Panic in IpfsDHT

[sidenaio]

Version information:

github.com/ipfs/go-ipfs v0.5.1-0.20200506014919-3339ce3b0e95

Description:

panic: send on closed channel

goroutine 2038717 [running]:
github.com/libp2p/go-libp2p-kad-dht.(*IpfsDHT).findProvidersAsyncRoutine.func1(0x21ed620, 0xc00b821000, 0xc0002a96e0, 0x22, 0xc00b2507b0, 0xb765d6, 0x1fd1540, 0xc000b141d8, 0x0)
        /home/mikhail/go/pkg/mod/github.com/libp2p/[email protected]/routing.go:582 +0x6d1
github.com/libp2p/go-libp2p-kad-dht.(*IpfsDHT).runLookupWithFollow

[aschmahmann]

@sidenaio thanks for the bug report. It looks like there's a cleanup issue when the FindProviders lookup is aborted early. Can you give any other context on what you were doing when you encountered the error?

[sidenaio]

We are using ipfs as a library in our project. We ran a test with 1k nodes. One node went down with this panic. We use ipfs for unixfs.add, unixfs.get, pin.add, pin.rm. Before the panic, our log contains read operation from ipfs.

[Stebalien]

This will happen when the context is canceled. The issue is that we:

1. Run queryFn in a goroutine as the "followup".
2. Abort early of the context is canceled or the "stop" condition is triggered.

Instead, we should just cancel the query and wait till all the query routines return.

https://github.com/libp2p/go-libp2p-kad-dht/blob/08ab423f2fc5f3c3e3e9452fc2358dbc0610d369/query.go#L131

[Stebalien]

Fixed in #7382.