System KPI-aided Diagnosis ​

Availability ​

This feature is available since openGauss 1.0.0.

Introduction ​

KPIs are views of key performance indicators for kernel components or the entire system. Based on KPIs, users can learn about the real-time and historical running status of the system.

Benefits ​

• Summarized system load diagnosis

  Precise alarms for system load exceptions (overload, stall, and SLA exceptions) and precise system load profile

• Summarized system time model diagnosis

  Instance-level and query-level time model segmentation, diagnosing the root causes of instance and query performance problems

• Query performance diagnosis

  Database-level query summary, including top SQL, SQL CPU usage, I/O consumption, execution plan, and excessive hard parsing

• Diagnosis of disk I/O, index, and buffer performance problems

• Diagnosis of connection and thread pool problems

• Diagnosis of checkpoint and redo (RTO) performance problems

• Diagnosis of system I/O, LWLock, and wait performance problems

  Diagnosis of over 60 modules and over 240 key operation performance problems

• Function-level performance monitoring and diagnosis (by GSTRACE)

  Tracing of over 50 functions at the storage and execution layers

Description ​

openGauss provides KPIs of 11 categories and 26 sub-categories, covering instances, files, objects, workload, communication, sessions, threads, cache I/O, locks, wait events, and clusters.

Figure 1 shows the distribution of kernel KPIs.

Figure 1 Distribution of kernel KPIs

Enhancements ​

None.

Constraints ​

• Utility statements do not support normalization. Non-DML statements, such as CREATE, DROP, COPY, and VACUUM, are not supported.
• Currently, only the top-level normalized SQL statements are recorded. SQL statements in a stored procedure are not normalized, and only the SQL statements that call the stored procedure are recorded.

Dependencies ​

None.