Datasets:

robworks-software
/

database-query-logs-synthetic

license: cc0-1.0
task_categories:
  - text-classification
  - text-generation
  - tabular-classification
  - tabular-regression
language:
  - en
tags:
  - synthetic
  - template-generated
  - database
  - sql
  - query-optimization
  - performance
  - nosql
  - mongodb
  - redis
  - cassandra
  - elasticsearch
  - mysql
  - postgresql
  - oracle
  - sql-server
pretty_name: Database Query Logs (synthetic)
size_categories:
  - 1K<n<10K

Database Query Logs (synthetic)

3,995 database query-log entries spanning 10 engines - MySQL, PostgreSQL, MongoDB, SQL Server, Oracle, MariaDB, SQLite, Cassandra, Redis, and Elasticsearch - with query text, type, complexity, execution timing, and row-count metadata.

These queries are synthetic

The queries were programmatically generated, not captured from production systems. They were produced by templating a set of query shapes across industry-flavored schema names.

The generation shows through:

3,195 training rows contain 2,243 distinct queries (70.2%); one Elasticsearch query body appears 35 times.
The same query recurs with only the schema name swapped - DELETE FROM telecom_logs and DELETE FROM construction_logs are otherwise byte-identical.
Some generated SQL is not valid. Schema names containing hyphens appear unquoted (DELETE FROM E-COMMERCE.AUDIT_LOGS), which will not parse.

A previous version of this card described the contents as "real-world and synthetic". There is no verified production-captured subset; treat the whole dataset as synthetic.

Loading

from datasets import load_dataset

ds = load_dataset("robworks-software/database-query-logs-synthetic")

Splits

Split	Rows
train	3,195
validation	396
test	404
total	3,995

Appropriate use

Parser and tokenizer testing across dialects (bearing in mind some queries are invalid).
Query-type and complexity classification.
Teaching examples of dialect syntax differences.

Inappropriate use

Performance modeling or optimizer research. execution_time_ms, rows_examined, and rows_returned are generated values. They were not measured on any real system and have no relationship to what these queries would actually cost. Do not train a cost model on them.
Workload characterization. The distribution reflects the generator's template mix, not any real application's query pattern.

Limitations

Synthetic throughout - see above.
Some queries are syntactically invalid.
70.2% distinct queries, with heavy near-duplication beyond exact repeats.
Timing and row-count fields are fabricated.
Schema names are industry labels (TELECOMMUNICATIONS, EDUCATION, LOGISTICS) applied cosmetically; they do not reflect real domain schemas.

License

CC0-1.0. Entirely generated content, dedicated to the public domain.

Citation

@dataset{database_query_logs_synthetic,
  title  = {Database Query Logs (synthetic)},
  author = {Robworks Software},
  year   = {2025},
  publisher = {Hugging Face},
  note   = {Programmatically generated queries; timing metadata is fabricated},
  url    = {https://huggingface.co/datasets/robworks-software/database-query-logs-synthetic}
}