Elasticsearch 6 · 7 · 8 · 9 · one SQL interface
Stop ETL’ing Elasticsearch into your warehouse just to JOIN it.
Single-cluster or cross-cluster JOINs on live Elasticsearch, through JDBC, Arrow Flight SQL or ADBC (Java). Pick your scale.
- 2 JOINs per query
- 10,000 rows per query
- 1 cluster · 1 materialized view
- 5 JOINs per query
- 1,000,000 rows per query
- 5 clusters · 50 materialized views
30-day trial, no card.
- Apache 2.0 core
- Elasticsearch 6 to 9
- JDBC Type 4
- Arrow Flight SQL
SELECT c.id, c.name,
SUM(o.amount) AS revenue
FROM customers c
JOIN orders o
ON o.customer_id = c.id
GROUP BY c.id, c.name
ORDER BY revenue DESC
LIMIT 3; | id | name | revenue |
|---|---|---|
| C-1042 | Northwind Ltd | 12,480.00 |
| C-0977 | Alder & Co | 10,915.50 |
| C-1210 | Brightline | 8,302.25 |
Works with
- Tested
- DBeaverDataGripSupersetArrow Flight SQLGrafanaArrow Flight SQL
- Compatible · not yet tested
- TableauPower BIArrow Flight SQL
For data engineers
No export job between Elasticsearch and your JOIN
SQL runs on the live indices (JOINs, subqueries, materialized views) through JDBC or Arrow Flight SQL.
Before
An export pipeline for every question
- Scroll the indices outPython scroll script
- Ship them to a warehouseLogstash or Spark job
- Keep the copy in syncschedules, retries, schema drift
- Run the JOIN therewarehouse SQL
One ticket per new question.
After
One SQL statement on live indices
SELECT c.tier, SUM(o.amount)
FROM orders o
JOIN customers c
ON o.customer_id = c.id
GROUP BY c.tier; Runs on Elasticsearch 6 to 9, where the data already is. Free in Community up to 2 JOINs per query.
orderscustomerslogs-*Pipelines in SQL
Build pipelines, not JSON
Ingest, transform and materialize data across Elasticsearch indices with standard SQL. No custom scripts, no REST API gymnastics.
- Cross-index JOIN: one query across indices, where Elasticsearch SQL can’t JOIN
- Materialized views: persist a pre-joined, pre-aggregated index
- COPY INTO: bulk load from S3 or local files (JSON, Parquet, Delta Lake)
- DML: INSERT, UPDATE, DELETE with SQL expressions
- Arrow Flight SQL: columnar streaming into DuckDB, pandas and Polars
-- Ingest from S3
COPY INTO raw_events
FROM 's3://lake/events.parquet'
FILE_FORMAT = 'PARQUET';
-- Materialize a cross-index JOIN
CREATE MATERIALIZED VIEW
order_details AS
SELECT o.id, c.name, SUM(o.amount)
FROM orders o
JOIN customers c
ON o.customer_id = c.id
GROUP BY o.id, c.name; Also for your team
Analysts and platform engineers use the same SQL
Connect your BI tools
Query Elasticsearch with the SQL you already know: window functions, nested fields, materialized views. Superset, Grafana and DBeaver are Tested; Tableau is Compatible, not yet tested.
Manage schemas with SQL
CREATE, ALTER and DROP indices with familiar DDL, from the REPL or any SQL client. One SQL surface from Elasticsearch 6 to 9, no version-specific JSON mappings.
From Elasticsearch to DuckDB, parsed once
Every stack pays the same Elasticsearch serialization step on the way out of the cluster: every format its search and SQL endpoints offer is a row-serialized document encoding. SoftClient4ES converts to Apache Arrow once, at the sidecar — and those exact columnar buffers are what land in DuckDB, pandas, or Polars. Zero re-serialization after the sidecar.
27.1 KBvs 1.39 GB, Trino connector
It moves what it must, not what it can
A GROUP BY returning 100 rows compiles into an Elasticsearch aggregation:
0.043 s, and 27.1 KB leaves the cluster. Trino’s
Elasticsearch connector performs predicate push-down only, per its documentation, so it
scans all ten million rows out of the cluster first — 5.50 s, and
1.39 GB of egress.
2.83 svs 24.05 s, Trino's documented client
It costs the client far less
Landing all ten million rows as a columnar table costs 2.83 s of client CPU against 24.05 s for Trino’s documented client, and fits in a 2 GB container where that client is OOM-killed even with 8 GB. In an 8 GB budget it completes five concurrent extractions where the documented client completes none and Trino’s fastest completes two.
11.91 svs 14.66 s, Trino's fastest client
And it is faster
11.91 s against 14.66 s for Trino’s fastest client and 44.70 s for its documented one — a factor of 1.23 to 3.75 depending on the route. The low end is the serious comparison: nobody extracts ten million rows through the documented client on purpose.
Where Trino is stronger: its fastest client uses less client memory than we do — 617 MB against our 921 MB. Elasticsearch’s own ES|QL pushes the same aggregation down and ties us on that cell.
Measured 2026-08-21 · Elasticsearch 8.18.3 · Trino 483 · sidecar 0.3.0 — scenes show one live run each; the quoted claim is the session median.
Correlate logs, metrics, and traces across regional ES clusters in a single SQL query.
One incident, several regions, several clusters. Instead of pivoting across N dashboards, JOIN the per-region indices in one query and read the timeline straight across.
Cross-cluster correlation runs on the federation server Pro, up to 5 clusters. Heterogeneous sources are upcoming in Q1 2027. Federation operator guide
-- Illustrative: use your own
-- per-region index names.
SELECT u.user_id, u.action,
e.error_code, e.region
FROM `prod_us`.user_events u
JOIN `prod_eu`.error_events e
ON u.user_id = e.user_id
WHERE e.ts >
NOW() - INTERVAL 15 MINUTE
ORDER BY e.ts; Free until you need scale
Every tier has every feature. The limits set the price, never seats or environments.
Compare all editions →Community
Free- JOINs per query
- 2
- Elasticsearch clusters
- 1
- Materialized views
- 1
- Rows per query
- 10,000
Pro
€119 /month- JOINs per query
- 5
- Elasticsearch clusters
- 5
- Materialized views
- 50
- Rows per query
- 1,000,000
Try it
Run your first JOIN in five minutes
$ curl -fsSL https://raw.githubusercontent.com/SOFTNETWORK-APP/SoftClient4ES/main/install.sh | bash $ ~/softclient4es/bin/softclient4es --host localhost --port 9200
Needs Java 11 or later (17 for Elasticsearch 9). On Windows the installer installs it if needed; on macOS and Linux it prints the install command.