Reactive Mongo Performance
Two solutions wrote in order to compare the performance of REST MongoDb with reactive and non-reactive solution.
Based on Rodrigo Chaves solution's (https://itnext.io/reactive-microservices-with-spring-5-95c5f8cd03d0) I wrote the non-reactive version of the solution account-servlet.
account-reactive
Reactive solution written by Rodrigo Chaves available also at https://github.com/LINKIT-Group/spring-rest-reactive.
account-servlet
A non-reactive adaptation of Rodrigo's solution
Performance Test
I tested the two solutions using gatling with this configuration.
Number of Users: 100 Number of requests per seconds: 100 req/sec Duration of test: 1 minute
import java.net.HttpURLConnection
import io.gatling.core.Predef._
import io.gatling.core.scenario.Simulation
import io.gatling.http.Predef._
import scala.util.Random
import scala.concurrent.duration._
class Mongo extends Simulation {
val rampUpTimeSecs = 5
val testTimeSecs = 60
val noOfUsers = 100
val noOfRequestPerSeconds = 100
val baseURL = "http://localhost:8080"
val accountResourcePath = "/accounts"
object LoadAccountByCurrencies {
val all = exec(http("LABAccounts")
.get(accountResourcePath + "/")
.check(status.is(HttpURLConnection.HTTP_OK)))
}
val httpProtocol = http
.baseURL(baseURL)
.acceptHeader("application/json")
.userAgentHeader("Gatling")
val testScenario = scenario("LoadTest")
.during(testTimeSecs) {
exec(
LoadAccountByCurrencies.all
)
}
setUp(
testScenario
.inject(atOnceUsers(noOfUsers)))
.throttle(
reachRps(noOfRequestPerSeconds) in (rampUpTimeSecs seconds),
holdFor(testTimeSecs seconds))
.protocols(httpProtocol)
}I invoked only the URL GET http://localhost:8080/accounts which gets all the records in the collection.
My system configuration provides 4 cores cpu.
Result
account-servlet
================================================================================
---- Global Information --------------------------------------------------------
> request count 1431 (OK=1431 KO=0 )
> min response time 142 (OK=142 KO=- )
> max response time 9099 (OK=9099 KO=- )
> mean response time 4200 (OK=4200 KO=- )
> std deviation 2933 (OK=2933 KO=- )
> response time 50th percentile 5783 (OK=5783 KO=- )
> response time 75th percentile 6439 (OK=6439 KO=- )
> response time 95th percentile 7533 (OK=7533 KO=- )
> response time 99th percentile 8863 (OK=8863 KO=- )
> mean requests/sec 22.359 (OK=22.359 KO=- )
---- Response Time Distribution ------------------------------------------------
> t < 800 ms 460 ( 32%)
> 800 ms < t < 1200 ms 6 ( 0%)
> t > 1200 ms 965 ( 67%)
> failed 0 ( 0%)
================================================================================
account-reactive
================================================================================
---- Global Information --------------------------------------------------------
> request count 1018 (OK=1018 KO=0 )
> min response time 213 (OK=213 KO=- )
> max response time 23257 (OK=23257 KO=- )
> mean response time 5943 (OK=5943 KO=- )
> std deviation 3190 (OK=3190 KO=- )
> response time 50th percentile 5631 (OK=5631 KO=- )
> response time 75th percentile 7998 (OK=7998 KO=- )
> response time 95th percentile 11423 (OK=11423 KO=- )
> response time 99th percentile 14498 (OK=14498 KO=- )
> mean requests/sec 15.662 (OK=15.662 KO=- )
---- Response Time Distribution ------------------------------------------------
> t < 800 ms 25 ( 2%)
> 800 ms < t < 1200 ms 25 ( 2%)
> t > 1200 ms 968 ( 95%)
> failed 0 ( 0%)
================================================================================
Conclusion
-
Reactive responses average result are slower than the non-reactive one.
-
The throughput is about 30% faster in non-reactive solution.
-
Making the test with only one request, reactive solution is faster than non-reactive (127ms and 260ms).
-
Also, reactive solution opens a huge number of connections to MongoDb.
Considering the evidence of the above test, I guess org.springframework.data.mongodb.repository.ReactiveMongoRepository has a noticeable scalability issue.
I hope someone could disprove my conclusion and let me know why I got this result!