指标类型设置存在问题

[JasperWey]

kubeskoop/pkg/exporter/probe/legacy.go

Lines 129 to 130 in 72ea289

	metrics <- prometheus.MustNewConstMetric(desc, prometheus.GaugeValue, float64(value), labelValues...)
	}

这里指定使用Gauge类型，导致所有的指标都是此类型，但是网络丢包统计最准确的是Counter类型吧？

- # TYPE kubeskoop_tcp_passiveopens gauge
- # TYPE kubeskoop_tcp_retranssegs gauge
- # TYPE kubeskoop_tcpext_delayedacklost gauge
- # TYPE kubeskoop_tcpext_embryonicrsts gauge
- # TYPE kubeskoop_tcpext_listendrops gauge
- # TYPE kubeskoop_tcpext_listenoverflows gauge
- # TYPE kubeskoop_tcpext_pawsactive gauge
- # TYPE kubeskoop_tcpext_pawsestab gauge
- # TYPE kubeskoop_tcpext_pfmemallocdrop gauge
- # TYPE kubeskoop_tcpext_syncookiesfailed gauge
- # TYPE kubeskoop_tcpext_syncookiesrecv gauge
- # TYPE kubeskoop_tcpext_syncookiessent gauge
- # TYPE kubeskoop_tcpext_tcpabortfailed gauge
- # TYPE kubeskoop_tcpext_tcpabortonclose gauge
- # TYPE kubeskoop_tcpext_tcpabortondata gauge
- # TYPE kubeskoop_tcpext_tcpabortonlinger gauge
- # TYPE kubeskoop_tcpext_tcpabortonmemory gauge
- # TYPE kubeskoop_tcpext_tcpabortontimeout gauge
- # TYPE kubeskoop_tcpext_tcpackskippedchallenge gauge
- # TYPE kubeskoop_tcpext_tcpackskippedfinwait2 gauge
- # TYPE kubeskoop_tcpext_tcpackskippedpaws gauge
- # TYPE kubeskoop_tcpext_tcpackskippedseq gauge
- # TYPE kubeskoop_tcpext_tcpackskippedsynrecv gauge
- # TYPE kubeskoop_tcpext_tcpackskippedtimewait gauge
- # TYPE kubeskoop_tcpext_tcpbacklogdrop gauge
- # TYPE kubeskoop_tcpext_tcpfastopenactivefail gauge
- # TYPE kubeskoop_tcpext_tcpfastretrans gauge
- # TYPE kubeskoop_tcpext_tcpkeepalive gauge
- # TYPE kubeskoop_tcpext_tcplossfailures gauge
- # TYPE kubeskoop_tcpext_tcplossproberecovery gauge
- # TYPE kubeskoop_tcpext_tcplossprobes gauge
- # TYPE kubeskoop_tcpext_tcplostretransmit gauge
- # TYPE kubeskoop_tcpext_tcpmemorypressures gauge
- # TYPE kubeskoop_tcpext_tcpmemorypressureschrono gauge
- # TYPE kubeskoop_tcpext_tcpmtupfail gauge
- # TYPE kubeskoop_tcpext_tcpmtupsuccess gauge
- # TYPE kubeskoop_tcpext_tcprcvcollapsed gauge
- # TYPE kubeskoop_tcpext_tcprcvqdrop gauge
- # TYPE kubeskoop_tcpext_tcpretransfail gauge
- # TYPE kubeskoop_tcpext_tcpslowstartretrans gauge
- # TYPE kubeskoop_tcpext_tcpsynretrans gauge
- # TYPE kubeskoop_tcpext_tcptimeouts gauge
- # TYPE kubeskoop_tcpext_tcpwinprobe gauge
- # TYPE kubeskoop_tcpext_tcpwqueuetoobig gauge
- # TYPE kubeskoop_tcpext_tcpzerowindowdrop gauge
- # TYPE kubeskoop_udp_ignoredmulti gauge
- # TYPE kubeskoop_udp_incsumerrors gauge
- # TYPE kubeskoop_udp_indatagrams gauge
- # TYPE kubeskoop_udp_inerrors gauge
- # TYPE kubeskoop_udp_memerrors gauge
- # TYPE kubeskoop_udp_noports gauge
- # TYPE kubeskoop_udp_outdatagrams gauge
- # TYPE kubeskoop_udp_rcvbuferrors gauge
- # TYPE kubeskoop_udp_sndbuferrors gauge

[Lyt99]

之前的评论似乎没发出来，今天才发现，不好意思。感谢您对KubeSkoop项目的关注！

您在issue里提到的指标都是直接从系统内核中直接透出的指标，是KubeSkoop直接从内核中透传出来的。如果使用counter类型，还需要在KubeSkoop中额外维护对应指标的增量。并且在如被采集的Pod重启或者重建等情况下，指标会清零，从而更加难以做指标的计算。

以上是KubeSkoop实现时使用gauge类型而非counter类型的理由。但是在实际指标透出时，可以考虑将其改为counter类型，便于查询和对指标作用的理解。这个我会在最近看一下。

再次感谢您对KubeSkoop项目的支持！