markhoerth opened a new pull request, #12328:
URL: https://github.com/apache/gravitino/pull/12328
### What changes were proposed in this pull request?
Completes the concept and API split started in #12326 and #12327, covering
the six
`manage-*` pages those did not reach.
New pages: `metalakes.md` and `partitions.md`. `model-catalog.md` is
rewritten as
the concept page for the model registry.
Reduced to the API: `manage-model-metadata-using-gravitino.md` (1284 lines
to 345),
`manage-jobs-in-gravitino.md` (673 to 217),
`manage-table-partition-using-gravitino.md`
(417 to 184), `manage-statistics-in-gravitino.md` (374 to 193),
`manage-metalake-using-gravitino.md` (321 to 218), and
`manage-view-metadata-using-gravitino.md` (289 to 174). Each keeps its slug
and
links to its concept page.
Several of these pages link to concept pages added in #12327, so this one
should
land after it.
### Why are the changes needed?
These six pages had no concept layer. The metalake page outsourced the
definition of
a metalake to `glossary.md#metalake`, which is the only inbound link to the
glossary.
The model catalog page linked four of its own sections to the API page
rather than
explaining anything, and never covered aliases or multiple URIs per version.
The
partition page presented a support matrix of four identical rows without
saying what
an Iceberg user should do instead.
Facts corrected against source, none of which the pages stated:
- Statistics go through `supportsStatistics()` and
`supportsPartitionStatistics()`
rather than directly on the table, and custom names take a `custom.`
prefix, not
`custom-`
- Cancelling a job is a `POST` to `runs/{jobId}`, with no `/cancel` suffix
- `createView` takes `Representation[]` built with
`SQLRepresentation.builder()`,
plus a default catalog and schema, not a query string and dialect
- Java overloads `getModelVersion` for alias lookup; there is no
`getModelVersionByAlias`
- Partition management is implemented by Hive, Glue, and Doris. Glue was
absent from
the old support matrix, and the other JDBC catalogs fail rather than
returning
empty
- Only service admins can create a metalake, and metalake operations use the
admin
client rather than the metalake client
- An alias belongs to one model version at a time, so moving it removes it
from the
version that held it
The partitions page also now says what an Iceberg user does instead:
partitioning is
set at table creation and evolved through an engine against the Iceberg REST
catalog
service, which handles the resulting `AddPartitionSpec` and
`SetDefaultPartitionSpec`
updates.
### Does this PR introduce _any_ user-facing change?
Documentation only. No API, property, or behavior changes. Two new pages;
every
existing page keeps its slug, so no links break. The new pages are not in
the sidebar
yet, since `sidebars.ts` lives in gravitino-site.
### How was this patch tested?
Every claim was checked against source: the statistics and partition
interfaces in
`org.apache.gravitino.stats` and `org.apache.gravitino.rel`, the job REST
paths in
`JobOperations`, `ViewCatalog.createView` and `SQLRepresentation`, the model
and
model version change types, the catalogs implementing `SupportsPartitions`,
the
Iceberg partition spec converters, and the `SERVICE_ADMIN` expression on
metalake
creation. UI behavior was checked against `web-v2`.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
