---
title: "JMeter CSV Data Set Config & Parameterization"
description: "Master data-driven testing in JMeter: CSV Data Set Config sharing modes (All threads vs Current thread group), EOF rules, and distributed CSV testing."
url: https://docs.jmeter.ai/topics/csv-data-set-config-guide/
lastUpdated: 2026-10-18
source: docs.jmeter.ai
---

# JMeter CSV Data Set Config & Parameterization

Data-driven testing ensures that virtual users execute requests using realistic, varied inputs (usernames, passwords, product IDs, search queries, payment tokens) rather than repeating identical static payloads.

The **CSV Data Set Config** element is Apache JMeter’s primary component for feeding external CSV data into test plans. This guide explains configuration parameters, **Sharing Modes**, End-of-File (EOF) behaviors, and best practices for distributed load testing.

---

## 1. Core Configuration Fields Reference

| Parameter | Recommended Setting | Description |
| --- | --- | --- |
| **Filename** | `users.csv` or `./data/users.csv` | Path to CSV file. Relative paths resolve from JMeter launch directory. |
| **File encoding** | `UTF-8` | Encoding format. Avoid leaving blank if files contain international characters. |
| **Variable Names** | `username,password,userRole` | Comma-separated variable names. If left blank, first row of CSV is read as headers. |
| **Ignore first line** | `True` (if CSV has header row) | Prevents the header text row from being processed as actual test data. |
| **Delimiter** | `,` (or `\t` for tab-separated) | Character separating values in each row. |
| **Allow quoted data?** | `True` | Allows values containing commas wrapped in quotes (`"New York, NY"`). |
| **Recycle on EOF?** | `True` or `False` | When the end of the file is reached, determines whether to loop back to top. |
| **Stop thread on EOF?** | `True` or `False` | Stops the virtual user thread when the file runs out of records. |
| **Sharing mode** | `All threads` / `Current thread group` | Controls how file cursors are shared across threads. |

---

## 2. Understanding Sharing Modes

Sharing Mode defines which virtual users share the CSV file read cursor:

### A. All Threads (Default & Most Common)

- **Behavior**: All virtual users across all Thread Groups read from a single shared file cursor.
- **Result**: Every row in the CSV file is read sequentially across the entire test. No two threads receive the same row on a given step.
- **Best for**: Unique test credentials, single-use coupon codes, unique transaction IDs.

### B. Current Thread Group

- **Behavior**: Each Thread Group maintains its own independent file read cursor.
- **Result**: Thread Group A and Thread Group B both start reading from row 1 of the file independently.
- **Best for**: Multi-stage workflows where each group needs access to the same baseline pool of customer IDs.

### C. Current Thread

- **Behavior**: Every virtual user thread opens its own separate file cursor and reads from row 1.
- **Result**: If you have 50 threads, each thread reads line 1 on iteration 1, line 2 on iteration 2.
- **Best for**: Per-thread dedicated datasets or isolated user state files.

### D. Identifier (Named Sharing Group)

- Assign an arbitrary string (e.g., `group_checkout`). All CSV Data Set configs sharing that identifier string share the same cursor.

---

## 3. End-of-File (EOF) Behaviors

| Recycle on EOF? | Stop thread on EOF? | Behavior when rows run out |
| --- | --- | --- |
| **True** | **False** | **Infinite Loop**: Automatically wraps back to line 1 and continues indefinitely. |
| **False** | **True** | **Graceful Shutdown**: The thread stops immediately upon exhausting data. |
| **False** | **False** | **Default Value (`<EOF>`)**: Variables are assigned literal string `<EOF>` on subsequent reads. |

---

## 4. Multi-Slave & Distributed Testing Considerations

When running JMeter in distributed master-slave mode (`jmeter -r`):

> **CSV Files Must Exist on Remote Injectors**
> JMeter does **not** automatically upload CSV data files to remote slave servers. You must copy the CSV file to the exact same relative directory path on every remote server before starting the test.

### Strategy 1: Partitioning Data Files

To ensure remote slaves do not use duplicate credentials:

- Split `users.csv` into `users_slave1.csv` and `users_slave2.csv`.
- In CSV Data Set Config, reference dynamic filename:

```text
users_\${__P(nodeId, 1)}.csv
```

### Strategy 2: Modular Offset Scripting

Use the `__CSVRead` function with thread and host offsets for mathematical partitioning without physical file splits.

---

## 5. Parameterization Best Practices

1. **Keep Paths Relative**: Always use `./data/users.csv` instead of absolute Windows paths (`C:\Users\...`) to ensure portability across developer machines, Docker containers, and CI/CD agents.
2. **Always Set Variable Names Explicitly**: Defining `username,password` in the element rather than relying on header rows avoids subtle bugs if CSV header formatting changes.
3. **Use Memory Buffers for Huge CSVs**: For datasets with millions of rows, ensure file storage has fast SSD read latency.

## Continue Learning

→

### Next Practical Step

Audit your CSV Data Set Config elements: convert absolute paths to relative paths and verify that EOF settings match your test duration goals.

📖

### Related Reference

- [Functions & Variables Guide](/topics/functions-and-variables/)
- [CSV Data Set Errors Playbook](/topics/errors/csv-data-set-file-not-found-sharing/)
- [CI/CD Load Testing](/topics/ci-cd-load-testing/)
- [Best Practices Manual](/user-manual/best-practices/)

⚠

### Common Mistakes

Using absolute file paths that break in CI/CD containers; leaving “Recycle on EOF” false with “Stop thread” false causing variables to be set to literal `<EOF>`.

🔧

### Troubleshooting

If threads immediately terminate upon launch, check whether your CSV file is empty or missing from the working directory.