A debugging to understand the slot mechanism of redis cluster

Once I was testing redis colony （ The redis The cluster uses a batch intel Performance hardware aep）, In order to find out the upper limit of performance before going online . So there is the following story .

Under normal circumstances , Because I don't have so many machines to fight redis colony , So I went to seek the help of the pressure testing platform of the testing department , To get large traffic （ In order to simulate the real online , The traffic recorded by bypass is used for playback ）, But because the pressure test platform is for business services （ request pb The format is customized by business ）, Without this redis Read write pressure measurement function . So we still have to rely on service , But there are not many service machines in our bypass test environment . So there are big data students' suggestions , Can your data all reach the same slot.

Well, this redis Of slot I heard , But I don't know much about , Because I only need one connection at ordinary times url+ The port is done . For example, the following configuration

So , Use gdb Open the service process , Go inside and spy slot How it works

gdb --args ./SearchUserFeatureServer --flagfile=conf/gflags.conf

Set breakpoints , Can press r, Start exploring

First of all, I found that when initializing the connection ,redis Will respond to , Return connection information like this

f2a9484ec91a9007863d015e2f72146940477efb 10.194.88.141:[email protected] master - 0 1656591917000 33 connected 656-1311

Copy

It describes

<id><ip:port><flags><master><ping-sent><pong-recv><config-epoch><link-state><slot><slot>...<slot>

So my code will look for this slot, Then proceed slot Of redis node Mapping of nodes . In the end, I'll put this 16384 The array of elements is full .

It's in the code int slot_ipport_index_vec[16384].

First of all, the cluster I connected has 24 individual node, One of them node The information is as follows ： His beginning slot yes 656, End slot yes 1311

Each paragraph slot The interval is 656, This 656 It's based on 16384 individual slot Count /24 individual node Got . Look at this 16384 An array , He is evenly distributed 24 individual node. Every node Monopolize a period slot.

First of all, I will go and redis Connect , Then he'll tell you redis How many node

And then this node The quantity will be used as the divisor ,16384/node Count =slot Section

Then each slot Intervals are all connected by the same node . Put this 16384 Divided into 0-655,656-1311,…

Subsequent requests will be based on key Of crc32 Figure out one. hash value , Project to this hash On the ring , Decide which node to use

This is a request key Come in , according to crc16 Do simple hash value calculation

The request will be based on hash Calculate to fall in 16384 The final node of an array of elements , It's like a hash Ring ,node Project once , request key Project once , Then find the intersection

But what I see is and redis establish init The process of , I don't know if it will be added or reduced later node How did you handle it , This needs further study ～