| --- |
| license: cc-by-nc-4.0 |
| task_categories: |
| - text-classification |
| - tabular-classification |
| - time-series-forecasting |
| language: |
| - en |
| tags: |
| - synthetic-data |
| - logs |
| - observability |
| - devops |
| - sre |
| size_categories: |
| - 100M<n<1B |
| --- |
| |
| # FreeSyntheticServerLogs100M |
|
|
| A free dataset of 100 million fully synthetic server and application log entries, built for developers and researchers who need realistic structured log data at scale — for testing log-analytics tools, alerting and observability pipelines, log-parsing systems, or training models to classify and triage logs. No real systems, users, or production data are represented in this data. |
|
|
| ## Schema |
|
|
| | Column | Type | Description | |
| |---|---|---| |
| | log_id | string | Unique log entry identifier | |
| | log_timestamp | string | Event time (YYYY-MM-DD HH:MM:SS) | |
| | service_name | string | Emitting service (auth-service, api-gateway, etc.) | |
| | log_level | string | INFO, DEBUG, WARN, ERROR, or FATAL | |
| | status_code | int | HTTP status code, correlated with log level | |
| | response_time_ms | int | Request duration in milliseconds | |
| | http_method | string | GET, POST, PUT, DELETE, PATCH | |
| | endpoint | string | API endpoint path | |
| | host | string | Host identifier (region + node) | |
| | message | string | Log message text | |
|
|
| ## Format |
|
|
| Single Parquet file, Snappy compression, ~2.0 GB, 100,000,000 rows. |
|
|
| ## Quick Start |
|
|
| **pandas** |
| ```python |
| import pandas as pd |
| df = pd.read_parquet("logs_100M.parquet") |
| ``` |
|
|
| **datasets** |
| ```python |
| from datasets import load_dataset |
| ds = load_dataset("ziadatalabs/FreeSyntheticServerLogs100M") |
| ``` |
|
|
| **duckdb** |
| ```python |
| import duckdb |
| duckdb.sql("SELECT * FROM 'logs_100M.parquet' LIMIT 10").show() |
| ``` |
|
|
| ## Notes |
|
|
| Status codes and response times are correlated with log level — INFO and DEBUG carry 2xx/3xx codes and fast responses, WARN skews toward 4xx, and ERROR and FATAL carry 5xx codes with slower response times — so the data reflects the real relationships a log-analytics or anomaly-detection model would learn. Log-level, service, method, and endpoint distributions reflect typical production traffic. All entirely synthetic. Note: at 100M rows, load in batches (e.g. pyarrow iter_batches or duckdb) rather than all at once on memory-limited machines. |
| |
| ## License & Usage |
| |
| Released under CC BY-NC 4.0 — personal, research, and educational use permitted, attribution required, no commercial use. |
| |
| --- |
| |
| Created by Zia Data Labs. Questions or feedback: zia.data.team@protonmail.com |
| |
