Introduction
Quick Start
Table Design
Data Loading
Data Export
Using StarRocks
Reference
- SQL Reference
  - User Account Management
    - ALTER USER
    - CREATE ROLE
    - CREATE USER
    - DROP ROLE
    - DROP USER
    - GRANT
    - REVOKE
    - SET PASSWORD
    - SET PROPERTY
    - SHOW GRANTS
    - SHOW ROLES
  - Cluster Management
  - DDL
  - DML
  - Data Type
    - Numeric
      - BIGINT
      - LARGEINT
      - SMALLINT
      - TINYINT
      - BOOLEAN
      - DECIMAL
      - DOUBLE
      - FLOAT
      - INT
    - String
      - CHAR
      - VARCHAR
      - STRING
    - Date
      - DATE
      - DATETIME
    - Others
      - HLL
      - BITMAP
  - Auxiliary Commands
    - DESCRIBE
- Function Reference
  - Date Functions
    - convert_tz
    - curdate
    - current_timestamp
    - curtime
    - datediff
    - date_add
    - date_format
    - date_sub
    - date_trunc
    - day
    - dayname
    - dayofmonth
    - dayofweek
    - dayofyear
    - from_days
    - from_unixtime
    - hour
    - minute
    - month
    - monthname
    - now
    - second
    - str_to_date
    - timediff
    - timestampadd
    - timestampdiff
    - to_date
    - to_days
    - unix_timestamp
    - utc_timestamp
    - weekofyear
    - year
  - Geographic Functions
  - String Functions
    - append_trailing_char_if_absent
    - ascii
    - char_length
    - concat
    - concat_ws
    - ends_with
    - find_in_set
    - get_json_double
    - get_json_int
    - get_json_string
    - group_concat
    - instr
    - lcase
    - left
    - length
    - locate
    - lower
    - lpad
    - ltrim
    - money_format
    - null_or_empty
    - regexp_extract
    - regexp_replace
    - repeat
    - reverse
    - right
    - rpad
    - split
    - split_part
    - starts_with
    - strleft
    - strright
  - Aggregation Functions
    - approx_count_distinct
    - avg
    - bitmap
    - count
    - hll_union_agg
    - max
    - min
    - percentile_approx
    - stddev
    - stddev_samp
    - sum
    - variance
    - var_samp
    - any_value
  - Bitmap Functions
  - Array Functions
  - cast function
  - hash function
  - Crytographic Functions
    - md5
    - sha2
- System Variables
- Error Code
- System Limit
Administration
- Build in Docker
- Manage a cluster
- Configuration
- Scale In and Out
- Backup and Restore
- Data Recovery
  - Data Recovery
  - Restore FEs
- Replica
- Query Planning
- Profiling
- Monitor and Alert
- User Privilege
- Authentication
- Resource Management
FAQ
- Deployment
- Data Migration
  - Data Ingestion
  - Data Export
- SQL
- Others FAQs
Benchmark
- SSB Benchmark
- TPC-H Benchmark
Release Notes
- v2.2
- v2.1
- v2.0
- v1.19

Architecture

System Architecture Diagram

architecture

Components

The StarRocks cluster is composed of FE and BE, and can be accessed via MySQL client.

FE

Connects to the MySQL client, parses and executes SQL statements.
Manages metadata, executes SQL DDL commands, and uses Catalog to record information about databases, tables, partitions, and tablet replications.
High-availability deployment. Use unique replication protocol to select master and synchronize metadata.
FE leader node modifies metadata; FE follower node performs read operations. Metadata read and write satisfies sequential consistency.
The number of FE nodes is 2n+1, which can tolerate n node failures. If the FE leader node fails, one of the follower nodes gets selected as the new leader node to complete the failover.
FE's SQL layer parses, analyzes, and rewrites users’ SQL statements. It also performs semantic analysis and relational algebra optimization, and produces logical execution plans.
FE Planner is responsible for transforming logical plans into physical plans that can be distributed to BE.
FE supervises BE, manages BE's online and offline, and maintains the number of tablet copies based on BE's health status.
FE coordinates data import to ensure data consistency.

BE

Manages tablet copies. Tablets are sub-tables with columnar storage, formed by table partitioning and hashing.
Creates or deletes sub-tables, instructed by FE.
Receives physical execution plans distributed by FE. The BE coordinator cooperates with other BE workers to complete executions.
Obtains data from the local column storage engine, and quickly filters it through indexing and predicate pushdown.
Executes compact tasks in the background to reduce read amplification during queries.
During data import, FE designates the BE coordinator to write fanout-form data to the BE where the multiple copies of the tablet are located.

Other Components

Hdfs Broker: used for importing data from Hdfs to StarRocks cluster, see Data Import section.

In this article

Architecture
- System Architecture Diagram
- Components