Yi Zhang created SPARK-41650:
--------------------------------
Summary: json expressions much slower in optimized mode
Key: SPARK-41650
URL: https://issues.apache.org/jira/browse/SPARK-41650
Project: Spark
Issue Type: Bug
Components: Spark Core, Structured Streaming
Affects Versions: 3.2.2
Reporter: Yi Zhang
I noticed spark structured streaming reading from Kafka json string into struct
type is much slower in spark-3.1+ than spark-3.0. Profiling reveals the json
expressions in spark-3.0 mostly on evaluate subExpr, while spark-3.1/3.2 spent
a lot time on writeField.
Suspect this may be related to SPARK-32948, so I tried with add a bogus option
from_json($"value", mySchema, Map("bogus_key"-> "bogus_value")
this turns off the optimization and the performance is much better. For
reference,
for same amount #records, it is 30 seconds vs. 3 minute on a task processing
500k records. This is big difference for a streaming job.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
