Hyukjin Kwon created SPARK-61315:
------------------------------------
Summary: Keep the Spark Connect listener bus consistent when a
listener fails, is removed or is re-added
Key: SPARK-61315
URL: https://issues.apache.org/jira/browse/SPARK-61315
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
A native listener that panics on the started event unwinds start()/toTable()
after the query exists on the server, so the caller gets no query handle.
Removing the last listener does not wait for a started event that another
thread is delivering. After a failed listener removal, or after a transport
failure of the listener stream, adding a listener again hangs, because the
server answers the new ADD with no response while it still holds the old
registration. The listener stream also calls the progress handlers with an
extra done notification. PySpark isolates a raising listener, waits for its bus
lock on removal, and has no Progress on the listener bus.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
