num

The maximum number of concurrent tasks per Routine Load job.

max_routine_load_job_num

The maximum total number of Routine Load jobs, including those in NEED_SCHEDULED, RUNNING, and PAUSE states.

period_of_auto_resume_min

The interval at which a paused Routine Load job is automatically resumed.

max_tolerable_backend_down_num

The maximum number of failed BE nodes that still allow automatic Routine Load resume. The default value of 0 means automatic resume is allowed only when all BE nodes are healthy.

label_keep_max_second

The maximum retention period for labels of deleted or canceled load jobs. After this period, labels are permanently removed and can be reused.

Reducing this value decreases FE node memory usage, because load job information is kept in memory until deleted. Under high concurrency with load job backlog, you may see a call frontend service failed error — in that case, check whether metadata writing is slow, and consider reducing this value to 12 or 6 hours.

streaming_label_keep_max_second

The maximum retention period for labels of high-frequency load jobs after they complete. This applies to INSERT INTO, Stream Load, and Routine Load jobs. Completed jobs and tasks are deleted after this period.

delete_thread_num

Concurrent threads for delete operations.

max_backup_restore_job_num_per_db

The maximum number of backup and restore job records stored per database.

backup_job_default_timeout_ms

The default timeout for backup jobs.

catalog_trash_expire_second

After a database, table, or partition is deleted, the data can be recovered using the RECOVER statement until this retention period expires. After the period elapses, the data is permanently deleted.

small_file_dir

The directory where small files managed by SmallFileMgr are stored.

max_small_file_size_bytes

The maximum size of a file stored in SmallFileMgr.

max_small_file_number

The maximum number of files stored in SmallFileMgr.

max_layout_length_per_row

The maximum memory layout length of a single row. The maximum RowBlock size on a BE node is 100 MB (configured by max_unpacked_row_block_size in be.conf), and each RowBlock holds 1024 rows — giving approximately 100 KB per row.

For example, a row with schema k1 INTEGER, v1 DECIMAL, v2 VARCHAR(2000) has a layout length of 2020 bytes: 4 (INTEGER) + 16 (DECIMAL) + 2000 (VARCHAR). Run HELP CREATE TABLE in the MySQL client for more details on memory layout lengths.

To support wider schemas, increase max_unpacked_row_block_size in be.conf, but the performance impact is unknown.

cache_result_max_row_count

The maximum number of rows that can be stored in the query result cache.

cache_last_version_interval_second

The interval used to distinguish between offline and real-time partition updates for cache freshness decisions.

cache_enable_partition_mode

When set to true, allows the FE node to retrieve cached data from BE nodes. This applies to real-time updates in specific partitions.

cache_enable_sql_mode

When set to true, allows the FE node to serve results from the SQL result cache. This applies to offline partition updates.

max_query_retry_time

The maximum number of times a query is retried when an RPC failure returns no results. Reduce this value to limit the avalanche effect under heavy load.

remote_fragment_exec_timeout_ms

The timeout for asynchronously executing a remote query fragment. Under high system load, increase this value.

enable_local_replica_selection

When set to true, the query planner prefers replicas hosted on the same machine as the FE node. This reduces network traffic in deployments where N hosts each run one BE node and one FE node, data has N replicas, and queries are distributed evenly across FE nodes.

When local replicas are unavailable, queries fall back to non-local replicas only if enable_local_replica_selection_fallback is also set to true.

enable_local_replica_selection_fallback

When set to true, queries fall back to non-local replicas when local replicas are unavailable. Requires enable_local_replica_selection to also be set to true.

max_distribution_pruner_recursion_depth

The maximum recursion depth for the hash distribution pruner. The pruner computes the cross-product of elements across distributed columns to determine which buckets to scan. For example, four distributed columns with 5, 4, 3, and 2 elements have a recursion depth of 5 × 4 × 3 × 2 = 120. Because 120 exceeds the default limit of 100, the pruner skips and scans all buckets.

Increasing this value enables pruning for more elements but consumes more CPU.

Because this parameter has MasterOnly: false, run ADMIN SET FRONTEND CONFIG on each FE node separately.

default_max_query_instances

The maximum number of query instances a single user can use simultaneously. A value of 0 or less means no limit.

rewrite_count_distinct_to_bitmap_hll

This is a session-level variable. For Aggregate model tables, it rewrites COUNT(DISTINCT c1) as follows:

• If c1 is a bitmap type: rewrites to BITMAP_UNION_COUNT(c1)

• If c1 is a HyperLogLog (HLL) type: rewrites to HLL_UNION_AGG(c1)

Set to false to disable rewriting.

max_connection_scheduler_threads_num

The maximum number of threads in the query scheduler.

qe_max_connection

The maximum number of concurrent connections per FE node.

use_compact_thrift_rpc

When set to true, query plan structs are sent in a compressed format, reducing their size by approximately 50% and preventing send-fragment timeout errors. In high-concurrency, small-query scenarios, concurrency may decrease by approximately 10%.

expr_depth_limit

The maximum depth of an expression tree. Queries with expression trees deeper than this limit may cause read lock timeouts in the database.

expr_children_limit

The maximum number of child nodes in an expression tree. Queries that exceed this limit may cause read lock timeouts.

disable_colocate_balance

When set to true, disables ColocateTableBalancer from relocating or rebalancing colocate tables.

query_colocate_join_memory_limit_penalty_factor

The memory limit ratio applied when a colocate join is executed.

colocate_group_relocate_delay_second

The delay before Apache Doris redistributes a colocation group after detecting a BE node failure. This prevents unnecessary redistributions when the BE node recovers within the delay period. Redistribution involves a large number of tablet migrations, so a conservative delay is preferred.

check_consistency_default_timeout_second

The default timeout for a tablet consistency check. Set this relative to the size of your tablets.

consistency_check_start_time

The hour (0–23) at which the periodic consistency check starts. If this value equals consistency_check_end_time, no consistency check runs.

consistency_check_end_time

The hour (0–23) at which the periodic consistency check ends. If this value equals consistency_check_start_time, no consistency check runs.

tablet_rebalancer_type

The rebalancing strategy. If the specified value cannot be parsed, BeLoad is used.

max_balancing_tablets

The maximum number of tablets that TabletScheduler balances concurrently. If the number of tablets pending balancing exceeds this value, the balance check is skipped.

max_scheduling_tablets

The maximum number of tablets that TabletScheduler schedules concurrently. If the number of tablets pending scheduling exceeds this value, the check is skipped.

disable_balance

When set to true, disables the balance check in TabletScheduler.

balance_load_score_threshold

The load score threshold for classifying BE node load. A BE node with a load score more than 10% below the average is considered low-load. Otherwise, it is considered high-load. This score is used to decide rebalancing moves.

schedule_slot_num_per_path

The number of tablet scheduler slots per disk path.

tablet_repair_delay_factor_second

The base delay factor for tablet repair. Repair timing depends on tablet priority:

disable_tablet_scheduler

When set to true, disables the tablet scheduler entirely, stopping both replica repair and balancing.

enable_force_drop_redundant_replica

When set to true, redundant replicas are dropped immediately during replica scheduling. This accelerates balancing and repair but may cause load jobs that are writing to those replicas to fail. Enable this temporarily when a large number of replicas are pending balancing or repair.

repair_slow_replica

When set to true, replicas with slow compaction are automatically detected and migrated to another node. Detection triggers when the minimum version count across replicas falls below min_version_count_indicate_replica_compaction_too_slow and the ratio of minimum to maximum version count exceeds valid_version_count_delta_ratio_between_replicas.

min_version_count_indicate_replica_compaction_too_slow

The version count threshold below which a replica is considered to have slow compaction.

valid_version_count_delta_ratio_between_replicas

The ratio threshold between the minimum and maximum version counts across replicas. Used together with repair_slow_replica to identify replicas with the slowest compaction.

min_bytes_indicate_replica_too_large

The data size threshold above which a replica is considered oversized.

skip_compaction_slower_replica

When set to true, replicas with slow compaction are skipped during query planning.

allow_replica_on_same_host

When set to true, multiple replicas of the same tablet can reside on the same host. Use this only in test environments where multiple BE nodes run on one host to verify multi-replica behavior.

drop_backend_after_decommission

When set to true, a BE node is removed from the cluster after it is fully decommissioned. When set to false, the BE node remains in DECOMMISSION state after its data is migrated.

Setting this to false is useful for manual disk balancing. In clusters where initial disks contain much more data than disks added later (because Apache Doris cannot automatically balance data between disks on a single BE node), follow this procedure for each BE node:

1. Set drop_backend_after_decommission to false.

2. Decommission the BE node to migrate its data to other BE nodes.

3. After decommission completes, cancel the decommission status so data migrates back. At this point, data is distributed evenly across all disks.

4. Repeat for all BE nodes.

replica_delay_recovery_second

The minimum interval before the FE node uses a clone job to recover a failed replica.

tablet_stat_update_interval_second

The interval at which all FE nodes collect tablet statistics from BE nodes.

tablet_delete_timeout_second

The timeout for deleting a tablet.

tablet_create_timeout_second

The timeout for creating a single replica.

max_create_table_timeout_second

The maximum timeout for a CREATE TABLE operation.

alter_table_timeout_second

The maximum timeout for ALTER TABLE operations. Set this relative to the data size of your tables.

max_backend_down_time_second

The maximum time a BE node can be down before the BACKEND_DOWN event is triggered.

es_state_sync_interval_second

The interval at which the FE node queries Elasticsearch to retrieve shard information for an index.

min_clone_task_timeout_sec

The minimum timeout for a clone job.

max_clone_task_timeout_sec

The maximum timeout for a clone job. In most cases, the actual timeout is estimated from data size at a minimum transfer speed of 5 MB/s. Adjust these limits manually if clone jobs time out unexpectedly.

clone_high_priority_delay_second

The delay before a high-priority clone job runs after it is created.

clone_normal_priority_delay_second

The delay before a normal-priority clone job runs. This parameter is ignored if its value is less than clone_checker_interval_second.

clone_low_priority_delay_second

The delay before a low-priority clone job runs. This prevents a large number of clone jobs from starting immediately when a host is briefly unavailable. This parameter is ignored if its value is less than clone_checker_interval_second.

clone_max_job_num

The maximum number of concurrent low-priority clone jobs. High-priority clone jobs have no concurrency limit.

clone_job_timeout_second

The timeout for a clone job. Set this relative to the size of your replicas — larger replicas take longer to clone.

clone_checker_interval_second

The interval at which the clone checker runs.

clone_distribution_balance_threshold

The threshold for balancing replica counts across BE nodes.

clone_capacity_balance_threshold

The threshold for balancing data capacity across BE nodes. Balancing uses these bounds:

• Upper limit = Average used capacity (AUC) × (1 + clone_capacity_balance_threshold)

• Lower limit = AUC × (1 − clone_capacity_balance_threshold)

The clone checker moves replicas from high-capacity BE nodes to low-capacity BE nodes.

capacity_used_percent_high_water

The disk usage threshold above which a BE node is considered high-load. Used when calculating the load score for BE nodes.

storage_flood_stage_usage_percent

Works together with storage_flood_stage_left_capacity_bytes. When a disk path reaches both thresholds simultaneously, load jobs and restore jobs are rejected for that path.

storage_flood_stage_left_capacity_bytes

Works together with storage_flood_stage_usage_percent. See the description above.

storage_high_watermark_usage_percent

Works together with storage_min_left_capacity_bytes. When a disk path exceeds both thresholds, it cannot be used as the destination for new replica writes. To protect data integrity, these thresholds can be exceeded temporarily during tablet restore operations.

storage_min_left_capacity_bytes

Works together with storage_high_watermark_usage_percent. See the description above.

storage_cooldown_second

When a table or partition is created with SSD as the storage medium, tablets remain on SSD for this duration before automatically moving to HDD. Override this in the CREATE TABLE statement.

default_storage_medium

The default storage medium for new tables and partitions when no storage medium is specified.

disable_storage_medium_check

When set to true, ReportHandler skips storage medium checks and disables the storage cooldown feature. Set this when you have no requirements for storage medium differentiation.

edit_log_type

The type of edit log. Use BDB to write logs to BDBJE. The LOCAL type is deprecated.

edit_log_port

The port used by BDBJE.

edit_log_roll_num

The maximum number of log entries written before the master FE node creates a new log file.

meta_dir

The directory where Apache Doris metadata is stored. For best performance and durability, store metadata on an SSD (for high write throughput) with RAID (for redundancy).

meta_delay_toleration_second

The maximum tolerable metadata replication lag. If follower or observer FE nodes lag beyond this threshold, they stop serving requests.

meta_publish_timeout_ms

The timeout for publishing metadata.

ignore_meta_check

When set to true, metadata lag between the master and follower or observer FE nodes is ignored, even if it exceeds meta_delay_toleration_second. Follower and observer FE nodes continue to serve read requests. Use this setting when the master FE node is stopped for an extended period but read access through follower or observer FE nodes is still needed.

metadata_failure_recovery

When set to true, the FE node resets the BDBJE replication groups and starts as the master. Use this only if all follower FE nodes fail to start: copy the metadata to another node and restart with this parameter set to true.

force_do_metadata_checkpoint

When set to true, the checkpoint thread creates a metadata checkpoint regardless of JVM memory usage. By default, checkpointing is skipped when JVM memory usage is high to prevent out-of-memory (OOM) errors.

metadata_checkpoint_memory_threshold

The JVM heap usage percentage above which the checkpoint thread pauses to prevent an OOM error.

bdbje_replica_ack_timeout_second

The maximum time the master FE node waits for acknowledgment from follower FE nodes when writing metadata. Increase this value if large metadata writes cause acknowledgment timeouts and master exits.

bdbje_lock_timeout_second

The lock timeout for BDBJE operations. If LockTimeoutException appears frequently in the FE WARN log, increase this value.

bdbje_heartbeat_timeout_second

The heartbeat timeout between the master and follower FE nodes in BDBJE. This matches the BDBJE default. Increase it to tolerate temporary network interruptions.

max_bdbje_clock_delta_ms

The maximum allowed clock offset between the master FE node and follower or observer FE nodes. If the offset exceeds this value when a follower or observer connects, the connection is rejected.

replica_ack_policy

The replica acknowledgment policy in BDBJE. For details, see Enum Durability.ReplicaAckPolicy.

replica_sync_policy

The synchronization policy for follower FE nodes in BDBJE.

master_sync_policy

The synchronization policy for the master FE node in BDBJE.

• With one follower FE node: use SYNC.

• With more than three follower FE nodes: set both master_sync_policy and replica_sync_policy to WRITE_NO_SYNC.

For details, see Enum Durability.SyncPolicy.

txn_rollback_limit

The maximum number of transactions that BDBJE can roll back.

catalog_try_lock_timeout_ms

The timeout for the tryLock method when acquiring the metadata lock. Do not change this except in testing scenarios.

history_job_keep_max_second

The maximum retention period for historical jobs such as schema changes and rollup jobs.

label_clean_interval_second

The interval at which the label cleaner removes expired load job labels.

delete_info_keep_max_second

The maximum retention period for delete operation metadata. Reducing this value decreases FE memory usage and image file size.

transaction_clean_interval_second

The interval at which visible or aborted transactions are cleared. Set this as short as practical to clear transactions promptly.

publish_version_interval_ms

The minimum interval between two version publish operations.

publish_version_timeout_second

The maximum timeout for publishing all versions of a transaction.

http_port

The HTTP server port. All FE nodes must use the same HTTP port.

rpc_port

The Thrift server port on the FE node.

query_port

The MySQL protocol port for connecting to the FE node.

http_backlog_num

The HTTP server backlog queue size. Increasing this value also requires increasing /proc/sys/net/core/somaxconn on Linux.

http_max_line_length

The maximum URL length in an HTTP request.

http_max_header_size

The maximum header length in an HTTP request.

http_max_chunk_size

The maximum HTTP chunk size.

mysql_nio_backlog_num

The MySQL server backlog queue size. Increasing this value also requires increasing /proc/sys/net/core/somaxconn on Linux.

thrift_backlog_num

The Thrift server backlog queue size. Increasing this value requires it to be greater than the value in /proc/sys/net/core/somaxconn on Linux.

thrift_client_timeout_ms

The timeout for Thrift server connections. The default value of 0 means connections never time out.

thrift_server_type

The service model used by the Thrift server. The value is case-insensitive.

thrift_server_max_worker_threads

The maximum number of worker threads for the Thrift server.

mysql_service_io_threads_num

The maximum number of threads the MySQL server uses for I/O event processing.

mysql_service_nio_enabled

Enables asynchronous I/O for the MySQL server.

max_mysql_service_task_threads_num

The maximum number of threads the MySQL server uses to process tasks.

jetty_server_acceptors

The number of Jetty acceptor threads, which accept new connections and hand them to selector threads.

jetty_server_selectors

The number of Jetty selector threads, which decompress HTTP requests using NIO. A small number of selector threads can handle many sockets. For most deployments, 2–4 selector threads are sufficient.

jetty_server_workers

The number of Jetty worker threads, which process HTTP requests using BIO. The appropriate count depends on your application's queries per second (QPS) and the proportion of I/O-bound operations. Higher QPS and more I/O-bound operations require more worker threads. Set this based on your workload.

jetty_server_max_http_post_size

The maximum size of files uploaded via PUT or POST requests.

jetty_server_max_http_header_size

The maximum HTTP request header size.

enable_http_server_v2

Enables HTTP Server 2.0, which is built on Spring Boot and uses a frontend-backend separated architecture. The new frontend UI requires HTTP Server 2.0 to be enabled.

max_agent_task_threads_num

The maximum number of threads in the agent task thread pool.

heartbeat_mgr_threads_num

The number of threads Heartbeat Manager uses to run heartbeat tasks.

heartbeat_mgr_blocking_queue_size

The size of the blocking queue used by Heartbeat Manager to store heartbeat tasks.

backend_rpc_timeout_ms

The timeout for the FE node to send an RPC request to a BE node's backend service.

priority_networks

A network selection policy for machines with multiple IP addresses. Specify one or more CIDR blocks separated by semicolons (;) — for example, 10.10.10.0/24. The FE node uses the IP address that matches a CIDR block. If no address matches, one is selected at random.

proxy_auth_magic_prefix

The magic prefix used for proxy authentication.

proxy_auth_enable

Enables proxy authentication.

grpc_max_message_size_bytes

The maximum gRPC message size. Increase this if query result sets are large.

enable_auth_check

Enables the authentication check feature. Set to false only to diagnose Doris permission system issues.

enable_bdbje_debug_mode

When set to true, the FE node runs in BDBJE debug mode. Navigate to System > bdbje in the web UI to view BDBJE protocol information.

enable_access_file_without_broker

When set to true, the FE node accesses Baidu Object Storage (BOS) or other cloud storage services directly without going through a broker proxy.

enable_outfile_to_local

When set to true, allows the OUTFILE clause to export query results to the local disk.

enable_metric_calculator

When set to true, a daemon thread collects system metrics at a fixed interval.

cluster_id

The ID of the Apache Doris cluster that this FE node belongs to. FE nodes and BE nodes with the same cluster ID form one cluster. The cluster ID is a random integer generated when the master FE node starts for the first time. Set this explicitly to assign a fixed ID.

cluster_name

The name of the Apache Doris cluster, displayed as the page title in the web UI.

auth_token

The authentication token used for identity verification within the Apache Doris cluster.

with_k8s_certs

When set to true, prepares a certificate for deploying Apache Doris on Kubernetes.

enable_deploy_manager

Specifies the third-party deployment manager. Set this when Apache Doris is deployed and managed by a deployment manager.

report_queue_size

The maximum number of report tasks pending in the FE node's report queue. Excessive pending tasks can cause OOM errors. Each BE node sends a tablet report every minute, so there is no need to receive reports without a limit. Keep this at the default value.

check_java_version

When set to true, Apache Doris checks Java version compatibility at startup. If the Java version is incompatible, an error is returned and startup stops.

dpp_hadoop_client_path

The path to the Hadoop client used for Data Processing Platform (DPP) operations.

dpp_bytes_per_reduce

The amount of data processed per reduce task in DPP jobs.

dpp_default_cluster

The default DPP cluster name. This is a historical name from the pre-Apache Doris era.

dpp_default_config_str

The default Hadoop configuration string for DPP jobs.

dpp_config_str

The DPP cluster configuration string.

sys_log_dir

The directory for FE system log files. Two log files are generated:

• fe.log: all FE node logs

• fe.warn.log: warning and error logs only

sys_log_level

The minimum severity level for system log output.

sys_log_roll_num

The maximum number of system log files to retain within each roll interval (as set by sys_log_roll_interval). The default retains up to 10 files per day.

sys_log_roll_interval

The log file rotation interval. Valid values:

sys_log_delete_age

The age after which system log files are deleted. Supported formats: 7d (days), 10h (hours), 60m (minutes), 120s (seconds).

sys_log_verbose_modules

The modules for which verbose (DEBUG-level) logs are generated. For example, setting this to org.apache.doris.catalog generates debug logs only for the catalog module.

log_roll_size_mb

The maximum size of a system log file or audit log file before it rolls over.

audit_log_dir

The directory for FE audit log files. The audit log file fe.audit.log records all requests, including user, host, execution time, and status.

audit_log_roll_num

The maximum number of audit log files to retain within each roll interval.

audit_log_roll_interval

The audit log file rotation interval. Valid values:

audit_log_delete_age

The age after which audit log files are deleted. Supported formats: 7d (days), 10h (hours), 60m (minutes), 120s (seconds).

audit_log_modules

The modules for which audit logs are generated.

qe_slow_log_ms

The response time threshold above which a query is recorded as a slow query in the audit log.

export_tablet_num_per_task

The number of tablets per execution plan in an export job.

export_task_default_timeout_second

The default timeout for export jobs.

export_running_job_num_limit

The maximum number of export jobs that can run concurrently. Set to 0 to remove the limit.

export_checker_interval_second

The interval at which the export checker runs.

sync_commit_interval_second

The maximum interval between transaction commits in a data synchronization channel. If data in the channel is not committed within this interval, the consumer notifies the channel to commit.

min_bytes_sync_commit

The minimum data size required to trigger a transaction commit. If the FE node receives less data than this threshold, it waits for the next sync_commit_interval_second cycle.

Set this value lower than the product of canal.instance.memory.buffer.size and canal.instance.memory.buffer.memunit in Canal (default: 16 MB). If this value exceeds that product, the FE node may attempt to read more data than the Canal store queue can hold, causing the queue to block until timeout.

max_bytes_sync_commit

The maximum number of threads in the data synchronization thread pool. There is one thread pool per FE node, used to dispatch sync tasks from the FE node to BE nodes.

plugin_dir

The directory where plugins are installed.

plugin_enable

Enables or disables the plugin system.

tmp_dir

The directory for temporary files, such as files generated during backup and restore operations. Files in this directory are cleared after each operation completes.

custom_config_dir

The directory where fe_custom.conf is stored. In some deployments, the conf/ directory is overwritten during system upgrades, which clears dynamically persisted parameters. Set this to a directory outside conf/ to protect those parameters from being overwritten.

small_file_dir

See Data loading.

spark_dpp_version

The Apache Spark version used for DPP jobs.

spark_home_default_dir

The directory containing the Spark client and Spark configuration files.

spark_launcher_log_dir

The directory for Spark launcher log files.

spark_resource_path

The directory containing Spark dependency packages.

yarn_config_dir

The directory containing YARN command configuration files. Verify that the required configuration files exist before running YARN commands; create them if they do not.

yarn_client_path

The path to the YARN client binary.

check_consistency_default_timeout_second

See Query and cache.

enable_deploy_manager

See Server and networking.

max_layout_length_per_row

See Data loading.