tiperf

Auto analyze prometheus metrics (and more) in TiDB, to help perf tuning

Usage

Analyze all

tiperf timeline all

Output (just for now, rapidly revolving)

[2020-04-19 04:57:03 => 2020-04-20 05:19:03]
    ** started by workload changed, similarity 0.16
    2020-04-20 02:05:03 [tikv] -> down 17.1.4.135
    2020-04-20 05:18:03 [tikv] -> up 172.1.4.148
    ** lasted 24h22m0s, ended by workload changed, similarity 0.22
[2020-04-20 05:19:03 => 2020-04-20 05:21:03]
    ** started by workload changed, similarity 0.22
    ** lasted 2m0s, ended by workload changed, similarity 0.16
[2020-04-20 05:21:03 => 2020-04-20 16:11:03]
    ** started by workload changed, similarity 0.16
    ** lasted 10h50m0s

Analyze jitter and pike

tiperf timeline jitter pike

Analyze all but jitter

tiperf timeline all ~jitter

Analyze all, in specified prometheus address

tiperf --host 11.22.33.44 --port 5566 timeline all ~jitter

Get help

tiperf timeline

How

First, tiperf split the timeline into many periods by analyzing workload changes, then run detecting functions (the function/feature names passed from command line by user) to create infomation of each period.

A detecting function is like

Detector(data sources.Sources, period base.Period, found FoundEvents, con base.Console) (Events, error)

• data data query client, include prometheus or other clients
• period the start and end time to be analyzed
• found the events other detecting functions collected, dependencies can be configured when registering this function
• con stdin/stdout
• Events return the events of this function collected

The whole process is simple, get involed if you are interested