Tap from web sometimes causes tap server panic #1504
Description
I've noticed that clicking the "Stop" button on the /tap and /top pages is usually laggy and often results in displaying this banner:
It looks like occasionally that correlates with an actual panic in the tap process in the controller pod:
time="2018-08-22T05:57:00Z" level=error msg="rpc error: code = Canceled desc = context canceled"
panic: send on closed channel
goroutine 139 [running]:
github.com/linkerd/linkerd2/controller/tap.(*server).tapProxy(0xc420442c20, 0x14843a0, 0xc420966fc0, 0x3f800000, 0xc420981510, 0xc4205fe740, 0xa, 0xc420a0c180)
/go/src/github.com/linkerd/linkerd2/controller/tap/server.go:278 +0x3b6
created by github.com/linkerd/linkerd2/controller/tap.(*server).TapByResource
/go/src/github.com/linkerd/linkerd2/controller/tap/server.go:94 +0x50b
This makes me think that the web server endpoint isn't properly handling websocket close, and is leaving its connection to the tap server in a bad state.