CryoThrust opened a new pull request, #13264:
URL: https://github.com/apache/seatunnel/pull/13264

   ### Purpose of this pull request
   
   This contribution to Apache SeaTunnel fixes #12766. Loading Hadoop XML 
currently does not log in from a keytab, so the Hudi catalog and writers can 
access a Kerberos-protected HDFS without an authenticated identity.
   
   Add the Hadoop-family options `kerberos_principal`, `kerberos_keytab_path` 
and `krb5_path` to both sink and catalog factories. Both credential options 
must be nonblank; the supplied Hadoop configuration must enable Kerberos. With 
neither credential option configured, the existing execution path is retained.
   
   Each catalog and direct/cached write-client provider owns a runtime UGI 
returned by `loginUserFromKeytabAndReturnUGI`. Initialization, catalog I/O, 
writes/deletes, automatic commits and cleanup use that identity. Check ticket 
renewal before operations; still attempt cleanup if renewal fails. Close the 
identity-owned filesystem cache and log out after client shutdown. Neither UGI 
nor its executor is serialized with the provider.
   
   Hudi 0.15's Java engine and `HoodieListData` use parallel streams, so 
wrapping only the submitting thread in `doAs` is insufficient when the common 
pool already exists. Authenticated operations run in a private ForkJoinPool 
whose worker lifetime is scoped to the returned UGI. Hudi's queue executor 
creates ordinary child threads within that scope. The existing configuration 
disables asynchronous cleaning and the embedded timeline server.
   
   The unused global-login helper remains uncalled. Hadoop security-mode 
initialization and the optional krb5 system property follow the synchronized 
HadoopLoginFactory precedent; those settings are still JVM-wide. The EN/ZH 
documentation describes this shared-worker constraint and a real-KDC 
verification recipe.
   
   ### Does this PR introduce any user-facing change?
   
   Yes. Hudi can explicitly authenticate with the three canonical options 
above. Partial/blank credentials and credentials combined with a non-Kerberos 
Hadoop configuration fail before filesystem/client initialization. The dotted 
JDBC options are not introduced as Hudi options. The authentication helper 
omits raw credential-bearing causes, and HudiSinkConfig.toString excludes the 
credential fields.
   
   No dependencies or default write/commit settings change. Open PR #12440 
changes the Hudi commit model and is not included here; whichever merges second 
will need to cover its aggregated-committer path too.
   
   ### How was this patch tested?
   
   - All 34 connector unit tests pass on both JDK 11 and 17. They cover option 
declaration/config text, conditional authentication, login-error sanitization, 
identity isolation, renewal failure and cleanup, an already initialized common 
pool, actual Hudi Java-engine/ListData parallel work, and Hudi child threads.
   - The direct and cached providers are exercised through HudiRecordWriter for 
client creation, upsert, delete/startCommit and close with identity assertions. 
Catalog tests exercise local filesystem initialization, database/table 
creation, listing, lookup and deletion with a returned UGI.
   - A real local Hudi client writes, automatically commits and deletes under a 
returned test UGI. The KDC login itself is stubbed; this is not a Kerberos/HDFS 
deployment test.
   - The existing standalone local-write fixture disables its embedded timeline 
server, avoiding dependence on a routable advertised host address.
   - Spotless and the connector's eight-module dependency reactor pass 
skip-tests verification, including Checkstyle/license checks.
   - The new negative case in the existing HudiIT passes on a real Zeta 
container: credentials with a non-Kerberos Hadoop configuration reject the job 
with the expected configuration diagnostic (1 test, no failures/errors/skips). 
The selected Zeta case uses direct test goals after building prerequisites, 
avoiding unrelated Spark 3.5 SDK staging.
   
   Portable connector commands (JDK 11):
   
   ```bash
   ./mvnw -pl seatunnel-connectors-v2/connector-hudi spotless:apply
   ./mvnw -pl seatunnel-connectors-v2/connector-hudi test
   ./mvnw -pl seatunnel-connectors-v2/connector-hudi -am -DskipTests verify
   RUN_ENGINES=zeta ./mvnw -pl 
seatunnel-e2e/seatunnel-connector-v2-e2e/connector-hudi-e2e \
     resources:testResources compiler:testCompile failsafe:integration-test 
failsafe:verify \
     -DskipIT=false 
-Dit.test=HudiIT#testRejectKerberosCredentialsWithoutSecurityConfiguration
   ```
   
   No real KDC login, Kerberized HDFS, live ticket-expiry/renewal or full 
multi-engine suite was run. These remain deployment validation, not inferred 
from the credential-free tests.
   
   ### Check list
   
   - [x] EN/ZH connector documentation updated.
   - [x] Unit tests and an existing E2E suite updated.
   - [x] No new dependency or binary package.
   - [x] No new connector registration or distribution entry required.
   
   


-- 
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