Vm auth: if the request does not fully cache the body to buf due to server connection disconnection, it will not be retried.

[shichanglin5]

Describe the bug

description in comment:

func (rtb *readTrackingBody) canRetry() bool {
   if rtb.cannotRetry {
   	return false
   }
// if the request does not fully cache the body to buf 
// due to server connection disconnection, it will not be retried.
   if len(rtb.buf) > 0 && !rtb.needReadBuf {
   	return false
   }
   return true
}

log

2024-06-06T20:30:37+08:00 2024-06-06T12:30:37.396Z	warn	VictoriaMetrics/app/vmauth/main.go:268	remoteAddr: "172.26.13.17:64104, X-Forwarded-For: 10.214.137.216"; requestURI: /api/v1/push?job=skyeye-skynet; cannot proxy the request to http://10-205-89-19.vminsert-skyeye.vm-cluster-skyeye.svc.cluster.local.:8480/insert/2:6/prometheus/api/v1/push?job=skyeye-skynet: readfrom tcp 10.205.89.16:42960->10.205.89.19:8480: write tcp 10.205.89.16:42960->10.205.89.19:8480: write: broken pipe

To Reproduce

1. do request with partial read body to buf
2. close client-server connection
3. log cannot proxy the request to xxx

Version

I built it with source code：
git revision: 1a1ee87

Logs

2024-06-06T20:30:37+08:00 2024-06-06T12:30:37.396Z warn VictoriaMetrics/app/vmauth/main.go:268 remoteAddr: "172.26.13.17:64104, X-Forwarded-For: 10.214.137.216"; requestURI: /api/v1/push?job=skyeye-skynet; cannot proxy the request to http://10-205-89-19.vminsert-skyeye.vm-cluster-skyeye.svc.cluster.local.:8480/insert/2:6/prometheus/api/v1/push?job=skyeye-skynet: readfrom tcp 10.205.89.16:42960->10.205.89.19:8480: write tcp 10.205.89.16:42960->10.205.89.19:8480: write: broken pipe

Screenshots

No response

Used command-line flags

No response

Additional information

No response

[shichanglin5]

In addition, I have observed that vm auth monitoring, go gc accounts for a relatively high proportion( 50%), and perhaps gc can be reduced by reusing buf to avoid frequent memory allocation.

[hagen1778]

#6533 has been merged and will be included into the next release.

[shichanglin5]

good!

[f41gh7]

We keep issues open until actual release. So re-opening it.