load tests and number of concurrent connections about rediscala HOT 11 CLOSED

Yep you should put val redis = RedisClient() outside of your main loop (here Action)

RedisClient() does a lot of thing (initialise a connection to redis server, pop 2 actors ...)

from rediscala.

Thanks for your reply, as the jedis client is not threadsafe, I put it at method level, and I did the same for your driver!
Trying now :)

from rediscala.

Now its works perfectly, and it's very fast!

My conclusions now are that the fastest test is redisscala with blocking code and a big thread pool (but the nb ofthreads will not be infinite...) , then the async redisscala, and then Jedis.

Thanks for your help!

from rediscala.

Do you have numbers ?

Could you show the "redisscala with blocking code" ?

from rediscala.

Yep,

In fact I've relunched the load test and it's quite similar in blocking or not blocking mode (I hope it's readable on github) :

blockingredissimulation :

STATISTICS
Requests Executions Response Time (ms)
Total OK KO Min Max Mean Std Dev 95th pct 99th pct Req/s
Global Information 10000 10000 0 0 80 1 5 10 30 906
request_blocking_redis 10000 10000 0 0 80 1 5 10 30 906

async redissimulation :

STATISTICS
Requests Executions Response Time (ms)
Total OK KO Min Max Mean Std Dev 95th pct 99th pct Req/s
Global Information 10000 10000 0 0 120 2 6 10 30 929
request_redis 10000 10000 0 0 120 2 6 10 30 929

The blocking redis code is very simple :

def blockingRedisQuery(q: String) = Action { 
    val result = Await.result(redis.get(q), 10 seconds).map(_.utf8String).getOrElse("not found")
    Ok(result) 
  }

It uses the default thread pool with 500 threads

from rediscala.

Thanks.
If you are going for async, you can reduce the thread pool size.

What was the number with Jedis?

from rediscala.

Also you should avoid doing many map on a Future when you can write it in one map :

.map(_.get).map(_.utf8String).map( result =>   
        Ok(result)
      )

.map(r => Ok(r.get.utf8String))

Each time you map, you add a Future to threadpool which will have to be scheduled

from rediscala.

Thanks for the tip, I'll try like this
For the thread pools, I have 500 threads for Jedis and "blocking redisscala", but for the async redisscala test, I have only 1 thread by cpu

from rediscala.

I've removed the useless map operation
Now I have EXACTLY the same performance with async and blockign simulations : 997 request/sec and 84% of requests <= 1ms

I can put the benchmark on github if you wish

from rediscala.

Yep good idea, it might help other.

from rediscala.

I've pushed everything here : https://github.com/loicdescotte/play-redis-benchmarks

from rediscala.