How does maxRequestsPerConnection of istio work? - kubernetes

Question

everyone.
I have been learning istio and to understand how maxRequestsPerConnection works, I applied the manifest below.
apiVersion: networking.istio.io/v1alpha3
kind: DestinationRule
metadata:
name: httpbin
spec:
host: httpbin
trafficPolicy:
connectionPool:
http:
maxRequestsPerConnection: 1
httpbin is a sample service of istio.
I thought maxRequestsPerConnection means how many http requests are allowed per one TCP Connection, and istio would close tcp connection after pod received one http request in this case.
After applying, I sent some http requests using telnet. I thought istio would accept the request once and then close the TCP connection, but istio didn't.
$ telnet httpbin 8000
Trying 10.76.12.133...
Connected to httpbin.default.svc.cluster.local.
Escape character is '^]'.
GET /get HTTP/1.1
User-Agent: Telnet [ja] (Linux)
Host: httpbin
HTTP/1.1 200 OK
server: envoy
date: Sun, 07 Nov 2021 14:14:16 GMT
content-type: application/json
content-length: 579
access-control-allow-origin: *
access-control-allow-credentials: true
x-envoy-upstream-service-time: 9
{
"args": {},
"headers": {
"Host": "httpbin",
"User-Agent": "Telnet [ja] (Linux)",
"X-B3-Parentspanid": "b042ad708e2a47a2",
"X-B3-Sampled": "1",
"X-B3-Spanid": "b6a08d45e1a1e15e",
"X-B3-Traceid": "fc23863eafb0322db042ad708e2a47a2",
"X-Envoy-Attempt-Count": "1",
"X-Forwarded-Client-Cert": "By=spiffe://cluster.local/ns/default/sa/httpbin;Hash=d9bb27f31fe44200f803dbe736419b4664b5b81045bb3811711119ca5ccf6a37;Subject=\"\";URI=spiffe://cluster.local/ns/default/sa/default"
},
"origin": "127.0.0.6",
"url": "http://httpbin/get"
}
GET /get HTTP/1.1
User-Agent: Telnet [ja] (Linux)
Host: httpbin
HTTP/1.1 200 OK
server: envoy
date: Sun, 07 Nov 2021 14:14:18 GMT
content-type: application/json
content-length: 579
access-control-allow-origin: *
access-control-allow-credentials: true
x-envoy-upstream-service-time: 3
{
"args": {},
"headers": {
"Host": "httpbin",
"User-Agent": "Telnet [ja] (Linux)",
"X-B3-Parentspanid": "85722c0d777e8537",
"X-B3-Sampled": "1",
"X-B3-Spanid": "31d2acc5348a6fc5",
"X-B3-Traceid": "d7ada94a092d681885722c0d777e8537",
"X-Envoy-Attempt-Count": "1",
"X-Forwarded-Client-Cert": "By=spiffe://cluster.local/ns/default/sa/httpbin;Hash=d9bb27f31fe44200f803dbe736419b4664b5b81045bb3811711119ca5ccf6a37;Subject=\"\";URI=spiffe://cluster.local/ns/default/sa/default"
},
"origin": "127.0.0.6",
"url": "http://httpbin/get"
}
After this, I sent http request ten times using fortio, and I got the same result.
$ kubectl exec "$FORTIO_POD" -c fortio -- /usr/bin/fortio load -c 1 -qps 0 -n 10 -loglevel Warning http://httpbin:8000/get
14:22:56 I logger.go:127> Log level is now 3 Warning (was 2 Info)
Fortio 1.17.1 running at 0 queries per second, 2->2 procs, for 10 calls: http://httpbin:8000/get
Starting at max qps with 1 thread(s) [gomax 2] for exactly 10 calls (10 per thread + 0)
Ended after 106.50891ms : 10 calls. qps=93.889
Aggregated Function Time : count 10 avg 0.010648204 +/- 0.01639 min 0.003757335 max 0.059256801 sum 0.106482036
# range, mid point, percentile, count
>= 0.00375734 <= 0.004 , 0.00387867 , 30.00, 3
> 0.004 <= 0.005 , 0.0045 , 70.00, 4
> 0.005 <= 0.006 , 0.0055 , 80.00, 1
> 0.012 <= 0.014 , 0.013 , 90.00, 1
> 0.05 <= 0.0592568 , 0.0546284 , 100.00, 1
# target 50% 0.0045
# target 75% 0.0055
# target 90% 0.014
# target 99% 0.0583311
# target 99.9% 0.0591642
Sockets used: 1 (for perfect keepalive, would be 1)
Jitter: false
Code 200 : 10 (100.0 %)
Response Header Sizes : count 10 avg 230.1 +/- 0.3 min 230 max 231 sum 2301
Response Body/Total Sizes : count 10 avg 824.1 +/- 0.3 min 824 max 825 sum 8241
All done 10 calls (plus 0 warmup) 10.648 ms avg, 93.9 qps
$
In my understanding, the message Sockets used: 1 (for perfect keepalive, would be 1) means fortio used only one TCP connection.
I guessed clients used different tcp connection for each http requests first, but if it is true, telnet connection was not closed by foreign host and fortio used ten tcp connections.
Please teach me what the function of maxRequestsPerConnection is.