Hyukjin Kwon created SPARK-60964:
------------------------------------
Summary: Retry each reattach, and release executions that fail at
once, in Spark Connect Rust as PySpark does
Key: SPARK-60964
URL: https://issues.apache.org/jira/browse/SPARK-60964
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
PySpark Connect retries each reattach of an execution with a retry policy of
its own ({{_has_next()}} makes a new {{Retrying}}), and releases an execution
whose first {{ExecutePlan}} call fails. The Rust client differs:
* It keeps one retry state for the whole stream: its attempts are never reset
and its elapsed-time limit counts from the start of the stream. A query, a
statement or the streaming listener that runs longer than the limit cannot
retry a transient {{UNAVAILABLE}} of a later reattach, so the call fails and
the execution is released.
* An error of the first {{ExecutePlan}} call, as an analysis or parse error is,
returns before the reattachable stream exists, so no {{ReleaseExecute}} is
sent, and the server keeps the execution until its detached timeout.
Retry each reattach, and release an execution that fails at once, as PySpark
does.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
