FE Configuration

FE Configuration

This document mainly introduces the relevant configuration items of FE.

The FE configuration file fe.conf is usually stored in the conf/ directory of the FE deployment path. In version 0.14, another configuration file fe_custom.conf will be introduced. The configuration file is used to record the configuration items that are dynamically configured and persisted by the user during operation.

After the FE process is started, it will read the configuration items in fe.conf first, and then read the configuration items in fe_custom.conf. The configuration items in fe_custom.conf will overwrite the same configuration items in fe.conf.

The location of the fe_custom.conf file can be configured in fe.conf through the custom_config_dir configuration item.

View configuration items

There are two ways to view the configuration items of FE:

FE web page

Open the FE web page http://fe_host:fe_http_port/variable in the browser. You can see the currently effective FE configuration items in Configure Info. 2. View by command

After the FE is started, you can view the configuration items of the FE in the MySQL client with the following command:

ADMIN SHOW FRONTEND CONFIG;

The meanings of the columns in the results are as follows:

Key: the name of the configuration item.
Value: The value of the current configuration item.
Type: The configuration item value type, such as integer or string.
IsMutable: whether it can be dynamically configured. If true, the configuration item can be dynamically configured at runtime. If false, it means that the configuration item can only be configured in fe.conf and takes effect after restarting FE.
MasterOnly: Whether it is a unique configuration item of Master FE node. If it is true, it means that the configuration item is meaningful only at the Master FE node, and is meaningless to other types of FE nodes. If false, it means that the configuration item is meaningful in all types of FE nodes.
Comment: The description of the configuration item.

Set configuration items

There are two ways to configure FE configuration items:

Static configuration

Add and set configuration items in the conf/fe.conf file. The configuration items in fe.conf will be read when the FE process starts. Configuration items not in fe.conf will use default values. 2. Dynamic configuration via MySQL protocol

After the FE starts, you can set the configuration items dynamically through the following commands. This command requires administrator privilege.

ADMIN SET FRONTEND CONFIG (" fe_config_name "=" fe_config_value ");

Not all configuration items support dynamic configuration. You can check whether the dynamic configuration is supported by the IsMutable column in theADMIN SHOW FRONTEND CONFIG;command result.

If the configuration item of MasterOnly is modified, the command will be directly forwarded to the Master FE and only the corresponding configuration item in the Master FE will be modified.

Configuration items modified in this way will become invalid after the FE process restarts.

For more help on this command, you can view it through the HELP ADMIN SET CONFIG; command. 3. Dynamic configuration via HTTP protocol

For details, please refer to Set Config Action

This method can also persist the modified configuration items. The configuration items will be persisted in the fe_custom.conf file and will still take effect after FE is restarted.

Examples

Modify async_pending_load_task_pool_size

Through ADMIN SHOW FRONTEND CONFIG; you can see that this configuration item cannot be dynamically configured (IsMutable is false). You need to add in fe.conf:

async_pending_load_task_pool_size = 20

Then restart the FE process to take effect the configuration. 2. Modify dynamic_partition_enable

Through ADMIN SHOW FRONTEND CONFIG; you can see that the configuration item can be dynamically configured (IsMutable is true). And it is the unique configuration of Master FE. Then first we can connect to any FE and execute the following command to modify the configuration:

1. ADMIN SET FRONTEND CONFIG ("dynamic_partition_enable" = "true"); `

Afterwards, you can view the modified value with the following command:

1. set forward_to_master = true; 2. ADMIN SHOW FRONTEND CONFIG;

After modification in the above manner, if the Master FE restarts or a Master election is performed, the configuration will be invalid. You can add the configuration item directly in fe.conf and restart the FE to make the configuration item permanent. 3. Modify max_distribution_pruner_recursion_depth

Through ADMIN SHOW FRONTEND CONFIG; you can see that the configuration item can be dynamically configured (IsMutable is true). It is not unique to Master FE.

Similarly, we can modify the configuration by dynamically modifying the configuration command. Because this configuration is not unique to the Master FE, user need to connect to different FEs separately to modify the configuration dynamically, so that all FEs use the modified configuration values.

Configurations

max_dynamic_partition_num

Default：500

IsMutable：true

MasterOnly：true

Used to limit the maximum number of partitions that can be created when creating a dynamic partition table, to avoid creating too many partitions at one time. The number is determined by “start” and “end” in the dynamic partition parameters..

grpc_max_message_size_bytes

Default：1G

Used to set the initial flow window size of the GRPC client channel, and also used to max message size. When the result set is large, you may need to increase this value.

min_replication_num_per_tablet

Default: 1

Used to set minimal number of replication per tablet.

max_replication_num_per_tablet

Default: 32767

Used to set maximal number of replication per tablet.

enable_outfile_to_local

Default：false Whether to allow the outfile function to export the results to the local disk.

enable_access_file_without_broker

Default：false

IsMutable：true

MasterOnly：true

This config is used to try skip broker when access bos or other cloud storage via broker

enable_bdbje_debug_mode

Default：false

If set to true, FE will be started in BDBJE debug mode

enable_alpha_rowset

Default：false

Whether to support the creation of alpha rowset tables. The default is false and should only be used in emergency situations, this config should be remove in some future version

enable_http_server_v2

Default：The default is true after the official 0.14.0 version is released, and the default is false before

HTTP Server V2 is implemented by SpringBoot. It uses an architecture that separates the front and back ends. Only when httpv2 is enabled can users use the new front-end UI interface.

jetty_server_acceptors

Default：2

jetty_server_selectors

Default：4

jetty_server_workers

Default：0

With the above three parameters, Jetty’s thread architecture model is very simple, divided into acceptors, selectors and workers three thread pools. Acceptors are responsible for accepting new connections, and then hand them over to selectors to process the unpacking of the HTTP message protocol, and finally workers process the request. The first two thread pools adopt a non-blocking model, and one thread can handle the read and write of many sockets, so the number of thread pools is small.

For most projects, only 1-2 acceptors threads are required, and 2 to 4 selectors threads are sufficient. Workers are obstructive business logic, often have more database operations, and require a large number of threads. The specific number depends on the proportion of QPS and IO events of the application. The higher the QPS, the more threads are required, the higher the proportion of IO, the more threads waiting, and the more total threads required.

Worker thread pool is not set by default, set according to your needs

jetty_server_max_http_post_size

Default：100 * 1024 * 1024 （100MB）

This is the maximum number of bytes of the file uploaded by the put or post method, the default value: 100MB

frontend_address

Status: Deprecated, not recommended use. This parameter may be deleted later Type: string Description: Explicitly set the IP address of FE instead of using InetAddress.getByName to get the IP address. Usually in InetAddress.getByName When the expected results cannot be obtained. Only IP address is supported, not hostname. Default value: 0.0.0.0

default_max_filter_ratio

Default：0

IsMutable：true

MasterOnly：true

Maximum percentage of data that can be filtered (due to reasons such as data is irregularly) , The default value is 0.

default_db_data_quota_bytes

Default：1PB

IsMutable：true

MasterOnly：true

Used to set the default database data quota size. To set the quota size of a single database, you can use:

1. Set the database data quota, the unit is:B/K/KB/M/MB/G/GB/T/TB/P/PB
2. ALTER DATABASE db_name SET DATA QUOTA quota;
3. View configuration
4. show data （Detail：HELP SHOW DATA）

default_db_replica_quota_size

Default: 1073741824

IsMutable：true

MasterOnly：true

Used to set the default database replica quota. To set the quota size of a single database, you can use:

1. Set the database replica quota
2. ALTER DATABASE db_name SET REPLICA QUOTA quota;
3. View configuration
4. show data （Detail：HELP SHOW DATA）

enable_batch_delete_by_default

Default：false

IsMutable：true

MasterOnly：true

Whether to add a delete sign column when create unique table

recover_with_empty_tablet

Default：false

IsMutable：true

MasterOnly：true

In some very special circumstances, such as code bugs, or human misoperation, etc., all replicas of some tablets may be lost. In this case, the data has been substantially lost. However, in some scenarios, the business still hopes to ensure that the query will not report errors even if there is data loss, and reduce the perception of the user layer. At this point, we can use the blank Tablet to fill the missing replica to ensure that the query can be executed normally.

Set to true so that Doris will automatically use blank replicas to fill tablets which all replicas have been damaged or missing

max_allowed_in_element_num_of_delete

Default：1024

IsMutable：true

MasterOnly：true

This configuration is used to limit element num of InPredicate in delete statement.

cache_result_max_row_count

Default：3000

IsMutable：true

MasterOnly：false

In order to avoid occupying too much memory, the maximum number of rows that can be cached is 2000 by default. If this threshold is exceeded, the cache cannot be set

cache_last_version_interval_second

Default：900

IsMutable：true

MasterOnly：false

The time interval of the latest partitioned version of the table refers to the time interval between the data update and the current version. It is generally set to 900 seconds, which distinguishes offline and real-time import

cache_enable_partition_mode

Default：true

IsMutable：true

MasterOnly：false

When this switch is turned on, the query result set will be cached according to the partition. If the interval between the query table partition time and the query time is less than cache_last_version_interval_second, the result set will be cached according to the partition.

Part of the data will be obtained from the cache and some data from the disk when querying, and the data will be merged and returned to the client.

cache_enable_sql_mode

Default：true

IsMutable：true

MasterOnly：false

If this switch is turned on, the SQL query result set will be cached. If the interval between the last visit version time in all partitions of all tables in the query is greater than cache_last_version_interval_second, and the result set is less than cache_result_max_row_count, the result set will be cached, and the next same SQL will hit the cache

If set to true, fe will enable sql result caching. This option is suitable for offline data update scenarios

	case1	case2	case3	case4
enable_sql_cache	false	true	true	false
enable_partition_cache	false	false	true	true

min_clone_task_timeout_sec 和 max_clone_task_timeout_sec

Default：Minimum 3 minutes, maximum two hours

IsMutable：true

MasterOnly：true

Type: long Description: Used to control the maximum timeout of a clone task. The unit is second. Default value: 7200 Dynamic modification: yes

Can cooperate with mix_clone_task_timeout_sec to control the maximum and minimum timeout of a clone task. Under normal circumstances, the timeout of a clone task is estimated by the amount of data and the minimum transfer rate (5MB/s). In some special cases, these two configurations can be used to set the upper and lower bounds of the clone task timeout to ensure that the clone task can be completed successfully.

agent_task_resend_wait_time_ms

Default：5000

IsMutable：true

MasterOnly：true

This configuration will decide whether to resend agent task when create_time for agent_task is set, only when current_time - create_time > agent_task_resend_wait_time_ms can ReportHandler do resend agent task.

This configuration is currently mainly used to solve the problem of repeated sending of PUBLISH_VERSION agent tasks. The current default value of this configuration is 5000, which is an experimental value.

Because there is a certain time delay between submitting agent tasks to AgentTaskQueue and submitting to be, Increasing the value of this configuration can effectively solve the problem of repeated sending of agent tasks,

But at the same time, it will cause the submission of failed or failed execution of the agent task to be executed again for an extended period of time

enable_odbc_table

Default：false

IsMutable：true

MasterOnly：true

Whether to enable the ODBC table, it is not enabled by default. You need to manually configure it when you use it. This parameter can be set by: ADMIN SET FRONTEND CONFIG(“key”=”value”)

enable_spark_load

Default：false

IsMutable：true

MasterOnly：true

Whether to enable spark load temporarily, it is not enabled by default

disable_storage_medium_check

Default：false

IsMutable：true

MasterOnly：true

If disable_storage_medium_check is true, ReportHandler would not check tablet’s storage medium and disable storage cool down function, the default value is false. You can set the value true when you don’t care what the storage medium of the tablet is.

drop_backend_after_decommission

Default：false

IsMutable：true

MasterOnly：true

This configuration is used to control whether the system drops the BE after successfully decommissioning the BE. If true, the BE node will be deleted after the BE is successfully offline. If false, after the BE successfully goes offline, the BE will remain in the DECOMMISSION state, but will not be dropped.

This configuration can play a role in certain scenarios. Assume that the initial state of a Doris cluster is one disk per BE node. After running for a period of time, the system has been vertically expanded, that is, each BE node adds 2 new disks. Because Doris currently does not support data balancing among the disks within the BE, the data volume of the initial disk may always be much higher than the data volume of the newly added disk. At this time, we can perform manual inter-disk balancing by the following operations:

Set the configuration item to false.
Perform a decommission operation on a certain BE node. This operation will migrate all data on the BE to other nodes.
After the decommission operation is completed, the BE will not be dropped. At this time, cancel the decommission status of the BE. Then the data will start to balance from other BE nodes back to this node. At this time, the data will be evenly distributed to all disks of the BE.
Perform steps 2 and 3 for all BE nodes in sequence, and finally achieve the purpose of disk balancing for all nodes

period_of_auto_resume_min

Default：5 （s）

IsMutable：true

MasterOnly：true

Automatically restore the cycle of Routine load

max_tolerable_backend_down_num

Default：0

IsMutable：true

MasterOnly：true

As long as one BE is down, Routine Load cannot be automatically restored

enable_materialized_view

Default：true

IsMutable：true

MasterOnly：true

This configuration is used to turn on and off the creation of materialized views. If set to true, the function to create a materialized view is enabled. The user can create a materialized view through the CREATE MATERIALIZED VIEW command. If set to false, materialized views cannot be created.

If you get an error The materialized view is coming soon or The materialized view is disabled when creating the materialized view, it means that the configuration is set to false and the function of creating the materialized view is turned off. You can start to create a materialized view by modifying the configuration to true.

This variable is a dynamic configuration, and users can modify the configuration through commands after the FE process starts. You can also modify the FE configuration file and restart the FE to take effect

check_java_version

Default：true

Doris will check whether the compiled and run Java versions are compatible, if not, it will throw a Java version mismatch exception message and terminate the startup

max_running_rollup_job_num_per_table

Default：1

IsMutable：true

MasterOnly：true

Control the concurrency limit of Rollup jobs

dynamic_partition_enable

Default：true

IsMutable：true

MasterOnly：true

Whether to enable dynamic partition, enabled by default

dynamic_partition_check_interval_seconds

Default：600 （s）

IsMutable：true

MasterOnly：true

Decide how often to check dynamic partition

disable_cluster_feature

Default：true

IsMutable：true

The multi cluster feature will be deprecated in version 0.12 ，set this config to true will disable all operations related to cluster feature, include: create/drop cluster add free backend/add backend to cluster/decommission cluster balance change the backends num of cluster link/migration db

force_do_metadata_checkpoint

Default：false

IsMutable：true

MasterOnly：true

If set to true, the checkpoint thread will make the checkpoint regardless of the jvm memory used percent

metadata_checkpoint_memory_threshold

Default：60 （60%）

IsMutable：true

MasterOnly：true

If the jvm memory used percent(heap or old mem pool) exceed this threshold, checkpoint thread will not work to avoid OOM.

max_distribution_pruner_recursion_depth

Default：100

IsMutable：true

MasterOnly：false

This will limit the max recursion depth of hash distribution pruner. eg: where a in (5 elements) and b in (4 elements) and c in (3 elements) and d in (2 elements). a/b/c/d are distribution columns, so the recursion depth will be 5 * 4 * 3 * 2 = 120, larger than 100, So that distribution pruner will no work and just return all buckets. Increase the depth can support distribution pruning for more elements, but may cost more CPU.

max_backup_restore_job_num_per_db

Default: 10

This configuration is mainly used to control the number of backup/restore tasks recorded in each database.

using_old_load_usage_pattern

Default：false

IsMutable：true

MasterOnly：true

Default：100

the max routine load job num, including NEED_SCHEDULED, RUNNING, PAUSE

max_running_txn_num_per_db

Default：100

IsMutable：true

MasterOnly：true

This configuration is mainly used to control the number of concurrent load jobs of the same database.

When there are too many load jobs running in the cluster, the newly submitted load jobs may report errors:

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

When this error is encountered, it means that the load jobs currently running in the cluster exceeds the configuration value. At this time, it is recommended to wait on the business side and retry the load jobs.

Generally it is not recommended to increase this configuration value. An excessively high number of concurrency may cause excessive system load

enable_metric_calculator

Default：true

If set to true, metric collector will be run as a daemon timer to collect metrics at fix interval

report_queue_size

Default： 100

IsMutable：true

MasterOnly：true

This threshold is to avoid piling up too many report task in FE, which may cause OOM exception. In some large Doris cluster, eg: 100 Backends with ten million replicas, a tablet report may cost several seconds after some modification of metadata(drop partition, etc..). And one Backend will report tablets info every 1 min, so unlimited receiving reports is unacceptable. we will optimize the processing speed of tablet report in future, but now, just discard the report if queue size exceeding limit. Some online time cost: 1. disk report: 0-1 msta 2. sk report: 0-1 ms 3. tablet report 4. 10000 replicas: 200ms

partition_rebalance_max_moves_num_per_selection

Default：10

IsMutable：true

MasterOnly：true

Valid only if use PartitionRebalancer，

partition_rebalance_move_expire_after_access

Default：600 (s)

IsMutable：true

MasterOnly：true

Valid only if use PartitionRebalancer. If this changed, cached moves will be cleared

Default：2

the default slot number per path in tablet scheduler , remove this config and dynamically adjust it by clone task statistic

tablet_repair_delay_factor_second

Default：60 （s）

IsMutable：true

MasterOnly：true

the factor of delay time before deciding to repair tablet. if priority is VERY_HIGH, repair it immediately.

HIGH, delay tablet_repair_delay_factor_second * 1;
NORMAL: delay tablet_repair_delay_factor_second * 2;
LOW: delay tablet_repair_delay_factor_second * 3;

es_state_sync_interval_second

Default：10

fe will call es api to get es index shard info every es_state_sync_interval_secs

disable_hadoop_load

Default：false

IsMutable：true

MasterOnly：true

Load using hadoop cluster will be deprecated in future. Set to true to disable this kind of load.

db_used_data_quota_update_interval_secs

Default：300 (s)

IsMutable：true

MasterOnly：true

For better data load performance, in the check of whether the amount of data used by the database before data load exceeds the quota, we do not calculate the amount of data already used by the database in real time, but obtain the periodically updated value of the daemon thread.

This configuration is used to set the time interval for updating the value of the amount of data used by the database

disable_load_job

Default：false

IsMutable：true

MasterOnly：true

if this is set to true

all pending load job will failed when call begin txn api
all prepare load job will failed when call commit txn api
all committed load job will waiting to be published

catalog_try_lock_timeout_ms

Default：5000 （ms）

IsMutable：true

The tryLock timeout configuration of catalog lock. Normally it does not need to change, unless you need to test something.

max_query_retry_time

Default：1

IsMutable：true

The number of query retries. A query may retry if we encounter RPC exception and no result has been sent to user. You may reduce this number to avoid Avalanche disaster

remote_fragment_exec_timeout_ms

Default：5000 （ms）

IsMutable：true

The timeout of executing async remote fragment. In normal case, the async remote fragment will be executed in a short time. If system are under high load condition，try to set this timeout longer.

enable_local_replica_selection

Default：false

IsMutable：true

If set to true, Planner will try to select replica of tablet on same host as this Frontend. This may reduce network transmission in following case:

N hosts with N Backends and N Frontends deployed.
The data has N replicas.
High concurrency queries are syyuyuient to all Frontends evenly
In this case, all Frontends can only use local replicas to do the query. If you want to allow fallback to nonlocal replicas when no local replicas available, set enable_local_replica_selection_fallback to true.

Default：true

if set to false, auth check will be disable, in case some goes wrong with the new privilege system.

tablet_stat_update_interval_second

Default：300，（5min）

update interval of tablet stat , All frontends will get tablet stat from all backends at each interval

storage_flood_stage_usage_percent

Default：95 （95%）

IsMutable：true

MasterOnly：true

storage_flood_stage_left_capacity_bytes

Default：

1. storage_flood_stage_usage_percent  : 95  (95%)

3. storage_flood_stage_left_capacity_bytes :  1 * 1024 * 1024 * 1024 (1GB)

IsMutable：true

MasterOnly：true

If capacity of disk reach the ‘storage_flood_stage_usage_percent’ and ‘storage_flood_stage_left_capacity_bytes’, the following operation will be rejected:

.load job
restore job

Default：false

If use k8s deploy manager locally, set this to true and prepare the certs files

dpp_hadoop_client_path

Default：/lib/hadoop-client/hadoop/bin/hadoop

dpp_bytes_per_reduce

Default：100 * 1024 * 1024L; // 100M

dpp_default_cluster

Default：palo-dpp

dpp_default_config_str

Default：{ “hadoop_configs : ‘“ “mapred.job.priority=NORMAL;” “mapred.job.map.capacity=50;” “mapred.job.reduce.capacity=50;” “mapred.hce.replace.streaming=false;” “abaci.long.stored.job=true;” “dce.shuffle.enable=false;” “dfs.client.authserver.force_stop=true;” “dfs.client.auth.method=0” “‘}

dpp_config_str

Default：{palo-dpp : {“ + “hadoop_palo_path : ‘/dir’,” + “hadoop_configs : ‘“ + “fs.default.name=hdfs://host:port;” + “mapred.job.tracker=host:port;” + “hadoop.job.ugi=user,password” + “‘}” + “}

enable_deploy_manager

Default：disable

Set to true if you deploy Palo using thirdparty deploy manager Valid options are:

disable: no deploy manager
k8s: Kubernetes
ambari: Ambari
local: Local File (for test or Boxer2 BCC version)

enable_token_check

Default：true

For forward compatibility, will be removed later. check token when download image file.

expr_depth_limit

Default：3000

IsMutable：true

Limit on the depth of an expr tree. Exceed this limit may cause long analysis time while holding db read lock. Do not set this if you know what you are doing

expr_children_limit

Default：10000

IsMutable：true

Limit on the number of expr children of an expr tree. Exceed this limit may cause long analysis time while holding database read lock.

proxy_auth_magic_prefix

Default：x@8

proxy_auth_enable

Default：false

meta_publish_timeout_ms

Default：1000 （ms）

The default user resource publishing timeout

disable_colocate_balance

Default：false

IsMutable：true

MasterOnly：true

This configs can set to true to disable the automatic colocate tables’s relocate and balance. If ‘disable_colocate_balance’ is set to true, ColocateTableBalancer will not relocate and balance colocate tables. Attention:

Under normal circumstances, there is no need to turn off balance at all.
Because once the balance is turned off, the unstable colocate table may not be restored
Eventually the colocate plan cannot be used when querying.

query_colocate_join_memory_limit_penalty_factor

Default：1

IsMutable：true

colocote join PlanFragment instance的memory_limit = exec_mem_limit / min (query_colocate_join_memory_limit_penalty_factor, instance_num)

max_connection_scheduler_threads_num

Default：4096

Default：5

Export checker’s running interval.

default_load_parallelism

Default: 1

IsMutable：true

MasterOnly：true

Default parallelism of the broker load execution plan on a single node. If the user to set the parallelism when the broker load is submitted, this parameter will be ignored.

max_broker_concurrency

Default：10

IsMutable：true

MasterOnly：true

Maximal concurrency of broker scanners.

min_bytes_per_broker_scanner

Default：67108864L (64M)

IsMutable：true

MasterOnly：true

Minimum bytes that a single broker scanner will read.

catalog_trash_expire_second

Default：86400L (1day)

IsMutable：true

MasterOnly：true

After dropping database(table/partition), you can recover it by using RECOVER stmt. And this specifies the maximal data retention time. After time, the data will be deleted permanently.

storage_cooldown_second

Default：30 * 24 * 3600L （30day）

Default：0.2

IsMutable：true

MasterOnly：true

Balance threshold of data size in BE. The balance algorithm is:

Calculate the average used capacity(AUC) of the entire cluster. (total data size / total backends num)
The high water level is (AUC * (1 + clone_capacity_balance_threshold))
The low water level is (AUC * (1 - clone_capacity_balance_threshold))
The Clone checker will try to move replica from high water level BE to low water level BE.

clone_checker_interval_second

Default：300 （5min）

Clone checker’s running interval

tablet_delete_timeout_second

Default：2

IsMutable：true

MasterOnly：true

Same meaning as tablet_create_timeout_second, but used when delete a tablet.

async_loading_load_task_pool_size

Default：10

IsMutable：false

MasterOnly：true

The loading_load task executor pool size. This pool size limits the max running loading_load tasks.

Currently, it only limits the loading_load task of broker load

async_pending_load_task_pool_size

Default：10

IsMutable：false

Default：PaloFe.DORIS_HOME_DIR + “/lib/yarn-config”

Default yarn config file directory ，Each time before running the yarn command, we need to check that the config file exists under this path, and if not, create them.

yarn_client_path

Default：DORIS_HOME_DIR + “/lib/yarn-client/hadoop/bin/yarn”

Default yarn client path

spark_launcher_log_dir

Default： sys_log_dir + “/spark_launcher_log”

The specified spark launcher log dir

spark_resource_path

Default：none

Default spark dependencies path

spark_home_default_dir

Default：DORIS_HOME_DIR + “/lib/spark2x”

Default spark home dir

spark_load_default_timeout_second

Default：86400 (1天)

IsMutable：true

MasterOnly：true

Default：100000

IsMutable：true

MasterOnly：true

Maximal memory layout length of a row. default is 100 KB. In BE, the maximal size of a RowBlock is 100MB(Configure as max_unpacked_row_block_size in be.conf). And each RowBlock contains 1024 rows. So the maximal size of a row is approximately 100 KB. eg. schema: k1(int), v1(decimal), v2(varchar(2000)) then the memory layout length of a row is: 4(int) + 16(decimal) + 2000(varchar) = 2020 (Bytes) See memory layout length of all types, run ‘help create table’ in mysql-client. If you want to increase this number to support more columns in a row, you also need to increase the max_unpacked_row_block_size in be.conf. But the performance impact is unknown.

load_straggler_wait_second

Default：300

IsMutable：true

MasterOnly：true

Maximal wait seconds for straggler node in load eg. there are 3 replicas A, B, C load is already quorum finished(A,B) at t1 and C is not finished if (current_time - t1) > 300s, then palo will treat C as a failure node will call transaction manager to commit the transaction and tell transaction manager that C is failed

1. This is also used when waiting for publish tasks
2. this parameter is the default value for all job and the DBA could specify it for separate job

thrift_server_max_worker_threads

Default：4096

The thrift server max worker threads

publish_version_interval_ms

Default：10 （ms）

minimal intervals between two publish version action

publish_version_timeout_second

Default：30 （s）

IsMutable：true

MasterOnly：true

Maximal waiting time for all publish version tasks of one transaction to be finished

max_create_table_timeout_second

Default：60 （s）

IsMutable：true

MasterOnly：true

In order not to wait too long for create table(index), set a max timeout.

tablet_create_timeout_second

Default：1（s）

IsMutable：true

MasterOnly：true

Maximal waiting time for creating a single replica. eg. if you create a table with #m tablets and #n replicas for each tablet, the create table request will run at most (m * n * tablet_create_timeout_second) before timeout.

max_mysql_service_task_threads_num

Default：4096

When FeEstarts the MySQL server based on NIO model, the number of threads responsible for Task events. Only mysql_service_nio_enabled is true takes effect.

（本文超出站点单文档体积上限，已截断）