search

tikv / client-java

TiKV Java Client
https://tikv.github.io/client-java/
Apache License 2.0
111 stars 109 forks source link

optimize the backoff mechanism #411

Open zz-jason opened 2 years ago

zz-jason commented 2 years ago

Enhancement

Summary

There are two backoffs in the current backoff and retry mechanism, the first one is inside the callWithRetry() function, the other one is outside the function, can be expressed like the following pseudo code:

callWithRetry()
    backoff
backoff

These backoffs can be integrated into one backoff like the following:

callWithRetry()
    backoff

Details

The following slow log illustrates the problem, the first backoff happens at 

        {
            "name":"backoff BoRegionMiss",
            "start":"16:21:51.605",
            "end":"16:21:51.626",
            "duration":"21ms"
        },

the second backoff happens at:

        {
            "name":"backoff BoRegionMiss",
            "start":"16:21:51.626",
            "end":"16:21:51.666",
            "duration":"40ms"
        },

As you can see, BoRegionMiss is executed twice and the total backoff time is increased to more than 60 seconds.

{
    "start":"16:21:51.605",
    "end":"16:21:51.666",
    "duration":"61ms",
    "func":"get",
    "region":"{Region[8590843696] ConfVer[10372] Version[153] Store[8589948024] KeyRange[k1]:[k2]}",
    "key":"k1",
    "spans":[
        {
            "name":"getRegionByKey",
            "start":"16:21:51.605",
            "end":"16:21:51.605",
            "duration":"0ms"
        },
        {
            "name":"callWithRetry tikvpb.Tikv/RawGet",
            "start":"16:21:51.605",
            "end":"16:21:51.626",
            "duration":"21ms"
        },
        {
            "name":"gRPC tikvpb.Tikv/RawGet",
            "start":"16:21:51.605",
            "end":"16:21:51.605",
            "duration":"0ms"
        },
        {
            "name":"backoff BoRegionMiss",
            "start":"16:21:51.605",
            "end":"16:21:51.626",
            "duration":"21ms"
        },
        {
            "name":"backoff BoRegionMiss",
            "start":"16:21:51.626",
            "end":"16:21:51.666",
            "duration":"40ms"
        },
        {
            "name":"getRegionByKey",
            "start":"16:21:51.666",
            "end":"16:21:51.666",
            "duration":"0ms"
        },
        {
            "name":"callWithRetry tikvpb.Tikv/RawGet",
            "start":"16:21:51.666",
            "end":"16:21:51.666",
            "duration":"0ms"
        },
        {
            "name":"gRPC tikvpb.Tikv/RawGet",
            "start":"16:21:51.666",
            "end":"16:21:51.666",
            "duration":"0ms"
        }
    ]
}
github-actions[bot] commented 2 years ago

This issue is stale because it has been open 30 days with no activity.