bug: Modifying the weight to 0 in APISIx version 3.9.1, there are still requests to be forwarded

[1450793561]

Current Behavior

There are 2 nodes in total using a round-robin algorithm. After modifying the weight of one node to 0, requests still continue to be forwarded to it for approximately 1 minute, with the business concurrency at 500 QPS.

The reproduction steps are as follows:

1. The upstream configuration is as follows

 {
  "nodes": [
    {
      "host": "10.2.137.151",
      "port": 9180,
      "weight": 1
    },
    {
      "host": "10.2.139.37",
      "port": 8888,
      "weight": 1
    }
  ],
  "timeout": {
    "connect": 60,
    "send": 60,
    "read": 60
  },
  "type": "roundrobin",
  "checks": {
    "active": {
      "concurrency": 10,
      "healthy": {
        "http_statuses": [
          200,
          302,
          201
        ],
        "interval": 2,
        "successes": 2
      },
      "http_path": "/",
      "req_headers": [
        "User-Agent: curl/7.29.0"
      ],
      "timeout": 2,
      "type": "tcp",
      "unhealthy": {
        "http_failures": 2,
        "http_statuses": [
          429,
          404,
          500,
          501,
          502,
          503,
          504,
          505
        ],
        "interval": 1,
        "tcp_failures": 2,
        "timeouts": 3
      }
    }
  },
  "scheme": "http",
  "pass_host": "pass",
  "name": "privcbs",
  "keepalive_pool": {
    "idle_timeout": 60,
    "requests": 1000,
    "size": 320
  }
}

2. Conduct ab pressure testing

 ab -n 10000 -c 20 http://xxxxx/assets/js/vendors~chunk-contract-sign~sign.196c44f2de1ff1d5063a.js

3. During the ab pressure test process, change the 10.2.139.37:8888 weight to 0
4. Check and view the apisix request log，that 10.2.139.37:8888 of node there still has requests
5. There are 3,195 non-200 requests，ab pressure test results

 ab -n 10000 -c 20 http://xxxx/assets/js/vendors~chunk-contract-sign~sign.196c44f2de1ff1d5063a.js
This is ApacheBench, Version 2.3 <$Revision: 1430300 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking sit-qysapp.cbs.bacic5i5j.com (be patient)
Completed 1000 requests
Completed 2000 requests
Completed 3000 requests
Completed 4000 requests
Completed 5000 requests
Completed 6000 requests
Completed 7000 requests
Completed 8000 requests
Completed 9000 requests
Completed 10000 requests
Finished 10000 requests


Server Software:        APISIX/3.9.1
Server Hostname:       xxxxxx
Server Port:            80

Document Path:          /assets/js/vendors~chunk-contract-sign~sign.196c44f2de1ff1d5063a.js
Document Length:        150 bytes

Concurrency Level:      20
Time taken for tests:   5.482 seconds
Complete requests:      10000
Failed requests:        6805
   (Connect: 0, Receive: 0, Length: 6805, Exceptions: 0)
Write errors:           0
Non-2xx responses:      3195
Total transferred:      29623980 bytes
HTML transferred:       24820735 bytes
Requests per second:    1824.05 [#/sec] (mean)
Time per request:       10.965 [ms] (mean)
Time per request:       0.548 [ms] (mean, across all concurrent requests)
Transfer rate:          5276.90 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.5      0      12
Processing:     1   11   8.4      9      61
Waiting:        1   10   8.3      9      61
Total:          1   11   8.4      9      61

Percentage of the requests served within a certain time (ms)
  50%      9
  66%     13
  75%     15
  80%     17
  90%     22
  95%     27
  98%     33
  99%     37
 100%     61 (longest request)

[Baoyuantop]

Hi @1450793561, can you test it with a new version? This version is a bit old, I will also try to reproduce it.

[1450793561]

Hi @1450793561, can you test it with a new version? This version is a bit old, I will also try to reproduce it.

We are currently using version 3.9.1 in our production environment. It’s difficult to assess the impact of upgrading, and we are also unsure whether the upgraded version will be problem-free.