istio - Envoy 中的“max_connections”到底是什么意思？

Question

我的问题是关于Envoy中“cluster.CircuitBreakers.Thresholds.max_connections”的真正含义。

特使文档中的cluster.CircuitBreakers.Thresholds.max_connections解释为：

Envoy 将与上游集群建立的最大连接数。如果未指定，则默认值为 1024。

Istio使用 envoy 作为 sidecar。最近我们尝试了断路示例，但总是发现连接数比我们配置的要多。

所以我们做另一个测试，如下所示：

在 istio 中添加两个服务：
- echo 客户端：1 个 pod，下游，将向 echo 服务器发送 HTTP 请求
- 回声服务器：2 个 pod，上游。

服务吊舱：

[root@k8s-master istio-1.0.3]# kubectl get pod -o wide
NAME                          READY   STATUS    RESTARTS   AGE     IP            NODE     NOMINATED NODE
echoclient-84485fbc5c-zxlv8   2/2     Running   0          8s      10.244.2.79   node02   <none>
echoserver-5655768fb9-smsvb   2/2     Running   0          23h     10.244.2.65   node02   <none>
echoserver-5655768fb9-srsq2   2/2     Running   0          7h52m   10.244.2.73   node02   <none>

为 echo server 配置的目标规则，下面显示了 envoy 中的相应信息。（最大连接数为 2）

istio 代理配置输出：

[root@k8s-master istio-1.0.3]# istioctl proxy-config clusters echoclient-84485fbc5c-zxlv8 --fqdn echoserver.default.svc.cluster.local -o json
[
    {
        "name": "outbound|8080||echoserver.default.svc.cluster.local",
        "type": "EDS",
        "edsClusterConfig": {
            "edsConfig": {
                "ads": {}
            },
            "serviceName": "outbound|8080||echoserver.default.svc.cluster.local"
        },
        "connectTimeout": "1.000s",
        "circuitBreakers": {
            "thresholds": [
                {
                    "maxConnections": 2
                }
            ]
        }
    }
]

从 echo 客户端到每个服务器同时发出多个请求（每个线程 40 个请求）。

结果：

[root@k8s-master istio-1.0.3]# kubectl exec -it echoclient-84485fbc5c-zxlv8 /bin/bash
Defaulting container name to echoclient.
Use 'kubectl describe pod/echoclient-84485fbc5c-zxlv8 -n default' to see all of the containers in this pod.
[root@echoclient-84485fbc5c-zxlv8 /]# /opt/jre/bin/java -cp /opt/echoclient-1.0-SNAPSHOT-jar-with-dependencies.jar hello.HttpSender "http://echoserver:8080/echo?name=peter" 10 40 0
using num threads: 10
Starting pool-1-thread-1 with numCalls=40 parallelSends=false delayBetweenCalls=0 url=http://echoserver:8080/echo?name=peter mixedRespTimes=false
Starting pool-1-thread-2 with numCalls=40 parallelSends=false delayBetweenCalls=0 url=http://echoserver:8080/echo?name=peter mixedRespTimes=false
Starting pool-1-thread-3 with numCalls=40 parallelSends=false delayBetweenCalls=0 url=http://echoserver:8080/echo?name=peter mixedRespTimes=false
Starting pool-1-thread-4 with numCalls=40 parallelSends=false delayBetweenCalls=0 url=http://echoserver:8080/echo?name=peter mixedRespTimes=false
Starting pool-1-thread-5 with numCalls=40 parallelSends=false delayBetweenCalls=0 url=http://echoserver:8080/echo?name=peter mixedRespTimes=false
Starting pool-1-thread-6 with numCalls=40 parallelSends=false delayBetweenCalls=0 url=http://echoserver:8080/echo?name=peter mixedRespTimes=false
Starting pool-1-thread-7 with numCalls=40 parallelSends=false delayBetweenCalls=0 url=http://echoserver:8080/echo?name=peter mixedRespTimes=false
Starting pool-1-thread-8 with numCalls=40 parallelSends=false delayBetweenCalls=0 url=http://echoserver:8080/echo?name=peter mixedRespTimes=false
Starting pool-1-thread-9 with numCalls=40 parallelSends=false delayBetweenCalls=0 url=http://echoserver:8080/echo?name=peter mixedRespTimes=false
Starting pool-1-thread-10 with numCalls=40 parallelSends=false delayBetweenCalls=0 url=http://echoserver:8080/echo?name=peter mixedRespTimes=false
pool-1-thread-7: successes=[40], failures=[0], duration=[481ms]
pool-1-thread-6: successes=[40], failures=[0], duration=[485ms]
pool-1-thread-4: successes=[40], failures=[0], duration=[504ms]
pool-1-thread-1: successes=[40], failures=[0], duration=[542ms]
pool-1-thread-9: successes=[40], failures=[0], duration=[626ms]
pool-1-thread-8: successes=[40], failures=[0], duration=[652ms]
pool-1-thread-2: successes=[40], failures=[0], duration=[684ms]
pool-1-thread-10: successes=[40], failures=[0], duration=[657ms]
pool-1-thread-5: successes=[40], failures=[0], duration=[678ms]
pool-1-thread-3: successes=[40], failures=[0], duration=[696ms]

检查从 echo 客户端到 echo 服务器的 HTTP 连接

来自 netstat 的连接信息：

[root@echoclient-84485fbc5c-zxlv8 /]# netstat -ano | grep 8080 | grep ESTABLISHED
tcp        0      0 10.244.2.79:58074       10.244.2.65:8080        ESTABLISHED off (0.00/0/0)
tcp        0      0 10.244.2.79:38076       10.244.2.73:8080        ESTABLISHED off (0.00/0/0)
tcp        0      0 10.244.2.79:58088       10.244.2.65:8080        ESTABLISHED off (0.00/0/0)
tcp        0      0 10.244.2.79:38080       10.244.2.73:8080        ESTABLISHED off (0.00/0/0)
tcp        0      0 10.244.2.79:58056       10.244.2.65:8080        ESTABLISHED off (0.00/0/0)
tcp        0      0 10.244.2.79:38094       10.244.2.73:8080        ESTABLISHED off (0.00/0/0)
tcp        0      0 10.244.2.79:38110       10.244.2.73:8080        ESTABLISHED off (0.00/0/0)
tcp        0      0 10.244.2.79:58076       10.244.2.65:8080        ESTABLISHED off (0.00/0/0)

来自特使集群的连接信息：

[root@echoclient-84485fbc5c-zxlv8 /]# curl -s http://localhost:15000/clusters | grep echoserver
outbound|8080||echoserver.default.svc.cluster.local::default_priority::max_connections::2
outbound|8080||echoserver.default.svc.cluster.local::default_priority::max_pending_requests::1024
outbound|8080||echoserver.default.svc.cluster.local::default_priority::max_requests::1024
outbound|8080||echoserver.default.svc.cluster.local::default_priority::max_retries::3
outbound|8080||echoserver.default.svc.cluster.local::high_priority::max_connections::1024
outbound|8080||echoserver.default.svc.cluster.local::high_priority::max_pending_requests::1024
outbound|8080||echoserver.default.svc.cluster.local::high_priority::max_requests::1024
outbound|8080||echoserver.default.svc.cluster.local::high_priority::max_retries::3
outbound|8080||echoserver.default.svc.cluster.local::added_via_api::true
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::cx_active::4
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::cx_connect_fail::0
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::cx_total::4
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::rq_active::0
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::rq_error::0
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::rq_success::200
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::rq_timeout::0
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::rq_total::200
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::health_flags::healthy
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::weight::1
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::region::
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::zone::
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::sub_zone::
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::canary::false
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.65:8080::success_rate::-1
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::cx_active::4
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::cx_connect_fail::0
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::cx_total::4
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::rq_active::0
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::rq_error::0
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::rq_success::200
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::rq_timeout::0
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::rq_total::200
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::health_flags::healthy
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::weight::1
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::region::
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::zone::
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::sub_zone::
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::canary::false
outbound|8080||echoserver.default.svc.cluster.local::10.244.2.73:8080::success_rate::-1

我们可以看到从 echoclient 到 echoserver(10.244.2.65,10.244.2.73)有8 个连接，但不是配置的 maxConnections 2。

为什么有 8 个连接而不是 2 个？是不是对 envoy 的 maxConnections 有什么误解？

score 2 · Accepted Answer

正如加勒特在评论中提到的那样：

max_connections或max_requests指每个childrenphp 池在重生之前可以进行的连接数，通常可以在/etc/etc/php/{version}/fpm/pool.d/www.conf max_children基于机器的内存中找到，一些有助于我理解的示例教程是Max Requests / Children

score 0 · Accepted Answer

我认为，您使用 1 个线程（即 1 个 HTTP 连接）调用了客户端，但它并行发送请求（默认为 10 个批次）

在特使中，max_connections适用于 http1 连接，在您的情况下，您只有一个 http 连接。

尝试查看max_request哪些可能适用于与 http2 密切相关的线程。

希望这可以帮助！！

score 0 · Accepted Answer

SSH 到客户端 pod 的节点，并执行到 istio-proxy 容器中

docker exec --privileged --user root -it <istio-proxy-container-id> bash

使用以下命令查找特使到上游的 tcp 连接

ss -pe | grep 8080 | grep envoy

istio - Envoy 中的“max_connections”到底是什么意思？

3 回答 3

Related

Reference