Revolyssup commented on PR #12353:
URL: https://github.com/apache/apisix/pull/12353#issuecomment-2999367932
@bzp2010 I ran a benchmark to test whether the `shdict:get` brings too much
performance degradation. But based on below results I don't think this should
be an issue.
## Configuration used
config.yaml
```yaml
apisix:
node_listen: 1984
deployment:
role: data_plane
role_data_plane:
config_provider: yaml
nginx_config:
worker_processes: 8
discovery:
nacos:
host:
- "http://nacos:[email protected]:8848";
prefix: "/nacos/v1/"
fetch_interval: 1
weight: 1
timeout:
connect: 2000
send: 2000
read: 5000
```
apisix.yaml
```yaml
routes:
-
uri: /hello
upstream:
service_name: APISIX-NACOS
discovery_type: nacos
type: roundrobin
#END
```
Here are the results.
```bash
❯ wrk -c 10 -t 5 -d 30s -R 50000 http://localhost:1984/hello
Running 30s test @ http://localhost:1984/hello
5 threads and 10 connections
Thread calibration: mean lat.: 2999.418ms, rate sampling interval: 10395ms
Thread calibration: mean lat.: 2885.546ms, rate sampling interval: 10125ms
Thread calibration: mean lat.: 2857.330ms, rate sampling interval: 10059ms
Thread calibration: mean lat.: 2967.000ms, rate sampling interval: 10231ms
Thread calibration: mean lat.: 2953.790ms, rate sampling interval: 10256ms
Thread Stats Avg Stdev Max +/- Stdev
Latency 11.12s 3.06s 16.47s 57.49%
Req/Sec 4.52k 82.05 4.63k 60.00%
689405 requests in 30.00s, 110.45MB read
Requests/sec: 22980.47
Transfer/sec: 3.68MB
```
```bash
❯ wrk -c 10 -t 5 -d 30s -R 50000 http://localhost:1984/hello
Running 30s test @ http://localhost:1984/hello
5 threads and 10 connections
Thread calibration: mean lat.: 3060.053ms, rate sampling interval: 10584ms
Thread calibration: mean lat.: 3084.401ms, rate sampling interval: 10772ms
Thread calibration: mean lat.: 3254.041ms, rate sampling interval: 11198ms
Thread calibration: mean lat.: 3093.539ms, rate sampling interval: 10985ms
Thread calibration: mean lat.: 3062.424ms, rate sampling interval: 10772ms
Thread Stats Avg Stdev Max +/- Stdev
Latency 11.89s 3.36s 17.74s 56.68%
Req/Sec 4.13k 39.11 4.18k 60.00%
623009 requests in 30.00s, 99.81MB read
Requests/sec: 20767.18
Transfer/sec: 3.33MB
```
## Result
The QPS dropped from 22980.47 to 20767.18 with ~2k difference. I think this
should be acceptable given the issue with events library this PR solves.
According to me, it's an okay tradeoff
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
