Zhihua Deng created HIVE-29583:
----------------------------------
Summary: Profile the underlying JDBC execution in Metastore
Key: HIVE-29583
URL: https://issues.apache.org/jira/browse/HIVE-29583
Project: Hive
Issue Type: Improvement
Reporter: Zhihua Deng
Assignee: Zhihua Deng
Currently, when Hive Metastore operations—such as get_partitions or drop_table
are slow, there is limited visibility into whether the delay is in the
Metastore logic, the DataNucleus ORM overhead, or the underlying RDBMS
execution.
A single call(e.g, get_table) can trigger complex SQL joins or multiple
round-trips to the backend database. Without explicit profiling of the JDBC
execution, administrators are forced to rely on database-side logs (like
Postgres slow query logs), which are often decoupled from the HMS context,
making it hard to correlate a slow SQL statement with a specific Hive user or
query.
By profiling the JDBC executions, Admins can immediately see if a database
index is missing or if the RDBMS is under-resourced and identify "expensive"
patterns (e.g., users querying tables with 100k+ partitions without filters).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
