Configuration Reference

This document is for operators and developers who use Doris CCR (Cross-Cluster Replication). It collects the configuration items at the FE, BE, and database/table levels that require attention or adjustment, and provides default values and applicable versions so you can quickly cross-check them during setup and operation.

Applicable Scenarios

Scenario	Configuration area to focus on
Initial setup of a CCR sync pipeline	FE basic parameters, database/table binlog properties
Synced tables contain inverted indexes or bitmap indexes	restore_reset_index_id
Upstream frequently creates temporary partitions, causing sync interruptions	ignore_backup_tmp_partitions
A single table has more than 100,000 tablets	restore_job_compressed_serialization, backup_job_compressed_serialization
Network bandwidth is ample and you want to speed up incremental sync	BE download-related parameters
Backup/restore jobs are large in scale and long-running	backup_job_default_timeout_ms, thrift_max_message_size

Prerequisites

• Two Doris clusters (upstream and downstream) have been deployed, and the CCR Syncer has been installed and connected.
• You have permission to modify the FE and BE configuration files fe.conf / be.conf of both upstream and downstream clusters.
• The current cluster version is known, so you can determine whether each configuration item takes effect.

Configuration Workflow Overview

1. Modify fe.conf on the FE nodes of both upstream and downstream clusters, and restart or apply dynamically as needed.
2. Modify be.conf on the BE nodes of both upstream and downstream clusters, and restart or apply dynamically as needed.
3. Set database/table binlog properties through CREATE TABLE / ALTER TABLE.
4. Start or restart the CCR sync task, observe its running state, and tune the parameters as needed.

FE Configuration

Configure in fe.conf, for example:

restore_reset_index_id = true

Name	Description	Default value	Version
restore_reset_index_id	If synced tables use inverted indexes or bitmap indexes, set this to false.	false	Starting from 2.1.8 and 3.0.4.
ignore_backup_tmp_partitions	To avoid sync interruptions caused by upstream creating tmp partition, set this to true.	false	Starting from 2.1.8 and 3.0.4.
max_backup_restore_job_num_per_db	The limit on the number of in-memory backup/restore jobs per DB. The recommended value is 2.	10	All versions.
label_num_threshold	Controls the number of TXN Labels to prevent transactions from being recycled too quickly. A value that is too large consumes more memory; a value that is too small may cause data duplication under abnormal conditions. The default value is sufficient in most cases.	2000	Starting from 2.1.
restore_job_compressed_serialization	Recommended to set to true when the tablet count exceeds 100,000. Before downgrading, disable this configuration and ensure the FE completes a checkpoint. When upgrading from 2.1 to 3.0, upgrade to at least 3.0.3.	false	Starting from 2.1.8 and 3.0.3.
backup_job_compressed_serialization	Recommended to set to true when the tablet count exceeds 100,000. Before downgrading, disable this configuration and ensure the FE completes a checkpoint. When upgrading from 2.1 to 3.0, upgrade to at least 3.0.3.	false	Starting from 2.1.8 and 3.0.3.
backup_job_default_timeout_ms	Timeout for backup/restore jobs. Must be configured on the FE of both the source and target clusters.	None	Set as needed
enable_restore_snapshot_rpc_compression	Enables snapshot info compression to reduce RPC message size. Recommended to set to true.	true	Starting from 2.1.8 and 3.0.3.

BE Configuration

Configure in be.conf, for example:

thrift_max_message_size = 2000000000

Name	Description	Default value	Version
thrift_max_message_size	Maximum size of a single RPC packet on the BE thrift server. When the CCR job involves a large number of tablets, set this to 2000000000.	100MB	All versions
be_thrift_max_pkg_bytes	Maximum size of BE Thrift RPC message packets.	20MB	Specific to 2.0.
max_download_speed_kbps	Download rate limit per download worker on the downstream BE. Default is 50MB/s per thread.	50MB/s	All versions
download_worker_count	Number of threads for download tasks. Set according to the network card, disk, and workload.	1	All versions

Database/Table Properties

Set through CREATE TABLE or ALTER TABLE to control binlog behavior at the database/table level.

Name	Description	Default value	Version
binlog.max_bytes	Maximum memory footprint of the binlog. Recommended to reserve at least 4GB.	Unlimited	All versions
binlog.ttl_seconds	Binlog retention time.	Unlimited before 2.0.5; 1 day (86400) starting from 2.0.5	All versions

Common Tuning Combinations

The table below summarizes common configuration combinations chosen based on runtime behavior, so you can apply them quickly as needed:

Symptom / Goal	Recommended configuration
Synced tables contain inverted index / bitmap index	FE: restore_reset_index_id = false
Upstream frequently creates temporary partitions, causing sync interruptions	FE: ignore_backup_tmp_partitions = true
Backup/restore jobs pile up on a single DB and FE memory is tight	FE: max_backup_restore_job_num_per_db = 2
A single table has more than 100,000 tablets	FE: restore_job_compressed_serialization = true, backup_job_compressed_serialization = true; also enable enable_restore_snapshot_rpc_compression = true
Backup/restore jobs are very large and the default timeout is insufficient	FE: increase backup_job_default_timeout_ms as needed (must be configured on both source and target clusters)
RPC packets are too large, causing backup/restore failures	BE: thrift_max_message_size = 2000000000
Downstream download speed is limited and incremental sync is slow	BE: increase download_worker_count, and adjust max_download_speed_kbps as needed
Control the binlog size of a single table to prevent memory bloat	Database/table: set binlog.max_bytes (recommended to reserve at least 4GB) and binlog.ttl_seconds

Notes on Version Upgrades and Downgrades

• After enabling restore_job_compressed_serialization or backup_job_compressed_serialization, if you need to downgrade, disable the corresponding configuration first and ensure the FE completes a checkpoint before performing the downgrade.
• When upgrading from 2.1 to 3.0, the target version must be at least 3.0.3 to match the implementation of the compressed serialization configurations.
• binlog.ttl_seconds defaulted to unlimited before 2.0.5 and defaults to 1 day (86400 seconds) starting from 2.0.5. After upgrading from an older version, set it explicitly according to your storage and replay requirements.