[
https://issues.apache.org/jira/browse/SPARK-61259?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tim Meehan updated SPARK-61259:
-------------------------------
Fix Version/s: connect-rust-4.3.0
> Raise KeyboardInterrupt on SIGINT during a running action in the Python
> client, as PySpark does
> -----------------------------------------------------------------------------------------------
>
> Key: SPARK-61259
> URL: https://issues.apache.org/jira/browse/SPARK-61259
> Project: Spark
> Issue Type: Bug
> Components: Connect
> Affects Versions: connect-rust-4.3.0
> Reporter: Tim Meehan
> Priority: Major
> Fix For: connect-rust-4.3.0
>
>
> In PySpark's Connect client, a SIGINT (Ctrl-C) during a long-running action
> raises KeyboardInterrupt at once, and the session keeps working. In the
> Python client of spark-connect-rust, the action keeps blocking and the signal
> is never handled.
> {code:python}
> import os, signal, threading
> from pyspark.sql import SparkSession, functions as F
> spark = SparkSession.builder.remote("sc://localhost:15002").getOrCreate()
> slow = spark.range(0, 10**12, 1, 2).agg(F.sum(F.sqrt("id")))
> threading.Timer(3.0, lambda: os.kill(os.getpid(), signal.SIGINT)).start()
> slow.collect()
> {code}
> Against an OSS Spark 4.3.0 Connect server on loopback:
> * pyspark-client 4.3.0: KeyboardInterrupt 3 seconds in;
> spark.range(3).count() then returns 3.
> * spark-connect-rust master 8768cc4: still blocked 12 seconds after the
> SIGINT. At 4ad36945 it was still blocked after 60 seconds.
> From source: the action waits on the Tokio runtime with the GIL released, and
> nothing under crates/ calls Python::check_signals (PyErr_CheckSignals) while
> it waits, so Python's SIGINT handler never runs. SPARK-61250, which runs the
> synchronous API on another thread, did not change this. spark.interruptAll()
> from another thread does cancel the query.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
