Hologres:Full-text search performance test on the PMC dataset

Prepare the test environment

Test resources:

• Hologres:

  • Compute resources: 48 CU

  • Version: V4.1.12

  • Number of shards: 6. If you increase the number of compute nodes, increase the number of shards linearly.

• ECS:

  • Instance type: ecs.c9i.16xlarge or ecs.g9i.16xlarge

  • Operating system: Debian 13.2 64-bit

Environment setup:

• Prepare a Hologres instance

  • Purchase a Hologres instance (V4.1) and create a database.

  • Create a user. For more information, see user management.

• Prepare an ECS instance

  • Purchase an ECS instance.

  • Install dependencies

    # Update the apt cache
    sudo apt update
    # Install the PostgreSQL client to connect to the database
    sudo apt install -y postgresql-client

  • Prepare the dataset: Download and decompress the pmc dataset from the official source:

    mkdir ~/data && cd ~/data
    wget http://benchmarks.elasticsearch.org.s3.amazonaws.com/corpora/pmc/documents.json.bz2
    bunzip2 -k documents.json.bz2

Run the performance test

An open-source benchmark tool developed by Hologres automates the performance test process described in this article, including table creation, data import, and index creation. For details about the tool, see the Git project alibabacloud-hologres-benchmark. For a table creation example, see the appendix.

• Install the benchmark tool

  # Create an isolated environment
  sudo apt install -y python3-venv
  python3 -m venv .venv
  
  # Activate the isolated environment
  source .venv/bin/activate
  python3 -m pip install -U pip
  
  # Install dependencies
  git clone https://github.com/aliyun/alibabacloud-hologres-benchmark.git
  cd alibabacloud-hologres-benchmark/fulltext_search/pmc
  pip3 install -r requirements.txt

• Modify the configuration file

  {
    "host": "<hologres_endpoint>",
    "port": <hologres_port>,
    "database": "<database_name>",
    "username": "<user_name>",
    "password": "<password>",
    "table_name": "pmc"
  }

• Run the benchmark script

  cd alibabacloud-hologres-benchmark/fulltext_search/pmc
  
  # This script runs the full benchmark process, including data import and queries. 
  # If the data already exists, the import step is skipped.
  python3 hologres_benchmark.py \
      --config config.json \
      --queries-config benchmark_queries.yaml \
      --data-dir ~/data

Test results

Results overview

Performance details: The following table shows the average response time in milliseconds for each query. Hologres delivers excellent performance for both full-text search queries (such as term and phrase) and aggregate analytics scenarios.

Appendix: Hologres table and index creation

• Create the test table in Hologres

  -- Create a new table group with 6 shards.
  CALL HG_CREATE_TABLE_GROUP ('tg_6', 6);
  
  -- Create the core table.
  CREATE TABLE pmc (
    id BIGINT PRIMARY KEY,
    name TEXT,
    journal TEXT,
    "date" TEXT,
    volume TEXT,
    issue TEXT,
    accession TEXT,
    "timestamp" timestamptz NOT NULL,
    pmid INTEGER,
    body TEXT
  ) WITH (
    table_group = 'tg_6',
    bitmap_columns = 'journal,"date",volume,issue,accession',
    segment_key = '"timestamp"',
    clustering_key = '"timestamp"',
    distribution_key = 'id'
  );

• Create full-text indexes in Hologres: Create inverted indexes on multiple fields.

  -- Create full-text indexes.
  CREATE INDEX pmc_accession_idx ON public.pmc USING fulltext(accession)
   WITH (
  tokenizer = 'keyword',
  analyzer_params = '{"tokenizer":{"type":"keyword"}}'
  );
  
  CREATE INDEX pmc_body_idx ON public.pmc USING fulltext(body)
   WITH (
  tokenizer = 'standard',
  analyzer_params = '{"filter":["lowercase"],"tokenizer":{"max_token_length":255,"type":"standard"}}'
  );
  
  CREATE INDEX pmc_date_idx ON public.pmc USING fulltext(date)
   WITH (
  tokenizer = 'standard',
  analyzer_params = '{"filter":["lowercase"],"tokenizer":{"max_token_length":255,"type":"standard"}}'
  );
  
  CREATE INDEX pmc_issue_idx ON public.pmc USING fulltext(issue)
   WITH (
  tokenizer = 'standard',
  analyzer_params = '{"filter":["lowercase"],"tokenizer":{"max_token_length":255,"type":"standard"}}'
  );
  
  CREATE INDEX pmc_journal_idx ON public.pmc USING fulltext(journal)
   WITH (
  tokenizer = 'standard',
  analyzer_params = '{"filter":["lowercase"],"tokenizer":{"max_token_length":255,"type":"standard"}}'
  );
  
  CREATE INDEX pmc_name_idx ON public.pmc USING fulltext(name)
   WITH (
  tokenizer = 'keyword',
  analyzer_params = '{"tokenizer":{"type":"keyword"}}'
  );
  
  CREATE INDEX pmc_volume_idx ON public.pmc USING fulltext(volume)
   WITH (
  tokenizer = 'standard',
  analyzer_params = '{"filter":["lowercase"],"tokenizer":{"max_token_length":255,"type":"standard"}}'
  );
  
  -- Perform a full index build.
  VACUUM pmc;