mirror of https://github.com/apache/druid.git
6c5a43106a
* SQL: Morph QueryMakerFactory into SqlEngine. Groundwork for introducing an indexing-service-task-based SQL engine under the umbrella of #12262. Also includes some other changes related to improving error behavior. Main changes: 1) Elevate the QueryMakerFactory interface (an extension point that allows customization of how queries are made) into SqlEngine. SQL engines can influence planner behavior through EngineFeatures, and can fully control the mechanics of query execution using QueryMakers. 2) Remove the server-wide QueryMakerFactory choice, in favor of the choice being made by the SQL entrypoint. The indexing-service-task-based SQL engine would be associated with its own entrypoint, like /druid/v2/sql/task. Other changes: 1) Adjust DruidPlanner to try either DRUID or BINDABLE convention based on analysis of the planned rels; never try both. In particular, we no longer try BINDABLE when DRUID fails. This simplifies the logic and improves error messages. 2) Adjust error message "Cannot build plan for query" to omit the SQL query text. Useful because the text can be quite long, which makes it easy to miss the text about the problem. 3) Add a feature to block context parameters used internally by the SQL planner from being supplied by end users. 4) Add a feature to enable adding row signature to the context for Scan queries. This is useful in building the task-based engine. 5) Add saffron.properties file that turns off sets and graphviz dumps in "cannot plan" errors. Significantly reduces log spam on the Broker. * Fixes from CI. * Changes from review. * Can vectorize, now that join-to-filter is on by default. * Checkstyle! And variable renames! * Remove throws from test. |
||
|---|---|---|
| .. | ||
| src/test | ||
| assembly.xml | ||
| pom.xml | ||