Pod 0->1, Delayed HTTP response

[Firlng]

When a Pod automatically scales from 0 to 1, why doesn't the HTTP response to the UI request get forwarded promptly after the Pod successfully starts, causing slow HTTP responses? Would the following parameters have any impact?
【
- name: KEDA_HTTP_CONNECT_TIMEOUT
value: 10000ms
- name: KEDA_HTTP_KEEP_ALIVE
value: 8s
- name: KEDA_RESPONSE_HEADER_TIMEOUT
value: 20000ms
- name: KEDA_CONDITION_WAIT_TIMEOUT
value: 8s
- name: KEDA_HTTP_ENDPOINTS_CACHE_POLLING_INTERVAL_MS
value: "50"
- name: KEDA_HTTP_FORCE_HTTP2
value: "false"
- name: KEDA_HTTP_MAX_IDLE_CONNS
value: "50"
- name: KEDA_HTTP_IDLE_CONN_TIMEOUT
value: 5s
- name: KEDA_HTTP_TLS_HANDSHAKE_TIMEOUT
value: 5s
- name: KEDA_HTTP_EXPECT_CONTINUE_TIMEOUT
value: 5s
】
This is the HTTP response time for the Sample xkcd pod when scaling from 0→1. After excluding the pod startup time of 4s, the HTTP Add-On forwarding time consumes about 15s. Can you tell me the specific reasons for the slow forwarding?
[
PS C:> # Measure the total time for the entire request process
PS C:> $startTime = Get-Date
PS C:> $response = Invoke-WebRequest -UseBasicParsing -Headers @{Host="myhost.com"} -Uri "http://xxxx.xx.xx.xx:8181/path1"
PS C:> $endTime = Get-Date
PS C:> $elapsedTime = ($endTime - $startTime).TotalMilliseconds
PS C:> Write-Host "Total request to response time: ${elapsedTime} milliseconds"
Total request to response time: 20302.7963 milliseconds
PS C:> Write-Host "Status code: $($response.StatusCode)"
Status code: 200
]

[linkvt]

Hi @Firlng .

I just tried accessing the xkcd app locally using 0.11.1 of the http addon (default helm deployment without any values set) with the following result:

# first request (cold start)
❯ curl http://10.89.0.30/path2 -H "Host: myhost.com" -w 'Total: %{time_total}s\n' -o /dev/null -s
Total: 3.410926s

# second request
❯ curl http://10.89.0.30/path2 -H "Host: myhost.com" -w 'Total: %{time_total}s\n' -o /dev/null -s
Total: 0.008262s

I think this might be an issue with your environment, maybe your application starts slowly or your readinessProbe is not configured in an optimal way.

[Firlng]

Hi @linkvt
In the file http-add-on-0.11.1\interceptor\config\timeouts.go:
[
Connect time.Duration envconfig:"KEDA_HTTP_CONNECT_TIMEOUT" default:"500ms"
]
In the file http-add-on-0.11.1\interceptor\main.go:
[
dialer := kedanet.NewNetDialer(timeouts.Connect, timeouts.KeepAlive)
]
My parameter is ：

• name: KEDA_HTTP_CONNECT_TIMEOUT
  value: 10000ms

If the timeouts.Connect value is too large, it will affect the connection.
If the timeouts.Connect value is 500ms，it will be OK.

timeouts.Connect controls the maximum wait time for TCP connections to prevent being stuck for extended periods due to network problems.

I think the timeout duration should not impact normal connection times. What are the reasons?

[linkvt]

Oh now I understand, you're absolutely right 👍

I was able to reproduce the problem and created a bug report: #1385

[Firlng]

Hi @linkvt

Thank you so much for your feedback and for creating the bug report!
Thanks again!