Intermittant crash

[achingbrain]

The following error occurs intermittently on Chrome and Firefox during IPFS CI:

ipfs:   1) interface-ipfs-core tests
ipfs:        .swarm.peers
ipfs:          should list peers only once even if they have multiple addresses:
ipfs:      AggregateError: 
ipfs:     Error: peer is not available
ipfs:         at Socket.<anonymous> (file:///home/travis/build/ipfs/js-ipfs/node_modules/libp2p-webrtc-star/src/index.js:175:67)
ipfs:         at Socket.Emitter.emit (file:///home/travis/build/ipfs/js-ipfs/node_modules/component-emitter/index.js:145:20)
ipfs:         at Socket.emitEvent (file:///home/travis/build/ipfs/js-ipfs/node_modules/socket.io-client-next/build/socket.js:264:20)
ipfs:         at Socket.onevent (file:///home/travis/build/ipfs/js-ipfs/node_modules/socket.io-client-next/build/socket.js:251:18)
ipfs:         at Socket.onpacket (file:///home/travis/build/ipfs/js-ipfs/node_modules/socket.io-client-next/build/socket.js:215:22)
ipfs:         at Manager.Emitter.emit (file:///home/travis/build/ipfs/js-ipfs/node_modules/component-emitter/index.js:145:20)
ipfs:         at Manager.ondecoded (file:///home/travis/build/ipfs/js-ipfs/node_modules/socket.io-client-next/build/manager.js:204:15)
ipfs:         at Decoder.Emitter.emit (file:///home/travis/build/ipfs/js-ipfs/node_modules/component-emitter/index.js:145:20)
ipfs:         at Decoder.add (file:///home/travis/build/ipfs/js-ipfs/node_modules/socket.io-client-next/node_modules/socket.io-parser/dist/index.js:117:23)
ipfs:         at Manager.ondata (file:///home/travis/build/ipfs/js-ipfs/node_modules/socket.io-client-next/build/manager.js:196:22)
ipfs:       at maybeSettle (file:///home/travis/build/ipfs/js-ipfs/node_modules/p-some/index.js:31:11)
ipfs:       at file:///home/travis/build/ipfs/js-ipfs/node_modules/p-some/index.js:69:23

The test is this one and the error message seems unrelated to the test code.

[hannahhoward]

Hi! Want to report my findings here:

I looked closely at the test here: https://github.com/ipfs/js-ipfs/blob/49f78807d7e26483bd926b45cc7e0f797d77e41b/packages/interface-ipfs-core/src/swarm/peers.js#L109-L139

I believe our answer may live somewhere in the todo referenced in these lines: https://github.com/ipfs/js-ipfs/blob/49f78807d7e26483bd926b45cc7e0f797d77e41b/packages/interface-ipfs-core/src/swarm/peers.js#L129-L131

I also see the error in question comes from here https://github.com/libp2p/js-libp2p-webrtc-star/blob/master/src/sig-server/routes-ws/next.js#L118 and I believe our problem is as follows:

• the test has two signalling servers
• the dialing peer is connected to signalling server a
• the dialed perer is connected to server a and to server b
• however, the it is listed on both server a and server b through it's address on server b -- this is cause the webrtc-star transport only maintains a single address, so the current one is over written here: https://github.com/libp2p/js-libp2p-webrtc-star/blob/master/src/listener.js#L38
• however, the registered address on the signaling server is set here: https://github.com/libp2p/js-libp2p-webrtc-star/blob/master/src/listener.js#L140 -- I can see a a potential race where signal server a's connect happens before the signally address is over written to the server b address. In this case, when the dialing peer attempts to connect at the address on server B, its not in the signalling servers list of peers.
• hence this error

Wow that's a mouthful. I have to admit -- this test feels very weird to me in this context. If the WS-Star transport only supports one address, we shouldn't be registering multiple and hoping the test works right? Maybe we ought to do the work to track which registered addresses are on which signalling servers?