The Unified Query API

GitLab data lives in many places - relational tables, a graph of connections, a full-text index, a code search engine, an analytics warehouse. Each has had its own query surface, its own performance profile, and its own UI. GLQL brings them together: one query language, one set of shared components, one place to ask a question and see the answer.

This challenge is a PoC of GLQL as the unified query layer, extended with new data providers so a single GLQL query can reach graph connections, full-text search, code search and scalable analytics - all from the same familiar surface.

A GLQL query, unchanged, now reaches further:

# One language across providers
type = Issue AND state = opened
  AND label = "group::search"
  AND search = "imagination"   -- Elasticsearch provider
ORDER BY updated DESC
LIMIT 20

✓ resolved across providers (db + elasticsearch) in 142ms

• One language to learn - the same GLQL grammar resolves against every provider.
• Providers are pluggable - adding a data source expands what GLQL can answer without a new UI.
• Performant and scalable by design - each query runs against the engine best suited to it.

Orbit - the Knowledge Graph - becomes a GLQL data provider. Instead of a flat list, a query can return the graph of connections between items and render it as a network you can explore and share. Relationships that were invisible in a table become obvious at a glance.

Search stops being a separate destination. With Elasticsearch and Zoekt as GLQL providers, you can express a full-text or code search as part of a GLQL query and filter, sort and display the results with the same components you already use everywhere else.

Elasticsearch: full-text search

Zoekt: code search

The Data Insights Platform, backed by ClickHouse, brings end-to-end, scalable analytics into the unified query layer. During the summit we made it far more flexible: instead of defining backend models up front, you can now run your own queries through the Query API - and they stay fast as the data grows.

Because GLQL is one query layer with one set of components, making it fast makes everything fast. The same engine that powers standalone queries also powers everyday UI - dropdowns, filters, pickers - so they get quicker and more consistent for free.

Work item parent dropdowns ⚡

Project filter dropdowns ⚡

Built by Platform Insights - the Foundation, Search and GLQL groups - as an R&D Summit challenge.

Jiaan Louw @jiaan EM John Mason @johnmason DRI Daniele Rossetti @drosse Robert Hunt @rob.hunt Ravi Kumar @rkumar555 Siddharth Dungarwal @sdungarwal Sreeram Narayanan @snarayanan_gl Vineeth Ravi @vineeth-r Ezekiel Kigbo @ekigbo Dmitry Gruzd @dgruzd

Thanks to everyone who joined the Unified Query API challenge at the R&D Summit.