Does Karate evaluate the entire JSON array passed to 'call' before iteration when paginating large datasets?

Answer to the question

Karate evaluates the entire JSON array that is passed to a call loop before any iteration starts. The array is materialised by the embedded JavaScript engine (GraalVM or Nashorn, depending on the version) and occupies heap space for the duration of the call.

Why this matters

Because the array is fully loaded, memory usage grows linearly with the number of elements. This mirrors the behaviour of large Examples tables and can lead to OutOfMemoryError if the array is very large.

Recommended pattern for large data sets

To avoid loading all pages at once, use a recursive feature call that fetches one page at a time. Each invocation passes the next offset and stops when the response indicates no more data.

# pagination.feature
Scenario: Process Pages
  * def offset = 0
  * def limit = 100
  * call read('pagination.feature') { offset: offset, limit: limit }

Scenario: Page Worker
  Given url 'https://api.example.com/data'
  And param offset = offset
  And param limit = limit
  When method get
  Then status 200
  * def hasMore = response.length == limit
  * if (hasMore) call read('pagination.feature') { offset: offset + limit, limit: limit }

How to verify

Run the feature while monitoring the JVM heap (e.g., with JVisualVM). Observe the spike when a large array is used vs. the steady profile of the recursive approach.