Svcwatcher core after losing master/leader

[TothFerenc]

Is this a BUG REPORT or FEATURE REQUEST?: bug

What happened: Svcwatcher Pod lost master for any reason, so the process was exiting:

E0704 17:44:12.997128       1 svcwatcher.go:93] Lost master
F0704 17:44:12.997152       1 svcwatcher.go:97] Lost lease
E0704 17:44:12.997232       1 event.go:269] Unable to write event: 'can't create an event with namespace 'default' in namespace 'kube-system'' (may retry after sleeping)
goroutine 1 [running]:
github.com/golang/glog.stacks(0xc000374400, 0xc00028e000, 0x3b, 0x9e)
        /go/src/github.com/nokia/danm/vendor/github.com/golang/glog/glog.go:769 +0xb8
github.com/golang/glog.(*loggingT).output(0x20605c0, 0xc000000003, 0xc00023c0e0, 0x1fce7c9, 0xd, 0x61, 0x0)
        /go/src/github.com/nokia/danm/vendor/github.com/golang/glog/glog.go:720 +0x372
github.com/golang/glog.(*loggingT).println(0x20605c0, 0xc000000003, 0xc00002feb0, 0x1, 0x1)
        /go/src/github.com/nokia/danm/vendor/github.com/golang/glog/glog.go:633 +0xe7
github.com/golang/glog.Fatalln(...)
        /go/src/github.com/nokia/danm/vendor/github.com/golang/glog/glog.go:1141
main.main()
        /go/src/github.com/nokia/danm/cmd/svcwatcher/svcwatcher.go:97 +0x9e4
E0704 17:44:20.320002       1 event.go:269] Unable to write event: 'can't create an event with namespace 'default' in namespace 'kube-system'' (may retry after sleeping)
glog: Flush took longer than 10s

What you expected to happen: No core dump before exit.

How to reproduce it: It happens frequently during deployment.

Anything else we need to know?:

Environment:

• DANM version (use danm -version):

# /opt/cni/bin/danm -version
2020/07/22 12:31:18 DANM binary was built from release: v4.2.0-0
2020/07/22 12:31:18 DANM binary was built from commit: c0a4c1570845556cf911a46df475c45a85941bb2

• Kubernetes version (use kubectl version):

# kubectl version
Client Version: version.Info{Major:"1", Minor:"18", GitVersion:"v1.18.4", GitCommit:"c96aede7b5205121079932896c4ad89bb93260af", GitTreeState:"clean", BuildDate:"2020-06-17T11:41:22Z", GoVersion:"go1.13.9", Compiler:"gc", Platform:"linux/amd64"}
Server Version: version.Info{Major:"1", Minor:"18", GitVersion:"v1.18.4", GitCommit:"c96aede7b5205121079932896c4ad89bb93260af", GitTreeState:"clean", BuildDate:"2020-06-17T11:33:59Z", GoVersion:"go1.13.9", Compiler:"gc", Platform:"linux/amd64"}

[Levovar]

so, this is the 97th line where it cores: https://github.com/nokia/danm/blob/master/cmd/svcwatcher/svcwatcher.go#L97 It is literally a library call without references to any objects I think I have already stated earlier that glog is shite :) maybe the non-newline API wouldn't core, but I absolutely refuse to deep dive into its code. solution is removing the usage of the whole library

the cannot create event remark above is more interesting for me

[Levovar]

reg the Eventing issue: the leader election library creates an event recorder without a namespace defined, so it defaults to default but our component runs in the kube-system, so when we really want to record an event it fails something like: https://github.com/tsuru/remesher/pull/5

which is funny because as far as I can tell the Events are raised using the meta of the provided EndPointsLock: https://github.com/kubernetes/client-go/blob/00dbcca6ee44c678754d3f5fda1bd0e704b26fe2/tools/leaderelection/resourcelock/endpointslock.go#L100, and lo and behold we do set the proper namespace into the lock: https://github.com/nokia/danm/blob/master/cmd/svcwatcher/svcwatcher.go#L74

soo...

[Levovar]

I guess others also have issues with the library :) https://bugzilla.redhat.com/show_bug.cgi?id=1842002

[Levovar]

@TothFerenc any comments on above? I'm kind of on the opinion that this is how stuff works, and we just need to live with it

[TothFerenc]

Maybe we can create a new TODO issue about log module harmonization (use the same logging engine across all DANM components), and this issue can depend on it. Of couse I will close this issue once client libraries are fixed in the meantime.