Kavpreet Grewal created SPARK-60231:
---------------------------------------
Summary: Improve the put/merge operation in ListState with TTL
when there are multiple values
Key: SPARK-60231
URL: https://issues.apache.org/jira/browse/SPARK-60231
Project: Spark
Issue Type: Bug
Components: Structured Streaming
Affects Versions: 5.0.0, 4.4.0
Reporter: Kavpreet Grewal
SPARK-54118 made ListStateImpl write a list with a single putList/mergeList,
but ListStateImplWithTTL still writes one value at a time: put(array),
appendList(array) and TTL eviction all do one put/merge per value. With
RocksDB, each merge first does a get to track the number of rows, and that get
folds every pending merge operand for the key, so writing N values costs O(N^2).
We should write the values with a single putList/mergeList, like ListStateImpl
does.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
