yuqi1129 opened a new pull request, #14635:
URL: https://github.com/apache/gravitino/pull/14635

   ### What changes were proposed in this pull request?
   
   Backport #14508 (`084754f2d3`) and #14607 (`c234862c72`) together to 
`branch-1.3`:
   
   - Import external tables under their own locks, preserve schema identities, 
and report partial source/store registration failures with recovery guidance.
   - Expand top-level column storage to 300 characters and add JDBC extension 
hooks and opt-in bounded physical connects.
   - Normalize table alter changes once in `TableNormalizeDispatcher`.
   
   Adapt installation/upgrade SQL to 1.3 and use catalog-row locking with the 
existing 1.3 storage implementation. Main-only Secret APIs and the quote-aware 
normalization SPI are not introduced; the normalization regression verifies the 
existing 1.3 folding path executes once.
   
   ### Why are the changes needed?
   
   Bring the metadata integrity, import concurrency, JDBC support, and 
duplicate-normalization fixes to the 1.3 maintenance branch without merging 
unrelated main history.
   
   ### Does this PR introduce _any_ user-facing change?
   
   Conflicting schema identities are rejected without deleting their children 
or governance metadata. Source creation can succeed before registration fails; 
reconcile conflicts or use load/import recovery as described in the error. 
External schema creation adds a registration preflight; warm external table 
loads perform two registration lookups, managed loads add a preflight lookup, 
and stale-registration repair may still read source metadata twice. Schema 
registrations within one catalog serialize on its database row. Existing CLI 
create commands retain their generic already-exists messages; detailed 
reconciliation guidance remains available in REST/client errors.
   
   Only top-level column names expand; catalog-specific limits still apply. 
Existing 1.3 databases require the documented stopped-server backup and manual 
column-capacity ALTER. Imported source properties, including inline sensitive 
values, are stored without new encryption; existing client filtering remains in 
place and schema alters retain the current source properties. JDBC 
hooks/connect deadlines are opt-in, with driver isolation and cleanup limits 
documented in `docs/jdbc-catalog-connection-validation.md`.
   
   ### How was this patch tested?
   
   - 309 targeted Java cases and 33 Python client cases passed, including 93 
storage/script cases across H2, MySQL 8 and PostgreSQL 13, 15 bounded-driver 
cases, catalog dispatchers, Java client, REST, and Iceberg REST regressions. 
The 33-case MySQL catalog integration suite also passed against an embedded 
server and MySQL 8.
   - Reverting the original 1.3 production paths and upgrade scripts reproduces 
all nine selected regression failures; restoring the fixes passes them. Native 
import cases for 128/129/300-character names executed and passed; restoring the 
old 128-character cap fails the 129/300 cases while keeping the 128 case green. 
Scoped Spotless and `git diff --check` passed.
   - No full core suite or deployed multi-node test was run.
   


-- 
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]

Reply via email to