Trying play framework and trying to understand the impact of blocking/non-blocking DB calls in play framework. Also trying to see the impact of using Akka actors (blocking/non-blocking) in playframework for DB calls.
This repo contains a sample play framework (2.7.3) project that exposes the below mentioned APIs. The APIs read a random node from Neo4j (Graph DB) and so Neo4j needs to be running on the same machine on the default port.
GET /test - This is a test API to check if the play server is running fine.
GET /getData - This is a test API which reads a node from Neo4j and uses a blocking call to Neo4j
GET /getDataAsync - This is a test API which reads a node from Neo4j and uses a non-blocking call to Neo4j
GET /getDataAsync2 - This is a test API which invokes the Neo4j read twice(non-blocking) and was created to understand the impact of throughput when we have 2 calls to the DB.
GET /getDataAkka - This is a test API that uses 2 AKKA actors in its processing. The GraphEngine actor is used to read data from Neo4j and the RequestRouter is used to route the requests to a pool of GraphEngine actors. The GraphEngine actor uses a blocking call to Neo4j.
GET /getDataAkkaAsync - This is a test API that uses an actor to fetch the node from Neo4j. The actor(GraphEngineAsync) is injected into the controller and the actor makes a non-blocking call to Neo4j
GET /getDataAkkaAsync2 - This API is similar to /getDataAkkaAsync with the only difference being that it makes 2 calls to the actor and thereby helps understand the impact of throughput when we have 2 calls to the DB.
- Neo4j and Playframework requires Java 8 to run so Java 8 has to be installed
- Download and run neo4j.
- Clone the repo and execute the following commands
cd clonedfolder/
mvn clean package
mvn play2:start
mvn play2:stop //to stop the server after your tests
- The objective was to understand the impact of non-blocking calls to the DB and also the impact of using Akka actors in playframework. Hence there were no optimizations applied on the java runtime/linux configuration but all the tests were run in an indentical environment.
- The tests were run on a laptop running Ubuntu 16.04 with 4 cores (Intel® Core™ i5-6300U CPU @ 2.40GHz) and 16 GB of RAM.
- For every test (TestID in the below table) the same scenario was executed twice and the RunID reflects the same. The play server was restarted after every TestID execution.
- Netty server was configured for the test instead of the default Akkahttp server that comes with Play 2.7.
- AB tool (ab -k -c 200 -n 100000 http://localhost:9000/test) was used to simulate the load.
- All test result times are in millisecond.
- All tests were run with 200 concurrent requests and 100000 requests in total.
| Test ID | Run ID | API | RPS | Mean Response Time | Max Response Time | 99 percentile |
|---|---|---|---|---|---|---|
| 1 | 1 | http://localhost:9000/test | 11595 | 17 | 2371 | 90 |
| 1 | 2 | http://localhost:9000/test | 39059 | 5 | 1142 | 16 |
| 2 | 1 | http://localhost:9000/getData (Blocking) | 2937 | 68 | 6028 | 293 |
| 2 | 2 | http://localhost:9000/getData (Blocking) | 3900 | 51 | 1478 | 159 |
| 3 | 1 | http://localhost:9000/getDataAsync | 2792 | 71 | 2115 | 330 |
| 3 | 2 | http://localhost:9000/getDataAsync | 5137 | 38 | 1112 | 75 |
| 4 | 1 | http://localhost:9000/getDataAsync2 | 2293 | 87 | 2273 | 391 |
| 4 | 2 | http://localhost:9000/getDataAsync2 | 3318 | 60 | 1170 | 99 |
| 5 | 1 | http://localhost:9000/getDataAkka (Blocking) | 2618 | 76 | 2372 | 300 |
| 5 | 2 | http://localhost:9000/getDataAkka (Blocking) | 3416 | 58 | 1117 | 211 |
| 7 | 1 | http://localhost:9000/getDataAkkaAsync | 2924 | 68 | 2241 | 318 |
| 7 | 2 | http://localhost:9000/getDataAkkaAsync | 4981 | 40 | 1170 | 87 |
| 8 | 1 | http://localhost:9000/getDataAkkaAsync2 | 2350 | 85 | 2096 | 417 |
| 8 | 2 | http://localhost:9000/getDataAkkaAsync2 | 3370 | 59 | 1123 | 94 |