Microservice Chassis for Reactive Restful API with Spring Boot.
- 🐘 Gradle
- 🐳 Docker Engine
Run the restful application server:
$ docker compose up --build -dStop the restful application server:
$ docker compose stop chassisStop and remove containers and networks:
$ docker compose downOpenJ9 is a JVM implementation originally developed by IBM. Built to run in devices with limited memory. Some states that OpenJ9 is a good fit to microservices running in cloud solutions such as Kubernetes. OpenJ9 is an open-source project and its source code can be found on the GitHub page https://github.com/eclipse/openj9.
To confirm which implementation of JVM would be more efficient for this current project, I build two docker images, one
using adoptopenjdk:16-jre-hotspot and the other adoptopenjdk:16-jre-openj9. For the performance test had being used
the Apache HTTP server benchmarking tool (ab), which shows how many requests per second your application is capable to
serve.
Container Resources Usage:
| NAME | CPU % | MEM USAGE / LIMIT | MEM % | NET I/O | BLOCK I/O |
|---|---|---|---|---|---|
| chassis-openj9 | 0.00% | 165.1MiB / 3.844GiB | 4.20% | 209kB / 8.1kB | 42.1MB / 123kB |
| chassis-hotspot | 0.51% | 307.9MiB / 3.844GiB | 7.82% | 210kB / 8.1kB | 42.8MB / 119kB |
Loading test command:
ab -c 50 -n 1000000 -r http://<container-url>:<container-port>/api/v1/chassis
OpenJ9 JVM load testing:
Document Path: /api/v1/chassis
Document Length: 100 bytes
Concurrency Level: 50
Time taken for tests: 2493.535 seconds
Complete requests: 1000000
Failed requests: 0
Total transferred: 198000000 bytes
HTML transferred: 100000000 bytes
Requests per second: 401.04 [#/sec] (mean)
Time per request: 124.677 [ms] (mean)
Time per request: 2.494 [ms] (mean, across all concurrent requests)
Transfer rate: 77.54 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 0 4.2 0 606
Processing: 3 124 134.9 67 1491
Waiting: 3 113 126.2 63 1455
Total: 3 125 134.9 68 1491
Percentage of the requests served within a certain time (ms)
50% 68
66% 101
75% 152
80% 205
90% 331
95% 419
98% 520
99% 594
100% 1491 (longest request)
Resource usage at its peak:
| NAME | CPU % | MEM USAGE / LIMIT | MEM % | NET I/O | BLOCK I/O |
|---|---|---|---|---|---|
| chassis-openj9 | 120.94% | 266MiB / 3.844GiB | 6.76% | 575MB / 676MB | 49.7MB / 12MB |
HotSpot JVM load testing:
Document Path: /api/v1/chassis
Document Length: 100 bytes
Concurrency Level: 50
Time taken for tests: 2383.039 seconds
Complete requests: 1000000
Failed requests: 0
Total transferred: 198000000 bytes
HTML transferred: 100000000 bytes
Requests per second: 419.63 [#/sec] (mean)
Time per request: 119.152 [ms] (mean)
Time per request: 2.383 [ms] (mean, across all concurrent requests)
Transfer rate: 81.14 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 1 23.3 0 1298
Processing: 3 118 132.0 63 1819
Waiting: 3 107 123.3 58 1817
Total: 3 119 133.8 64 1819
Percentage of the requests served within a certain time (ms)
50% 64
66% 96
75% 142
80% 190
90% 319
95% 415
98% 517
99% 587
100% 1819 (longest request)
Resource usage at its peak:
| NAME | CPU % | MEM USAGE / LIMIT | MEM % | NET I/O | BLOCK I/O |
|---|---|---|---|---|---|
| chassis-hotspot | 136.09% | 488.5MiB / 3.844GiB | 12.41% | 572MB / 673MB | 49.8MB / 7.61MB |
Based on the testing results, it can be observed a smaller consumption of CPU of about 15% on the OpenJ9 JVM implementation while its throughput is about 18% higher than the Hotspot one. However, where the OpenJ9 really standouts is its performance regarding the memory usage. OpenJ9 consumes about 45% less memory than HotSpot implementation.
Every project has its own requirements and demands hence the JVM implementation chosen must align with its needs. For this one in particular which is using the Spring Boot Webflux (a reactive-stack web application built on a Reactive Streams API and running on the non-blocking Netty server) the OpenJ9 seems to be the best fit.
Bugs, feature requests and more, in GitHub Issues.