Use a feeder to inject dynamic data to your Gatling simulations

Feeder is a type alias for Iterator[Map[String, T]], meaning that the component created by the feed method will poll Map[String, T] records and inject its content.

It’s very simple to build a custom one. For example, here’s how one could build a random email generator:

import scala.util.Random
val feeder = Iterator.continually(Map("email" -> (Random.alphanumeric.take(20).mkString + "@foo.com")))

The structure DSL provides a feed method.


This defines a workflow step where every virtual user feed on the same Feeder.

Every time a virtual user reaches this step, it will pop a record out of the Feeder, which will be injected into the user’s Session, resulting in a new Session instance.

If the Feeder can’t produce enough records, Gatling will complain about it and your simulation will stop.

feed(feeder, 2)


Gatling provides multiple strategies for the built-in feeders:

.queue // default behavior: use an Iterator on the underlying sequence
.random // randomly pick an entry in the sequence
.shuffle // shuffle entries, then behave like queue
.circular // go back to the top of the sequence once the end is reached


An Array[Map[String, T]] or a IndexedSeq[Map[String, T]] can be implicitly turned into a Feeder. For example:

val feeder = Array(
  Map("foo" -> "foo1", "bar" -> "bar1"),
  Map("foo" -> "foo2", "bar" -> "bar2"),
  Map("foo" -> "foo3", "bar" -> "bar3")

File Based Feeders

Gatling provides various file based feeders.

When using the bundle distribution, files must be in the user-files/resources directory. This location can be overridden, see [configuration`.

When using a build tool such as maven, files must be placed in src/main/resources or src/test/resources.

In order the locate the file, Gatling try the following strategies in sequence:

  1. as a classpath resource from the classpath root, eg data/file.csv for targeting the src/main/resources/data/file.csv file. This strategy is the recommended one.
  2. from the filesystem, as a path relative to the Gatling root dir. This strategy should only be used when using the Gatling bundle.
  3. from the filesystem, as an absolute path. Use this strategy if you want your feeder files to be deployed separately.

CSV feeders

Gatling provides several built-ins for reading character-separated values files.

Our parser honors the RFC4180 specification.

The only difference is that header fields get trimmed of wrapping whitespaces.

val csvFeeder = csv("foo.csv") // use a comma separator
val tsvFeeder = tsv("foo.tsv") // use a tabulation separator
val ssvFeeder = ssv("foo.ssv") // use a semicolon separator
val customSeparatorFeeder = separatedValues("foo.txt", '#') // use your own separator

Loading mode

CSV files feeders provide several options for how data should be loaded in memory.

eager loads the whole data in memory before the Simulation starts, saving disk access at runtime. This mode works best with reasonably small files that can be parsed quickly without delaying simulation start time and easily sit in memory. This behavior was the default prior to Gatling 3.1 and you can still force it.

val csvFeeder = csv("foo.csv").eager.random

batch works better with large files whose parsing would delay simulation start time and eat a lot of heap space. Data is then read by chunks.

val csvFeeder = csv("foo.csv").batch.random
val csvFeeder2 = csv("foo.csv").batch(200).random // tune internal buffer size

Default behavior is an adaptive policy based on (unzipped, sharded) file size, see gatling.core.feederAdaptiveLoadModeThreshold in config file. Gatling will use eager below threshold and batch above.

Zipped files

If your files are very large, you can provide them zipped and ask gatling to unzip them on the fly:

val csvFeeder = csv("foo.csv.zip").unzip

Supported formats are gzip and zip (but archive most contain only one single file).

Distributed files (Gatling Enterprise only)

If you want to run distributed with Gatling Enterprise and you want to distribute data so that users don’t use the same data when they run on different cluster nodes, you can use the shard option. For example, if you have a file with 30,000 records deployed on 3 nodes, each will use a 10,000 records slice.

val csvFeeder = csv("foo.csv").shard

JSON feeders

Some might want to use data in JSON format instead of CSV:

val jsonFileFeeder = jsonFile("foo.json")
val jsonUrlFeeder = jsonUrl("http://me.com/foo.json")

For example, the following JSON:


will be turned into:

record1: Map("id" -> 19434, "foo" -> 1)
record2: Map("id" -> 19435, "foo" -> 2)

Note that the root element has of course to be an array.

JDBC feeder

Gatling also provide a builtin that reads from a JDBC connection.

// beware: you need to import the jdbc module
import io.gatling.jdbc.Predef._

jdbcFeeder("databaseUrl", "username", "password", "SELECT * FROM users")

Just like File parser built-ins, this return a RecordSeqFeederBuilder instance.

  • The databaseUrl must be a JDBC URL (e.g. jdbc:postgresql:gatling),
  • the username and password are the credentials to access the database,
  • sql is the query that will get the values needed.

Only JDBC4 drivers are supported, so that they automatically registers to the DriverManager.

Sitemap Feeder

Gatling supports a feeder that reads data from a Sitemap file.

// beware: you need to import the http module
import io.gatling.http.Predef._

val feeder = sitemap("/path/to/sitemap/file")

The following Sitemap file:

<?xml version="1.0" encoding="UTF-8"?>
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">



will be turned into:

record1: Map(
  "loc" -> "http://www.example.com/",
  "lastmod" -> "2005-01-01",
  "changefreq" -> "monthly",
  "priority" -> "0.8")
record2: Map(
  "loc" -> "http://www.example.com/catalog?item=12&amp;desc=vacation_hawaii",
  "changefreq" -> "weekly")

record3: Map(
  "loc" -> "http://www.example.com/catalog?item=73&amp;desc=vacation_new_zealand",
  "lastmod" -> "2004-12-23",
  "changefreq" -> "weekly")

Redis feeder

This feature was originally contributed by Krishnen Chedambarum.

Gatling can read data from Redis using one of the following Redis commands.

  • LPOP - remove and return the first element of the list
  • SPOP - remove and return a random element from the set
  • SRANDMEMBER - return a random element from the set

By default RedisFeeder uses LPOP command:

import io.gatling.redis.Predef._

import com.redis._

val redisPool = new RedisClientPool("localhost", 6379)

// use a list, so there's one single value per record, which is here named "foo"
// same as redisFeeder(redisPool, "foo").LPOP
val feeder = redisFeeder(redisPool, "foo")

You can then override the desired Redis command:

// read data using SPOP command from a set named "foo"
val feeder = redisFeeder(redisPool, "foo").SPOP
// read data using SRANDMEMBER command from a set named "foo"
val feeder = redisFeeder(redisPool, "foo").SRANDMEMBER

Note that since v2.1.14, Redis supports mass insertion of data from a file. It is possible to load millions of keys in a few seconds in Redis and Gatling will read them off memory directly.

For example: a simple Scala function to generate a file with 1 million different urls ready to be loaded in a Redis list named URLS:

import java.io.{ File, PrintWriter }

import io.gatling.redis.util.RedisHelper._

def generateOneMillionUrls(): Unit = {
  val writer = new PrintWriter(new File("/tmp/loadtest.txt"))
  try {
    for (i <- 0 to 1000000) {
      val url = "test?id=" + i
      // note the list name "URLS" here
      writer.write(generateRedisProtocol("LPUSH", "URLS", url))
  } finally {

The urls can then be loaded in Redis using the following command:

cat /tmp/loadtest.txt | redis-cli --pipe


Sometimes, you might want to convert the raw data you got from your feeder.

For example, a csv feeder would give you only Strings, but you might want to convert one of the attribute into an Int.

convert(conversion: PartialFunction[(String, T), Any]) takes:

  • a PartialFunction, meaning that you only define it for the scope you want to convert, non matching attributes will be left unchanged
  • whose input is a (String, T) couple where the first element is the attribute name, and the second one the attribute value
  • and whose output is Any, whatever you want

For example:

csv("myFile.csv").convert {
  case ("attributeThatShouldBeAnInt", string) => string.toInt

Grabbing Records

Sometimes, you just might want to reuse or convenient built-in feeders for custom needs and get your hands on the actual records.

readRecords returns a Seq[Map[String, Any]].

val records: Seq[Map[String, Any]] = csv("myFile.csv").readRecords

Non Shared Data

Sometimes, you could want all virtual users to play all the records in a file, and Feeder doesn’t match this behavior.

Still, it’s quite easy to build, thanks to flattenMapIntoAttributes e.g.:

val records = csv("foo.csv").readRecords

foreach(records, "record") {

User Dependent Data

Sometimes, you could want to filter the injected data depending on some information from the Session.

Feeder can’t achieve this as it’s just an Iterator, so it’s unaware of the context.

You’ll then have to write your own injection logic, but you can of course reuse Gatling parsers.

Consider the following example, where you have 2 files and want to inject data from the second one, depending on what has been injected from the first one.

In userProject.csv:

user, project
bob, aProject
sue, bProject

In projectIssue.csv:


Here’s how you can randomly inject an issue, depending on the project:

import java.util.concurrent.ThreadLocalRandom

import io.gatling.core.feeder._

// index records by project
val issuesByProject: Map[String, Seq[Any]] =
    .groupMap(record => record("project").toString)(record => record("issue"))

// inject project
  .exec { session =>
    // fetch project from  session
    session("project").validate[String].map { project =>
      // fetch project's issues
      val issues = issuesByProject(project)

      // randomly select an issue
      val selectedIssue = issues(ThreadLocalRandom.current.nextInt(issues.length))

      // inject the issue in the session
      session.set("issue", selectedIssue)

Edit this page on GitHub