FisherYu/fuse-query

FuseQuery is a Cloud-Native SQL Query Engine at scale

FuseQuery

FuseQuery is a Cloud Distributed SQL Query Engine at scale.

Cloud-Native and Distributed ClickHouse from scratch in Rust.

Give thanks to ClickHouse and Arrow.

Features

• High Performance

  • Everything is Parallelism

• High Scalability

  • Everything is Distributed

• High Reliability

  • True Separation of Storage and Compute

Architecture

Crates

Crate	Description	Status
distributed	Distributed scheduler and executor for planner	WIP
optimizers	Optimizer for Distributed&Local plan	WIP
datablocks	Vectorized data processing unit	WIP
datastreams	Async streaming iterators	WIP
datasources	Interface to the datasource(system.numbers for performance/Fuse-Store)	WIP
executors	Executor(EXPLAIN/SELECT) for the Pipeline	WIP
functions	Scalar and Aggregation Functions	WIP
processors	Dataflow Streaming Processor	WIP
planners	Distributed&Local planners for building processor pipelines	WIP
servers	Server handler(MySQL/HTTP)	MySQL
transforms	Data Stream Transform(Source/Filter/Projection/AggregatorPartial/AggregatorFinal/Limit)	WIP

Status

SQL Support

• Projection
• Filter
• Limit
• Aggregate
• Functions
• Filter Push-Down
• Projection Push-Down (TODO)
• Distributed Query (WIP)
• Sorting (TODO)
• Joins (TODO)
• SubQueries (TODO)

Performance

• Memory SIMD-Vector processing performance only
• Dataset: 10,000,000,000 (10 Billion)
• Hardware: 8vCPUx16G Cloud Instance
• Rust: rustc 1.50.0-nightly (f76ecd066 2020-12-15)
• Build with Link-time Optimization and Using CPU Specific Instructions

Query	FuseQuery (v0.1)	ClickHouse (v19.17.6)
SELECT avg(number) FROM system.numbers_mt	(1.32 s.)	×1.29 (1.70 s) 5.90 billion rows/s., 47.16 GB/s
SELECT sum(number) FROM system.numbers_mt	(1.35 s.)	×0.99 (1.34 s) 7.48 billion rows/s., 59.80 GB/s
SELECT min(number) FROM system.numbers_mt	(1.34 s.)	×1.17 (1.57 s.) 6.36 billion rows/s., 50.89 GB/s
SELECT max(number) FROM system.numbers_mt	(1.32 s.)	×1.77 (2.33 s.) 4.34 billion rows/s., 34.74 GB/s
SELECT max(number+1) FROM system.numbers_mt	(3.77 s.)	×0.87 (3.29 s.) 3.04 billion rows/s., 24.31 GB/s
SELECT count(number) FROM system.numbers_mt	(1.31 s.)	×0.51 (0.67 s.) 15.00 billion rows/s., 119.99 GB/s
SELECT sum(number+number+number) FROM numbers_mt	(4.05 s.)	×1.22 (4.95 s.) 2.02 billion rows/s., 16.17 GB/s
SELECT sum(number) / count(number) FROM system.numbers_mt	(1.32 s.)	×0.97 (1.28 s.) 7.84 billion rows/s., 62.73 GB/s
SELECT sum(number) / count(number), max(number), min(number) FROM system.numbers_mt	(1.76 s.)	×2.29 (4.03 s.) 2.33 billion rows/s., 18.61 GB/s

Note:

• ClickHouse system.numbers_mt is 8-way parallelism processing
• FuseQuery system.numbers_mt is 8-way parallelism processing

How to Run?

Fuse-Query Server

Run from source

$ make run

12:46:15 [ INFO] Options { log_level: "debug", num_cpus: 8, mysql_handler_port: 3307 }
12:46:15 [ INFO] Fuse-Query Cloud Compute Starts...
12:46:15 [ INFO] Usage: mysql -h127.0.0.1 -P3307

or Run with docker(Recommended):

$ docker pull datafusedev/fuse-query
...

$ docker run --init --rm -p 3307:3307 datafusedev/fuse-query
05:12:36 [ INFO] Options { log_level: "debug", num_cpus: 6, mysql_handler_port: 3307 }
05:12:36 [ INFO] Fuse-Query Cloud Compute Starts...
05:12:36 [ INFO] Usage: mysql -h127.0.0.1 -P3307

Query with MySQL client

Connect

$ mysql -h127.0.0.1 -P3307

Explain

mysql> explain select (number+1) as c1, number/2 as c2 from system.numbers_mt(10000000) where (c1+c2+1) < 100 limit 3;
+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| explain                                                                                                                                                                                                                                                                                                               |
+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| └─ Limit: 3
  └─ Projection: (number + 1) as c1:UInt64, (number / 2) as c2:UInt64
    └─ Filter: (((c1 + c2) + 1) < 100)
      └─ ReadDataSource: scan parts [8](Read from system.numbers_mt table)                                                                                                                   |
| 
  └─ LimitTransform × 1 processor
    └─ Merge (LimitTransform × 8 processors) to (MergeProcessor × 1)
      └─ LimitTransform × 8 processors
        └─ ProjectionTransform × 8 processors
          └─ FilterTransform × 8 processors
            └─ SourceTransform × 8 processors                                |
+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2 rows in set (0.00 sec)

Select

mysql> select (number+1) as c1, number/2 as c2 from system.numbers_mt(10000000) where (c1+c2+1) < 100 limit 3;
+------+------+
| c1   | c2   |
+------+------+
|    1 |    0 |
|    2 |    0 |
|    3 |    1 |
+------+------+
3 rows in set (0.06 sec)

How to Test?

$ make test

Roadmap

• 0.1 support aggregation select
• 0.2 support distributed query (WIP)
• 0.3 support group by, order by
• 0.4 support join
• 0.5 support sub queries
• 0.6 support TPC-H benchmark