lhy1024/bench

bench workload

bench's Issues

Some reformat code for report

In the first two weeks, in order to ensure the fast running of the demo, we have merged some code that is not elegant enough. Now it’s time to refactor them.

• provide an interface for Prometheus to reduce code, such as L170-L178 and L184-L192

• Improve the readability of judging whether to balance

• format code and improve readability for merge report

Add bench tso

Introduce config file

we have too many env params and config

More github command params

we need to support more github command params

• specify the tikv branch
• test case

Push Docker image to Docker Hub doesn't normal run sometimes.

Support simulator mode

• tiup with only pd
• add workload with simulator

package bin tool to docker

we use some other tools, such as tpcc, sysbench, go-ycsb, it maybe a good idea to involve them by docker

add empty region scheduler simulator case

Add more metrics for scale out

we also need to add some other metrics, such as apply log duration per server and apply db mutex. Because we found that store which move peer out has more high apply db mutex

More desc about simulator

Add Github action to build docker image

We also need to add GitHub actions to build a docker image automated. I hope you can help me to do it.
@ZenoTan

Add exporter

we need to export data.
I plan to make it with importer and backup by BR to finish it.
So we need to implement BR client and to determine the needed parameter, it can be obtained through environment variables or through api server
@ZenoTan

Add load for scale out

Currently our scale-out test does not have external load, we want to add it later. How to determine the size of the load is a problem. Previously, we achieved it through continuous testing. The qps increased but the delay remained unchanged to determine a suitable external load.

Regression Testing

Simulator bugs

• GetLeader panic(nil)
• Merge number incorrect
• Merge schedule limit undefined
• Pd-ctl cannot set store limit

Refine readability of pd bench result

Current implementation only provides a simple report. We should add statistics module to let people understand better. Also, report code still needs refactoring.

Add mock server for test

Keep monitor pod to debug

I plan to provide this functionality through the api_server, the detailed process is:

1. After each invocation of the workload finished, api_server uses the rsync to backup Prometheus data and Grafana dashboard configuration files to an S3 backend.
2. Api_server supplies an API to search for all related monitoring data backups.
3. Api_server supplies an API to launch the Prometheus and Grafana dashboard pods on a K8s cluster and uses the backup data to restore the metrics.

Add hot test case

• add tpcc test