[
https://issues.apache.org/jira/browse/IMPALA-14963?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18082820#comment-18082820
]
ASF subversion and git services commented on IMPALA-14963:
----------------------------------------------------------
Commit 6899d95119d2aaaa515d7be222d8c4d29cd72545 in impala's branch
refs/heads/master from Gokul Kolady
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=6899d9511 ]
IMPALA-14963: Query Profile AI Analysis Tool
The query profile AI analysis component provides an end-to-end workflow
for generating bottleneck analysis directly from the query details page,
so users can inspect root-cause hypotheses without leaving the Impala UI.
Implementation adds a new /query_ai_analysis tab and template, a
/query_ai_analysis_generate endpoint, and a backend pipeline that loads
the query profile JSON, redacts sensitive identifiers, runs an iterative
tool-calling analysis loop through ai_generate_text, and unredacts the
final response for display. It also extends OpenAI response parsing to
handle native tool_calls payloads and increases AI connection timeout to
better support longer analysis runs.
Tests added:
Unit tests verify that the AI-analysis helper logic can reliably parse
noisy model responses and extract assistant content across expected and
malformed payload shapes. They also validate the concurrency guard by
confirming multithreaded slot acquisition never exceeds the configured
max.
Manual Tests:
Manually verified the full end-to-end AI analysis flow on a local Impala
setup using a TPC-DS query profile to validate LLM interaction and
output correctness.
Code Generated-by: Cursor AI
Change-Id: Ia9376b8c56df75032f51abbb76b3ac6927f9107e
Reviewed-on: http://gerrit.cloudera.org:8080/24284
Reviewed-by: Jason Fehr <[email protected]>
Reviewed-by: Yida Wu <[email protected]>
Tested-by: Impala Public Jenkins <[email protected]>
> Query Profile AI Analyzer
> -------------------------
>
> Key: IMPALA-14963
> URL: https://issues.apache.org/jira/browse/IMPALA-14963
> Project: IMPALA
> Issue Type: Task
> Reporter: Gokul Kolady
> Assignee: Gokul Kolady
> Priority: Major
>
> Add an interface that takes in a JSON query profile, redacts it (using the
> profile redaction interface), and communicates with an AI endpoint via Impala
> AI Functions over multiple rounds to reach a conclusion about performance
> bottlenecks. During these communication rounds, the AI endpoint should have
> several tool call options to choose from in order to retrieve parsed sections
> of the query profile to drive its analysis. This AI analysis flow should be
> runnable from the Impala Coordinator Web UI in a given query's query details
> page.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
