Redis 客户端可以同时更新缓存导致错误的 state 被保存

Question

I have been building a simple application that uses Redis as cache to store data regarding a game where each user has a score and after a user completes a task the score is updated for the user.

My problem is when a user completes a task his score is updated which means that it will update the record in redis by replacing the previous value with the new one (in my case it will replace the entire room object with the new one even though the room has not changed but only the score of the player inside the room has changed).

The thing is if multiple users complete a task at the same time they will send each at the same time the new record to redis and only the last one will receive the update.

For example:

In the redis cache this is the starting value: { roomId: "...", score:[{ "player1": 0 }, { "player2": 0 }] }

Player 1 completes a task and sends:

{ roomId: "...", score:[{ "player1": 1 }, { "player2": 0 }] }

At the same time Player 2 completes a task and sends:

{ roomId: "...", score:[{ "player1": 0 }, { "player2": 1 }] }

In the redis cache first it will be saved the value received from Player1 let's say and then the value from player 2 which means that the new value in the cache will be:

{ roomId: "...", score:[{ "player1": 0 }, { "player2": 1 }] }

Even though this is wrong because the correct value would be: { roomId: "...", score:[{ "player1": 1 }, { "player2": 1 }] } where both changes are present.

At the moment I am also using a pub/sub system to keep track of changes so that does are reflected to every server and each user connected to the server.

What can I do to fix this? For reference consider the following image as the architecture of the system:

Answer 1

The issue appears to be that you are interleaving one read/write set of operation with others, which leads to using stale data while updating keys. Fortunately, the fix is (relatively) easy: just combine your read/write chunk of operations into a single atomic unit, using either a Lua script , a transaction or, even easier, through a single RedisJSON command.

Here is an example using RedisJSON. Prepare your JSON key/document which will hold all the scores for the room first, using the JSON.SET command:

> JSON.SET room:foo $ '{ "roomId": "foo", "score": [] }'
OK

After that, use the JSON.ARRAPPEND command once you need to append an item to the score array:

> JSON.ARRAPPEND room:foo $.score '{ "player1": 123 }'
1
...
> JSON.ARRAPPEND room:foo $.score '{ "player2": 456 }'
2

Getting back the whole JSON document is as easy as running:

> JSON.GET room:foo
"{\"roomId\":\"foo\",\"score\":[{\"player1\":123},{\"player2\":456}]}"