This guide records the earlier scalar configuration interface. The current interface prefers complete file content; see Configuration File Editors.
The Kyuubi, Airflow, Celeborn, DolphinScheduler, Trino, Doris, Elasticsearch and MinIO packs expose native configuration through Ambari. Basic settings remain compatible with earlier releases. Service-owned configuration types travel with each Agent task and are consumed by the runtime renderer.
Save a configuration version with a note, review any advisor warnings, and restart the affected service. Configuration saving does not itself restart a service. The installed service‘s native settings, configuration history and restart indicators remain Ambari’s normal interfaces.
For an existing installation, update the management definition first, reopen Configs and save the newly exposed defaults together with the existing values. Do not replace existing cluster configurations wholesale with XML defaults. The basic settings retain their names, so existing credentials, paths, ports and topology do not require renaming. Existing custom properties are preserved.
| Service | Configuration types and behavior |
|---|---|
| Kyuubi | kyuubi-defaults writes every supplied native property; kyuubi-environment writes literal shell exports into kyuubi-env.sh; kyuubi-log4j2 writes the complete logging file. JVM options are editable. |
| Trino | trino-config, trino-node, trino-log, trino-jvm and trino-catalogs manage the corresponding files. Each catalog is a name-to-properties entry in the JSON editor. |
| Doris | doris-fe and doris-be overlay the pinned vendor files. Replaced keys are removed before new values are written, preserving vendor comments and version-specific defaults. |
| Elasticsearch | elasticsearch-settings writes YAML settings using quoted scalars and JSON-compatible arrays; elasticsearch-jvm writes an options fragment; elasticsearch-log4j2 manages the complete vendor logging document. |
| DolphinScheduler | Separate master, worker, API and alert types write Spring application.properties overrides beside vendor YAML. The common type overlays common.properties. Each role has its own full Logback editor; the runtime and JVM types configure process memory and options. |
| MinIO | minio-environment supplies native MINIO_* variables to the running server through a protected runtime document. Basic credentials and listener arguments remain authoritative. |
| Airflow | airflow-settings uses section.option keys to manage airflow.cfg; airflow-webserver manages the full FAB Python configuration. |
| Celeborn | celeborn-defaults writes native properties; celeborn-environment manages additional exports; celeborn-log4j2 manages logging. Master and worker assignments remain in basic settings. |
Common fields are a curated subset for the pinned software versions. Additional upstream options use the custom-property editor; the package does not enumerate every connector, plugin or future upstream option. Arbitrary user-supplied options remain subject to the application's own validation.
Configuration types are scoped to their service; unrelated cluster credentials are not forwarded to runtime helpers. Properties are validated and escaped, including Java property Unicode escapes. Structured action results retain execution identity and do not echo the input configuration.
Basic topology, installation paths and service identity take precedence through explicit conflict rejection. Heap options for Trino and Elasticsearch belong to Basic Settings; their JVM editors accept additional options. DolphinScheduler database and registry settings use the existing basic fields. Invalid JVM input, malformed catalog JSON, duplicate catalog names and unsafe catalog paths fail.
Trino records its owned catalog names before writing files, allowing retry after an interrupted write. It removes catalogs deleted from the editor and refuses to adopt pre-existing unowned files. The TPCH catalog is retained for the service check.
Kyuubi waits for a reachable JDBC listener before attempting the SQL service check. A listening socket is only a prerequisite: success still requires the real SQL result and matching session, operation and execution identities.
These configuration improvements retain the previously accepted single-node or small-cluster topology. HA, new authentication modes and additional runtime dependencies require their corresponding lifecycle support. For example, changing an arbitrary property does not provision an Airflow Celery broker, enable Kyuubi Kerberos or distribute Elasticsearch certificates.
The complete store remains a transport bundle of management definitions. Software artifacts continue to use their pinned upstream sources. Published archive bytes are immutable; configuration changes ship as new definition versions. No RPM rebuild is required for these definition updates.
Acceptance evidence and exact final versions are recorded in native configuration acceptance.