The GitHub Actions job "flink-connector-aws: build on pull request" on 
flink-connector-aws.git/new-combined-implementation has failed.
Run started by GitHub user fmorillo7694 (triggered by fmorillo7694).

Head commit for run:
613ab852fc1cecc3a03328cb0bf2accad9b79b06 / Francisco <[email protected]>
[FLINK-30481][Connectors/AWS] Add AWS Glue Data Catalog implementation 
(FLIP-277)

Adds flink-catalog-aws-glue, a Flink Catalog backed by the AWS Glue Data
Catalog, plus the flink-sql-catalog-aws-glue uber jar, documentation (en
and zh) and end-to-end tests.

Catalog
- Database, table, view, function and partition operations through the
  AWS SDK v2 Glue client, with pagination for every list call.
- Names are stored lowercase as Glue requires; original column names and
  the full Flink schema (types, nullability, computed and metadata
  columns, watermarks, primary keys, column order) are persisted as Glue
  parameters keyed by column name so the schema round-trips exactly.
- Tables written by other engines (Athena, Spark, Hive crawlers) are
  readable: Glue type strings are parsed case-insensitively, including
  nested and parameterized types, EXTERNAL_TABLE and VIRTUAL_VIEW types,
  missing storage descriptors and partition keys duplicated in the
  storage descriptor.
- Malformed metadata fails fast instead of being skipped: unknown or
  malformed Glue types, struct fields without a type, partitions whose
  value count does not match the table keys.
- getTable only surfaces options the write path consumes, so
  getTable(a) -> createTable(b) copies are symmetric.
- Partition locations follow the HiveCatalog contract (exposed as the
  location property on read, honoured on create and alter) with a guard
  that rejects a location spelling a different partition of the same
  table.
- AWS client configuration (region, endpoint, credentials, http client
  type and timeouts) reuses the shared connector-aws-base utilities;
  http-client.type is validated and the timeout options are wired.
- Unsupported operations (statistics, listPartitionsByFilter) are
  documented in the Limitations section.

Tests
- Unit tests run against an in-memory FakeGlueClient with real Glue
  semantics (lowercasing, location rules, UpdateTable, cascade deletes,
  fault injection) and can be switched to real Glue with credentials.
- Moto-backed ITCases exercise the SQL surface; credential-gated e2e
  tests run against a real Glue account, including tables created by
  other engines and a Kinesis read/filter/write through catalog
  resolved tables against Localstack.
- Documentation is executed as a test input: every CREATE CATALOG
  example in the docs pages is run through the factory, and the type
  mapping table is checked against the converter in both directions,
  including a composed-type sweep to depth 3.
- Architecture tests are wired for the module.

Co-authored-by: Samrat Deb <[email protected]>
Co-authored-by: Anthony Pounds-Cornish <[email protected]>

Report URL: 
https://github.com/apache/flink-connector-aws/actions/runs/36979892478

With regards,
GitHub Actions via GitBox

Reply via email to