Metric Issues
Does an Increase in Swap Usage Affect Workloads?
According to the Linux kernel swap policy, the system may initiate swapping to relieve the memory pressure even when the overall memory usage is low. After memory is swapped out, it is not proactively released, which leads to a continuous increase in the monitored swap usage. This increase is considered normal and does not affect workloads.
When the Metric "Time Percentage of Disk I/O in a Non-Idle State" Reaches 100%, Does Disk Load Reach Its Upper Limit?
This metric measures the percentage of time when the disk is not idle (there are I/O activities). It indicates how busy the disk is. The disk can process I/O requests in parallel. Even if the value of this metric reaches 100%, the disk may not reach its maximum processing capability. To determine if disk load has reached its upper limit, you need to consider the instance's storage type, IOPS, and disk read/write throughput. For details, see Instance Storage Types.
Why Does the Network Output Throughput Increase Periodically?
This metric measures the average amount of traffic sent from the measured host's NICs per second. Traffic generated during the backup of logs (such as binlogs and audit logs) is positively correlated with the database workload, which may cause Network Output Throughput to rise periodically. Such increases are normal.
What Are the Differences and Troubleshooting Suggestions for the Metrics "Row Locks Waits Transactions", "Current Row Lock Waits", and "Average Row Lock Wait Time"?
| Metric ID | Metric Name | Description | Source | Metric Type | Troubleshooting Suggestions |
|---|---|---|---|---|---|
| rds_innodb_lock_waits | Row Locks Waits Transactions | Number of InnoDB transactions waiting for row locks | For version 5.x, statistics are retrieved from the information_schema.innodb_lock_waits table. For version 8.0, statistics are retrieved from the performance_schema.data_lock_waits table. The performance_schema parameter must be enabled for version 8.0. | Instantaneous value | Check for blocked and waiting sessions and terminate blocked sessions, if any, to quickly restore operations. For details, see Managing Locks & Transactions. |
| rds_innodb_row_lock_current_waits | Current Row Lock Waits | Number of current InnoDB row lock waits | Status variable: Innodb_row_lock_current_waits | Instantaneous value | |
| rds_innodb_row_lock_time_avg | Average Row Lock Wait Time | Average time spent waiting for an InnoDB row lock, in milliseconds | Status variable: Innodb_row_lock_time_avg | Accumulated value (reset after the instance is rebooted) |
What Are the Differences and Troubleshooting Suggestions for Connection Metrics?
| Metric ID | Metric Name | Description | Source | Metric Type | Troubleshooting Suggestions |
|---|---|---|---|---|---|
| rds006_conn_count | Total Connections | Total number of connections that attempt to connect to the MySQL server | Status variable: Threads_connected | Instantaneous value |
|
| rds072_conn_usage | Connection Usage | Percent of used MySQL connections to the total number of connections | Ratio of the status variable Threads_connected to the parameter max_connections | Instantaneous value | |
| rds007_conn_active_count | Current Active Connections | Number of connections that are not in the sleep state | Status variable: Threads_running | Instantaneous value |
|
| rds_conn_active_usage | Active Connection Usage | Usage of active connections | Ratio of the status variable Threads_running to the parameter max_connections | Instantaneous value |
Why Is the Real-Time Replication Delay Still Displayed as 0 When the Latest Data Cannot Be Queried from a Read Replica?
The real-time replication delay reflects an instantaneous value. If monitoring by seconds is not enabled, the system collects replication metrics once per minute. The collected value only represents the delay at the moment of sampling and cannot indicate whether any delay occurred between sampling points.
If data is written to the primary instance but has not yet been replayed on the read replica between two sampling moments, the actual replication delay is not zero and the read replica cannot return the latest data. However, because no sampling occurs at that exact moment, the system cannot observe the delay. As a result, the replication delay metric still displays 0.
What is your overall rating for this page?
Thank you very much for your feedback. We will continue working to improve the documentation.See the reply and handling status in My Cloud VOC.
For any further questions, feel free to contact us through the chatbot.
Chatbot
