From/size vs search_after vs scroll in Elasticsearch: Key Differences and Usage

In Elasticsearch, from/size is simple pagination but inefficient for deep pages, search_after is a performant cursor-based pagination for sorted results, and scroll is designed for retrieving large datasets efficiently in batches without real-time relevance. Choose based on data size and use case.

Quick Comparison

This table summarizes the main differences between from/size, search_after, and scroll in Elasticsearch.

Feature	from/size	search_after	scroll
Use case	Simple pagination for small offsets	Efficient deep pagination with sorting	Extracting large datasets in batches
Performance	Slower for large offsets	Fast and efficient	Optimized for large data retrieval
Statefulness	Stateless	Stateless	Stateful (keeps context)
Sorting requirement	No strict sorting needed	Requires unique sort keys	No strict sorting needed
Data freshness	Real-time	Real-time	Snapshot at scroll start
Complexity	Simple to use	Moderate complexity	More complex, needs scroll context

Key Differences

from/size is the simplest way to paginate results by skipping from records and returning size records. However, it becomes slow and resource-heavy for large from values because Elasticsearch must skip many records internally.

search_after uses a cursor-like approach by passing the last sort values from the previous page to fetch the next page. It requires a unique and consistent sort order but is much faster and more efficient for deep pagination since it avoids skipping.

scroll is designed for extracting large volumes of data in batches. It creates a snapshot of the data at the start and keeps a context open on the server. This method is stateful and not suitable for real-time user pagination but excellent for exporting or reindexing large datasets.

Code Comparison

Example of paginating with from/size to get the second page of 10 results sorted by timestamp:

json

{
  "query": { "match_all": {} },
  "sort": [ { "timestamp": "asc" } ],
  "from": 10,
  "size": 10
}

Output

Returns documents 11 to 20 sorted by timestamp ascending.

search_after Equivalent

Example of paginating with search_after using the last hit's sort values ["2023-01-01T12:00:00", "abc123"]:

json

{
  "query": { "match_all": {} },
  "sort": [ { "timestamp": "asc" }, { "_id": "asc" } ],
  "size": 10,
  "search_after": ["2023-01-01T12:00:00", "abc123"]
}

Output

Returns next 10 documents after the given timestamp and ID, sorted ascending.

When to Use Which

Choose from/size for simple, shallow pagination with small offsets where performance impact is minimal.

Choose search_after for efficient deep pagination on sorted data when you need fast response times and consistent ordering.

Choose scroll when you need to retrieve large datasets in bulk, such as for data export or reindexing, and real-time freshness is not critical.

Key Takeaways

from/size is easy but slow for deep pages.

search_after is efficient cursor-based pagination requiring unique sorting.

scroll is best for large bulk data retrieval, not real-time pagination.

Use search_after for deep pagination with sorting, from/size for simple cases.

scroll keeps server state and snapshots data at query start.