[
https://issues.apache.org/jira/browse/SPARK-60742?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hyukjin Kwon resolved SPARK-60742.
----------------------------------
Resolution: Fixed
Issue resolved by pull request 142
[https://github.com/apache/spark-connect-rust/pull/142]
> Build SQL function exports in linear time in Spark Connect Rust
> ---------------------------------------------------------------
>
> Key: SPARK-60742
> URL: https://issues.apache.org/jira/browse/SPARK-60742
> Project: Spark
> Issue Type: Improvement
> Components: Connect
> Affects Versions: connect-rust-4.3.0
> Reporter: DB Tsai
> Assignee: DB Tsai
> Priority: Major
> Labels: pull-request-available
> Fix For: connect-rust-4.3.0
>
>
> pyspark.sql.functions constructs __all__ by checking callable names against a
> growing list. The current facade exports 568 unique names, making this
> bookkeeping quadratic in the number of exported names.
> Use a membership set while retaining the public list, exact export order,
> callable/alias identities, signatures and
> functions.builtin/functions.partitioning compatibility. Keep the change
> confined to export construction. SPARK-60683 / PR126 already defers
> SQL-function error imports; its export loop remains unchanged, so this is a
> separate improvement.
> Read-only attribution on the current published-change stack measured about
> 0.650 ms for the unchanged export assignment/loop across five warm batches of
> 100 iterations. Fresh import-only samples after library prerequisites were
> 1.820–2.433 ms. These measurements have different boundaries from whole-plan
> benchmarks and do not establish a candidate speedup.
> Validate matched, frozen baseline/candidate environments and the current
> combined stack. Preserve all 568 exports in the same list order, builtin's
> separate export list, aliases, generated signatures, star imports and
> serialized plans. Compare the unchanged cold-plan benchmark and warm guards
> with correctness/RPC validators, and report source/wheel provenance and
> cold-import boundaries. No JVM is required for these client/mock measurements.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
