soin08 commented on PR #29268: URL: https://github.com/apache/flink/pull/29268#issuecomment-5859986695
@gaborgsomogyi @gyfora, I performed the following test for [PR #29268](https://github.com/apache/flink/pull/29268): - Generated a savepoint containing 100 million keyed-state entries (`BIGINT` keys and `INT` values). - Built two Flink images from `master`: - **Before:** [`b25e303d5353e9e1a252712e746eb1b372df5729`](https://github.com/apache/flink/commit/b25e303d5353e9e1a252712e746eb1b372df5729), immediately preceding the PR. - **After:** [`6d87fcc958a1def740991523f44c715e8e69cc0d`](https://github.com/apache/flink/commit/6d87fcc958a1def740991523f44c715e8e69cc0d), including the PR changes. - Set up two otherwise identical Flink clusters and SQL Gateway endpoints using the respective images. - Queried the savepoint in batch mode with parallelism 100: ~~~sql SET 'parallelism.default' = '100'; LOAD MODULE state; SET 'execution.runtime-mode' = 'BATCH'; CREATE TABLE long_int_state ( state_key BIGINT, state_value INT, PRIMARY KEY (state_key) NOT ENFORCED ) WITH ( 'connector' = 'savepoint', 'state.backend.type' = 'rocksdb', 'state.path' = 's3a://flink-common/long-int-100m', 'operator.uid' = 'long-int-kv-operator', 'fields.state_value.state-name' = 'long-int-value' ); SELECT COUNT(*) FROM long_int_state; ~~~ Before the PR, managed memory was not visible as allocated to the job in the Flink UI, and the RocksDB metrics were empty. After the PR, the job used the available managed-memory budget and RocksDB metrics were reported correctly. This confirms that the PR fixes managed-memory integration and metrics exposure for this use case. However, the test did not show a performance improvement: both queries completed in approximately the same time. **Before the PR:**  **After the PR:**  -- 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]
