How to use the latency insights feature - Tair (Redis® OSS-Compatible)

The latency insights feature of CloudDBA collects latency statistics for all commands and custom events in a Tair (Redis OSS-compatible) database. The statistics are accurate to the microsecond. You can use this feature to troubleshoot failures and performance degradation in Tair (Redis OSS-compatible) databases.

Overview

Redis 2.8.13 introduced a latency monitoring feature. This feature uses an event-based mechanism to help you find and troubleshoot potential latency problems. It can only retrieve data from the last 160 seconds and stores only the event with the highest latency per second.

Latency insights is an advanced latency statistics feature provided by Tair (Redis OSS-compatible). It records the execution time of up to 27 events and all Redis commands. It also saves all latency statistics from the last three days. Latency insights has the following features:

Persistent: supports data persistence and latency spike tracing.
High-precision: allows full events to be monitored within milliseconds.
High-performance: supports asynchronous implementations with minimal impact on performance.
Real-time: supports real-time data queries and aggregation operations.
Multidimensional: provides comprehensive latency data that allows you to analyze an instance based on events, time, and latency.

Prerequisites

The Tair (Redis OSS-compatible) instance must use one of the following minor versions. For more information about how to update a minor version, see Update the minor version of an instance.

Minor version 1.6.9 or later if the instance is a Tair (Enterprise Edition) memory-optimized instance. If you want to collect statistics about Tair module commands, update the minor version to 1.7.28 or later.
Minor version 5.1.4 or later if the instance is a Redis Open-Source Edition 5.0 instance.
Minor version 0.1.15 or later if the instance is a Redis Open-Source Edition 6.0 instance.
Minor version 7.0.0.6 or later if the instance is a Redis Open-Source Edition 7.0 instance.

Billing

This feature is available free of charge.

Procedure

Log on to the console and go to the Instances page. In the top navigation bar, select the region in which the instance is deployed. Then, find the instance and click its ID.
In the navigation pane on the left, click CloudDBA > Latency Insights.
On the Latency Insights page, select a time range and click Search. By default, statistics from the last 5 minutes are displayed.
For cluster and read/write splitting instances, you can view the statistics for Data Node and Proxy Node.
Note
You can query data from the last three days. The selected time range cannot exceed one hour.

Click an event name or a number in the list to view the trend of the corresponding metric over time.

In the trend chart, you can also select the corresponding metrics to view their trends over time.

Note

Only commands or events that take longer than the specified threshold are recorded and displayed. If you encounter instance latency issues, see Suggestions for handling common latency events.

pqus

Metric	Description
Event	The name of the event. Examples include ExpireCycle, EventLoop, Ping, Scan, Commands, and Info. For more information, see Appendix: Common special events.
Total	The number of events.
Average Latency (us)	The average latency of the event, in microseconds (us).
Maximum Latency (us)	The maximum latency of the event, in microseconds (us).
Aggregated count (< 1 ms)	The aggregated count of events with a latency of less than 1 ms. Click the icon to view statistics for smaller time ranges, such as <1 us, <2 us, <4 us, <8 us, <16 us, <32 us, <64 us, <128 us, <256 us, <512 us, and <1 ms. Note Counting rule: For example, <1 us counts events with a latency between 0 us and 1 us. <2 us counts events with a latency between 1 us and 2 us.
<2ms <4ms ... >33s	The number of events in this latency range. Note Counting rule: For example, <2 ms counts events with a latency between 1 ms and 2 ms. >33s counts events with a latency greater than 33s.

Appendix: Common special events

Category	Event	Threshold	Description
Memory eviction	EvictionDel	30 ms	The amount of time required to delete evicted keys in a specific eviction cycle.
	EvictionLazyFree	30 ms	The amount of time that background threads take to release the memory in a specific eviction cycle.
	EvictionCycle	30 ms	The amount of time required to perform an eviction. An eviction cycle involves the time required to select and delete data that you want to evict and the time spent on waiting for background threads.
Memory defragmentation	ActiveDefragCycle	100 ms	The amount of time required to defragment memory.
Rehash	Rehash	100 ms	The amount of time required to perform a rehash.
Data structure upgrade	ZipListConvertHash	30 ms	The amount of time required to convert a ziplist to a dictionary by means of hash encoding.
	IntsetConvertSet	30 ms	The amount of time required to convert an intset to a set by means of set encoding.
	ZipListConvertZset	30 ms	The amount of time required to convert a ziplist to a skiplist by means of ziplist encoding.
Append-only file (AOF)	AofWriteAlone	30 ms	The amount of time required to write an AOF as expected.
	AofWrite	30 ms	The amount of time required to write an AOF. Each time an AOF is successfully written, an AofWrite event and one of the following three events are recorded: AofWriteAlone, AofWriteActiveChild, and AofWritePendingFsync.
	AofFstat	30 ms	The amount of time required to obtain the information about an AOF.
	AofRename	30 ms	The amount of time required to rename an AOF.
	AofReWriteDiffWrite	30 ms	The amount of time consumed by an incremental AOF write performed by a parent process after its child process rewrites an AOF.
	AofWriteActiveChild	30 ms	The amount of time required to write an AOF to a disk. Other child processes may write data to the disk when the AOF is written.
	AofWritePendingFsync	30 ms	The amount of time required to write an AOF. A background process may be performing a fsync operation when the AOF is written.
Redis database (RDB) file	RdbUnlinkTempFile	50 ms	The amount of time required to delete a temporary RDB file after a bgsave child process is terminated.
Others	Commands	30 ms	The amount of time required to run a command that is not tagged with fast.
	FastCommand	30 ms	The amount of time required to run a command that is tagged with fast and has the time complexity of O(1) or O(log N). For more information about such commands, see the FastCommands section of this topic.
	EventLoop	50 ms	The amount of time required to run a main event loop.
	Fork	100 ms	The amount of time required to call a fork operation.
	Transaction	50 ms	The actual amount of time consumed by a transaction.
	PipeLine	50 ms	The amount of time consumed by a multi-threaded pipeline.
	ExpireCycle	30 ms	The amount of time required to clear expired keys at a time.
	ExpireDel	30 ms	The amount of time required to delete expired keys in a specific cycle for clearing the keys.
	SlotRdbsUnlinkTempFile	30 ms	The amount of time required to delete a temporary RDB file from a slot after a bgsave child process is terminated.
	LoadSlotRdb	100 ms	The amount of time required to load an RDB file from a slot.
	SlotreplTargetcron	50 ms	The amount of time required to load an RDB file from a slot to a temporary database and then migrate the file to a destination database by using a child process.