check queue is not nil before accessing

[C-Pro]

I have got a panic in production service.

panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x1 addr=0x0 pc=0x2157d16]

goroutine 26814443 [running]:
github.com/centrifugal/centrifuge-go.(*cbQueue).pushOrClose(0x0, 0xc0024a4cd8, 0x0)
	/home/runner/go/pkg/mod/github.com/centrifugal/[email protected]/queue.go:64 +0x36
github.com/centrifugal/centrifuge-go.(*cbQueue).push(...)
	/home/runner/go/pkg/mod/github.com/centrifugal/[email protected]/queue.go:55
github.com/centrifugal/centrifuge-go.(*Client).runHandlerSync(0xc001146780, 0xc0024a4cc0)
	/home/runner/go/pkg/mod/github.com/centrifugal/[email protected]/client.go:635 +0xbb
github.com/centrifugal/centrifuge-go.(*Subscription).handlePublication(0xc000814e00, 0xc004855440)
	/home/runner/go/pkg/mod/github.com/centrifugal/[email protected]/subscription.go:590 +0x107
github.com/centrifugal/centrifuge-go.(*Client).handlePush(0xc001146780, 0xc003f4d380)
	/home/runner/go/pkg/mod/github.com/centrifugal/[email protected]/client.go:719 +0x105
github.com/centrifugal/centrifuge-go.(*Client).handle(0xc001146780, 0xc000bd8f00)
	/home/runner/go/pkg/mod/github.com/centrifugal/[email protected]/client.go:682 +0x151
github.com/centrifugal/centrifuge-go.(*Client).readOnce(0xc001146780, {0x2e920e0?, 0xc0012b8cf0?})
	/home/runner/go/pkg/mod/github.com/centrifugal/[email protected]/client.go:618 +0xa9
github.com/centrifugal/centrifuge-go.(*Client).reader(0xc001146780, {0x2e920e0, 0xc0012b8cf0}, 0x1?)
	/home/runner/go/pkg/mod/github.com/centrifugal/[email protected]/client.go:625 +0x5f
created by github.com/centrifugal/centrifuge-go.(*Client).startReconnecting in goroutine 26814431
	/home/runner/go/pkg/mod/github.com/centrifugal/[email protected]/client.go:961 +0x669

I suspect the centrifuge connection was closed and Client.moveToClosed has set c.cbQueue to nil before Subscription.handlePublication callback was executed and tried to push to the queue.

So I am adding a check that queue is not nil before calling push. I suspect there should be a better way involving status field, but I am not sure what the lifecycle is supposed to be, so taking most straightforward approach.

[FZambia]

Thanks! I'll take a closer look soon.

Maybe it's possible to build a reproducer using integration tests which are using real Centrifugo instance now:

docker run -p 8000:8000 centrifugo/centrifugo:v5 centrifugo --client_insecure

I.e. publish to some channel and close the client. I think this should be visible when running a test under stress tool for example. If you will have time to try making it - would be nice, if not - I'll try myself as soon as I start looking.

[C-Pro]

Maybe it's possible to build a reproducer using integration tests which are using real Centrifugo instance now:

Added a test in a separate PR #95 to make it easier to check on both master and this branch.
For me it fails in first couple of runs on master and does not fail in the branch of this pr.
I am running multiple times to catch the race with high confidence:

go test -v -count=25 -failfast -timeout=1h -run TestConcurrentPublishSubscribeDisconnect .

[FZambia]

Many thanks, already started looking whether there is a better way to fix, if won't find a better approach – will proceed with your changes. Hopefully during next several days.

[FZambia]

Opened #96, it already includes #95 - it addresses the issue by not calling callbacks at all after close, waiting for currently queued callbacks to fire.