@Rocky Are there any benchmarks or workload generators for testing StarRocks' data loading and compaction performance?

StarRocks provides several tools and standard benchmarks to test data loading and compaction performance. While there isn't a single tool exclusively for "compaction," compaction is typically tested by generating high-frequency or high-volume loading workloads using standard benchmark suites. 1. Benchmark Suites & Workload Generators StarRocks maintains an official benchmarking repository and supports several industry-standard datasets: * Official Benchmark Tool (GitHub): The starrocks-benchmark repository contains scripts for end-to-end testing, including: * SSB (Star Schema Benchmark): Used for testing flat-table and multi-table query performance. It includes

gen-ssb.sh

for data generation and

stream_load.sh

for high-speed ingestion testing. * TPC-H & TPC-DS: Scripts like

gen-tpch.sh

and

stream_load.sh

are provided to generate datasets (SF1 to SF1000+) and load them via Stream Load. * Built-in Data Generation (v3.3+): In recent versions, StarRocks has introduced native support for generating TPC-H and TPC-DS data directly within the system using External Catalogs (e.g.,

CREATE EXTERNAL CATALOG tpch PROPERTIES ("type" = "TPCH");

). This eliminates the need for external

dbgen

tools and local disk space. * DuckDB for TPC-DS: A common community practice is using DuckDB to generate TPC-DS data in Parquet format, which can then be loaded into StarRocks using the

INSERT INTO ... SELECT * FROM FILES()

syntax. 2. Testing Data Loading Performance To stress-test loading specifically, you can use the following methods: * Stream Load Transaction Interface: For testing high-concurrency and atomicity, use the Stream Load API. * StarRocks Flink/Kafka Connectors: These can be configured with high parallelism to simulate real-time streaming workloads, which is the most common way to stress the compaction engine. * DataX / SMT: For bulk migration testing from other databases. 3. Measuring & Stressing Compaction Compaction performance is typically evaluated by monitoring how well the background threads keep up with a heavy loading workload. * Compaction Score: This is the primary metric. You can monitor it using:

sql
    SHOW PROC '/compactions';
    -- or for shared-data clusters
    SELECT * FROM information_schema.be_cloud_native_compactions;

If the Compaction Score exceeds 100, it indicates that compaction cannot keep up with the ingestion rate, and StarRocks will begin to throttle or reject new loads. * Manual Triggering: To test compaction performance in isolation (without loading), you can manually trigger it:

sql
    ALTER TABLE <table_name> COMPACT;

* Metrics for Compaction: Use the StarRocks Grafana Dashboard to monitor

Cumulative Compaction

and

Base Compaction

rates, as well as the "Tablet Version Count." Summary of Recommended Workflow 1. Generate Workload: Use the

starrocks-benchmark

scripts to run a high-frequency Stream Load test. 2. Monitor Pressure: Use

SHOW PROC '/transactions/{db}/running'

to see if "Partition's compaction score is larger than 100.0" appears in the error message, signaling a bottleneck. 3. Tuning: If compaction is the bottleneck, increase

compact_threads

or

update_compaction_num_threads_per_disk

in the BE configuration (

be.conf

). References • administration/management/compaction.md • knowledge/admin_and_operation/shared_data_compaction.md • best_practices/primarykey_table.md • faq/shared_data_faq.md