Hi All,

Following the earlier discussion about a Hadoop-independent GCS filesystem
[1], I would like to start the formal discussion for FLIP-603: GCS
SDK-native filesystem [2].  FLIP-603 proposes a new flink-gcs-fs-native
plugin that implements the gs:// filesystem directly using the Google Cloud
Storage Java SDK, without depending on Hadoop or gcs-connector.

The existing flink-gs-fs-hadoop plugin is already partly native: ordinary
filesystem operations are delegated to GoogleHadoopFileSystem, while its
RecoverableWriter uses the GCS SDK and server-side compose directly. The
proposal replaces the Hadoop-backed operations while retaining the existing
compose-based recovery design.

The initial proposal deliberately keeps the following outside its scope:
  - Deprecating flink-gs-fs-hadoop before the native implementation reaches
feature parity and production maturity.
  - Client-side encryption, which belongs to the shared SDK-native
filesystem work.
  - PathsCopyingFileSystem support.
  - GCS Hierarchical Namespace acceleration; standard copy-and-delete
rename remains the compatibility baseline.
  - A cloud-generic object-storage operations interface.


Looking forward to feedback an suggestion on the following:
  1. Should we retain the existing gs.* configuration namespace for
migration compatibility, or use a separate namespace for native-only
options?
  2. Should HNS-enabled buckets and atomic folder rename be supported in
the first version or added later as an optimisation?
  3. Are there existing gcs-connector directory, authentication, or
endpoint behaviours that must be included in the initial compatibility
target?
  4. Is preserving the current GCS recoverable-state format sufficient,
including the requirement to retain temporary-bucket and entropy settings
during migration?
  5. Should GCS bulk-copy support remain a follow-up, or is it required for
initial feature parity?


Bests,
Samrat

[1] https://lists.apache.org/thread/lhvkzjrwh7jqn1oj2m4g7dlhgggfrtrf
[2]
https://cwiki.apache.org/confluence/spaces/FLINK/pages/444334267/FLIP-603+FileSystems+GCS+SDK-native+filesystem
[3]
https://cwiki.apache.org/confluence/spaces/FLINK/pages/435028214/FLIP-597+Filesystems+Hadoop-less+Flink+filesystems

Reply via email to