Skip to content

chore: release 0.2.0 - #97

Merged
fightBoxing merged 1 commit into
lance-format:mainfrom
fightBoxing:release/v0.2.0
Nov 9, 2026
Merged

fightBoxing merged 1 commit into
lance-format:mainfrom
fightBoxing:release/v0.2.0

Conversation

@fightBoxing

Copy link
Copy Markdown
Collaborator

Summary

Bump the connector to 0.2.0, the first release shipping Apache Flink 2.1 support.

This is a release-only change: version numbers plus the user-facing docs refresh.

Changes

  • Set the parent version and the lance-flink.version property to 0.2.0.
  • Replace the 1.18 / 1.19 module list in the docs with the maintained 1.20 / 2.1 matrix — the 1.18 / 1.19 sources were already dropped from the build (<modules> and the CI matrix only cover 1.20 and 2.1).
  • Drop the OVERWRITE parallelism limitation note from the README; it is superseded by the job-id tagged commit protocol from fix(sink): move overwrite replace-once to committer in Flink 2.1 Sink V2 #96.
  • Point the install / vector-search docs at the lance-flink-2.1 artifact.

Verification

  • mvn clean package (JDK 17): BUILD SUCCESS
    • lance-flink-1.20: Tests run: 273, Failures: 0, Errors: 0, Skipped: 18
    • lance-flink-2.1: Tests run: 332, Failures: 0, Errors: 0, Skipped: 18
  • Artifacts: lance-flink-1.20-0.2.0.jar, lance-flink-2.1-0.2.0.jar

Release notes draft (v0.2.0)

New features

  • Apache Flink 2.1 support — FLIP-27 Source V2 and Sink V2 (Sink + SupportsCommitter) implementations in the new lance-flink-2.1 module (JDK 17+, Scala-free), including an async LanceSplitEnumerator.start().
  • Primary-key upsert / delete with composite keys, plus ALTER TABLE schema evolution.
  • Time travel via read.version / read.as-of-timestamp.
  • Read support for Lance datasets on Hadoop-compatible filesystems (hdfs://, tbdsfs://, viewfs://).
  • Documentation site covering install, configuration, DDL / DML / DQL and performance.

Bug fixes

  • Sink V2 no longer leaves orphan .lance files: fragment materialization is deferred out of write().
  • FragmentMetadata is serialized with an explicit field layout instead of Java native serialization.
  • write.mode=overwrite replaces the dataset once per job via the flink.job-id transaction property, and every commit in that mode carries the tag — parallel overwrite jobs no longer lose rows, and later checkpoints no longer overwrite earlier ones.
  • ARRAY<DOUBLE> reads go through Arrow's Float8Vector; RestNamespace sends the table schema as an Arrow IPC stream.

Version matrix

Module Flink version JDK
lance-flink-1.20 1.20.0 11+
lance-flink-2.1 2.1.1 17+

Dependencies: lance-core 7.0.0, lance-namespace 0.7.5, Apache Arrow 18.3.0.

Bump the connector to 0.2.0 and refresh the user-facing docs for the
Flink 2.1 release:

- update the parent version and lance-flink.version property;
- replace the 1.18 / 1.19 module list with the maintained 1.20 / 2.1
  matrix (1.18 and 1.19 sources were already dropped from the build);
- drop the OVERWRITE parallelism limitation note, superseded by the
  job-id tagged commit protocol.

Co-authored-by: Cursor <cursoragent@cursor.com>
@github-actions github-actions Bot added the chore Maintenance and chores label Nov 9, 2026
@fightBoxing
fightBoxing merged commit 1cb4636 into lance-format:main Nov 9, 2026
7 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

chore Maintenance and chores

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant