mateczagany opened a new pull request, #28929:
URL: https://github.com/apache/flink/pull/28929
## What is the purpose of the change
On every watermark, `emitResultAndCleanUpState` will scan the full
`leftState` of every key, which includes a full deserialization, including all
the elements which have have higher timestamps than the current watermark. This
work is completely unneeded, and for some specific Flink jobs, the speedup this
new operator brings can be very much noticable.
With RocksDB and ForSt, when iterating the state, the keys returned will
always be ordered in a serialized-key byte order, so it's easy to take an
advantage of that, and exit early from the iteration of the state.
## Brief change log
- Add `TemporalRowTimeJoinOperatorV2` (flink-table-runtime):
- probe side keyed by nested `LeftTimeIndexKey`
- on RocksDB/ForSt (detected via `getBackendTypeIdentifier()`, same idiom
as `InputSortBuffer` / `LateralSnapshotJoinOperator`), both states are iterated
in ascending time order with early exit at the watermark
- probe rows are re-sorted by arrival index before emission, preserving
the exact output order of the old operator
- unordered backends (heap) fall back to full scans equivalent to the old
operator
- Add `@ExecNodeMetadata(version = 2)` to `StreamExecTemporalJoin`.
`createJoinOperator` picks the operator by `getVersion()`. Version 1 plans keep
instantiating `TemporalRowTimeJoinOperator`
- Add restore test fixtures (compiled plan + savepoint) for
`stream-exec-temporal-join_2` for all four temporal join test programs.
- Register `LeftTimeIndexKeySerializer` in `TypeSerializerTestCoverageTest`.
## Verifying this change
- `TemporalRowTimeJoinOperatorV2Test`: all scenarios of
`TemporalRowTimeJoinOperatorTest` mirrored with identical inputs/expected
outputs
- each run on both heap and RocksDB backends
- asserting the sorted fast path is actually taken on RocksDB
- plus new scenarios for arrival-order emission with out-of-order probe
rows
- `LeftTimeIndexKeySerializerTest`: `SerializerTestBase` battery plus a
property test asserting unsigned lexicographic byte order equals numeric
`(timestamp, index)` order.
- `TemporalJoinRestoreTest`: version 1 savepoints restore into the old
operator, version 2 savepoints into the new one (both on RocksDB)
- `TemporalJoinITCase` passes unmodified; fresh plans now compile to
`stream-exec-temporal-join_2` and run the new operator end-to-end.
- The existing `TemporalRowTimeJoinOperatorTest` is untouched and still
passes
## Does this pull request potentially affect one of the following parts:
- Dependencies (does it add or upgrade a dependency): no
- The public API, i.e., is any changed class annotated with
`@Public(Evolving)`: no
- The serializers: no
- The runtime per-record code paths (performance sensitive): yes
- Anything that affects deployment or recovery: JobManager (and its
components), Checkpointing, Kubernetes/Yarn, ZooKeeper: no
- The S3 file system connector: no
## Documentation
- Does this pull request introduce a new feature? no
- If yes, how is the feature documented? not applicable
---
##### Was generative AI tooling used to co-author this PR?
<!--
If generative AI tooling has been used in the process of authoring this PR,
please
change the checkbox below to `[X]` and replace the placeholder in the
"Generated-by"
line with the tool name and version. Otherwise remove the "Generated-by"
line.
See the ASF Generative Tooling Guidance for details:
https://www.apache.org/legal/generative-tooling.html
You are responsible for the quality and correctness of every change in this
PR
regardless of the tooling used. Low-effort AI-generated PRs will be closed.
See
AGENTS.md for the full guidance.
-->
- [ ] Yes (please specify the tool below)
Generated-by: [Tool Name and Version]
--
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]
