Reasoning behind this panic

[anacrolix]

Reasoning behind this panic

[anacrolix]

Thanks for the report. Good to hear you're having luck with the client. Are you loading 10k+ torrents into a single Client? Could you estimate how many simultaneous UTP connections you have?

At a quick glance, I think this panic would occur if you received a SYN packet, that would result in a connection with a receive ID that matches the connection the initiator is trying to establish, but was initiated locally, and accepted locally. If you put log.Println(c.send_id, syn.conn_id) before the panic, you should find that send_id is 2 higher than conn_id if my theory is correct. It's probable that you're initiating an awful lot of connections between 2 clients to trigger this condition. Let me know the results of this, if it's not too much hassle to try and trigger it. I'll go into more detail if the data fits my prediction.

The other message you refer to is related, assuming the aforementioned theory. A fix for this will kill both birds.

[Emyrk]

Currently the largest test I have loads up to 25k torrents into a single client (should I launch more than 1 client to handle this load?). The number I am at when I get the panic is anywhere from 14-25K. As for simultaneous UTP connections, I am unsure. Is there a way I could estimate that, or see the current number of connections?

My test network setup is a docker network over weave, and I control the torrent tracker and all nodes, so everything is internal with no external nodes. I create the torrents internally too. I usually run 3-4 nodes with 1 having all the torrents initially. If that helps with any estimates, but I am unsure how I would be able to make that guess.

To your statement on launching a lot of connections between 2 clients, that is true. I have 25,000 torrents, and only 4 nodes, and most of the time only 1 node has the given torrent.

I'll put in the log statement and report back when I reproduce and get the log printout.

[Emyrk]

As you suspected:
log.Println(c.send_id, syn.conn_id)
Produces:
43150 43148

43150 43148
panic: :|
	panic: sync: Unlock of unlocked RWMutex

goroutine 28 [running]:
panic(0xbdc980, 0xc44a519eb0)
	/usr/local/go/src/runtime/panic.go:500 +0x1a1
sync.(*RWMutex).Unlock(0xc42020ef30)
	/usr/local/go/src/sync/rwmutex.go:118 +0xef
github.com/FactomProject/factomd-torrent/vendor/github.com/anacrolix/sync.(*RWMutex).Unlock(0xc42020ee00)
	/home/steven/go/src/github.com/FactomProject/factomd-torrent/vendor/github.com/anacrolix/sync/rwmutex.go:26 +0x59
panic(0xbdc980, 0xc44a519ea0)
	/usr/local/go/src/runtime/panic.go:458 +0x243
github.com/FactomProject/factomd-torrent/vendor/github.com/anacrolix/utp.(*Socket).ackSyn(0xc42019a540, 0xa88c0001, 0xc4a4767200, 0x10, 0x0, 0x0)
	/home/steven/go/src/github.com/FactomProject/factomd-torrent/vendor/github.com/anacrolix/utp/socket.go:485 +0x38f
github.com/FactomProject/factomd-torrent/vendor/github.com/anacrolix/utp.(*Socket).Accept(0xc42019a540, 0x0, 0x0, 0x0, 0x0)
	/home/steven/go/src/github.com/FactomProject/factomd-torrent/vendor/github.com/anacrolix/utp/socket.go:507 +0xd8
github.com/FactomProject/factomd-torrent/vendor/github.com/Anacrolix/torrent.(*Client).acceptConnections(0xc42020ee00, 0x13e4e40, 0xc42019a540, 0x13e5401)
	/home/steven/go/src/github.com/FactomProject/factomd-torrent/vendor/github.com/Anacrolix/torrent/client.go:383 +0xb8
created by github.com/FactomProject/factomd-torrent/vendor/github.com/Anacrolix/torrent.NewClient
	/home/steven/go/src/github.com/FactomProject/factomd-torrent/vendor/github.com/Anacrolix/torrent/client.go:290 +0x67d

[anacrolix]

Thanks for the feedback. I'll write a unit test and a fix when I get time. In the meanwhile, you can just comment out that panic on your servers, it will behave correctly.

[Emyrk]

Thanks!

[anacrolix]

Oops, I think this was fixed in 78c7e0d.