:Intelligent O&M: Getting started with EMR AI Assistant

Function

Use natural language to quickly get information about StarRocks product features, architecture, version differences, best practices, and cloud deployment solutions.

Typical questions

• What issues are fixed in the latest version of StarRocks?

• What are the instances under my account?

• What does the pipeline_exec_thread_pool_thread_num parameter do?

For example, after a user enters the question What are the advantages of the storage-compute separation architecture of Serverless StarRocks?, the AI assistant retrieves relevant documents and returns a structured answer covering four core advantages: 1. Inexpensive and seamlessly scalable storage (supports object storage such as Alibaba Cloud OSS, S3, and HDFS, offering low storage costs and nearly unlimited capacity expansion); 2. Elastically scalable computing power (CN nodes are stateless and can be dynamically added or removed in seconds without data redistribution or cross-node shuffles); 3. High-performance cache mechanism (Hot data is cached on local disks. When a cache hit occurs, performance is comparable to storage-compute integrated architectures. The cache lifecycle is managed automatically); 4. Enhanced resource isolation and stability (Compute and storage resources are completely decoupled).

Function

• Check the health status of an instance using natural language. The assistant analyzes issues like CPU spikes, memory leaks, full disks, query timeouts, bad tablets, and metadata lock conflicts, then provides repair steps.

• Retrieve the running status of SQL queries in an instance by using natural language. It provides in-depth insights by analyzing top SQL queries, aggregating different SQL types, and tracking daily/weekly trends in SQL execution. It also supports detailed diagnosis and analysis of individual SQL queries.

Typical questions

• The CPU usage spiked around 10:00. What was the cause?

• What are the top 10 slow SQL queries for the instance today, and how do they compare to yesterday?

After EMR AI Assistant determines the cluster type and performs a Top SQL query, it returns a Top SQL Query Summary (including cluster name, time range, and sort order) and a Top 10 SQL Summary Table. The table columns include QueryID, SQL fingerprint (first 80 characters), CPU (ms), Mem (MB), Scan (MB), and Status. The table is sorted by CPU Time in descending order, allowing you to quickly see which SQL queries consume the most resources.

Function

• Intelligently parses slow query logs to highlight hotspots and data skew in SQL execution plans, and generates optimization suggestions for indexes, partitions, and materialized views.

• Intelligently analyzes the current running status and load of an instance and provides parameter tuning suggestions.

Typical questions

• QueryID: 6cb2***-****-****-**-******8716 is currently slow. How can I optimize it?

• Disk usage has been consistently high for the past week. How can I resolve this?

EMR AI Assistant performs a Query Profile analysis on the StarRocks query (Query ID: 056b***4359) and identifies a serious performance issue. Query Performance Overview: Total execution time is 4.571 seconds, and the CPU time is 3 minutes and 20 seconds (98.31%). It scans 592 million rows and returns 591 million rows (almost a full table scan), reads 3.14 GB from disk, and produces a result size of 25.51 GB. Key Performance Bottleneck Analysis: 1. OLAP_SCAN operator (3.163 seconds) - Scans nearly 600 million rows without an effective predicate filter. The root cause is a missing WHERE clause or a filter condition that cannot be pushed down. 2. PROJECT operator (3.416 seconds) - Processes a projection for 591 million rows. The root cause is likely a SELECT statement containing a large number of columns or complex expressions.