Hyukjin Kwon created SPARK-61241:
------------------------------------
Summary: A panicking Rust progress handler or listener aborts the
process, wedges the listener bus, or cuts a prefetched local iterator short
Key: SPARK-61241
URL: https://issues.apache.org/jira/browse/SPARK-61241
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
- A progress handler that panics on an update and again on the final done call
aborts the process ("panic in a destructor during cleanup"): ProgressTracker's
Drop calls the handlers while the thread is unwinding. Repro: a handler that
panics on every call, with `SELECT reflect('java.lang.Thread','sleep',4000L)`.
- With to_local_iterator(true), a progress handler that panics on the prefetch
task ends the iterator early with Ok and the rows so far (a silent truncation).
- A native Rust listener that panics on an event unwinds its worker: the
registrations stay on the bus, no stream error is reported, and a listener
added later never gets events.
- Fix: no handler call from Drop while panicking; a callback panic is a
subscription failure (reported like a returned error); a new subscription
removes the server listener first.
- Official PySpark has no Rust callback path, so this is Rust-only; the Python
bridge's behavior is unchanged.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
