gaborgsomogyi opened a new pull request, #26134:
URL: https://github.com/apache/flink/pull/26134
## What is the purpose of the change
State processor API with RocksDB is extremely slow because it does the
following for key deduplication:
* Get keys from `columnFamility1`
* Get keys from `columnFamility2`
* Iterate over the keys
* Calls `KeyedStateReaderFunction.readKey`
* Removes the key from both column families
Since key removal is slow compared to other operations the overall
performance is unacceptable.
Instead of the mentioned method we've migrated to the following:
* Get keys from `columnFamility1`
* Get keys from `columnFamility2`
* Added `RocksMultiStateKeysIterator` which iterates on all column families
at the same time in an ordered way
As a result key deduplication happens on iteration time in a streaming
manner.
Important note that this change effects only state processor API, existing
checkpointing save/restore code paths are untouched.
## Brief change log
Changed state processor API key iteration deduplication algorithm.
## Verifying this change
Existing unit tests + performance/correctness tests.
## 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
--
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]
