Error: write EPIPE

[achingbrain]

This sometimes happens during CI for js-IPFS, usually after a test has passed and it's tearing down the nodes in an after or afterEach block.

ipfs: [stdout] Error: write EPIPE
ipfs:     at afterWriteDispatched (internal/stream_base_commons.js:154:25)
ipfs:     at writeGeneric (internal/stream_base_commons.js:145:3)
ipfs:     at Socket._writeGeneric (net.js:784:11)
ipfs:     at Socket._write (net.js:796:8)
ipfs:     at doWrite (_stream_writable.js:403:12)
ipfs:     at writeOrBuffer (_stream_writable.js:387:5)
ipfs:     at Socket.Writable.write (_stream_writable.js:318:11)
ipfs:     at /Users/travis/build/ipfs/js-ipfs/node_modules/stream-to-it/sink.js:90:21
ipfs:     at processTicksAndRejections (internal/process/task_queues.js:97:5)
ipfs:     at async sink (/Users/travis/build/ipfs/js-ipfs/node_modules/libp2p-tcp/src/socket-to-conn.js:31:9) {
ipfs:   errno: 'EPIPE',
ipfs:   code: 'ERR_UNSUPPORTED_PROTOCOL',
ipfs:   syscall: 'write'
ipfs: }

This is with [email protected].

[jacobheun]

Is this causing CI to fail when shutting down? The remote end is likely terminating while we're trying to write but we should be catching the error.

[achingbrain]

Yes, it causes CI failures.

[achingbrain]

Any idea where this should be handled? I'm seeing this maybe 1 job in 10, and considering each build has 18 jobs that means I see it pretty often.

[jacobheun]

Bumping this to a P0 and looking into it as it's currently a huge pain in CI

[jacobheun]

Based on the added code it looks like the error is coming from the newStream function at https://github.com/libp2p/js-libp2p/blob/v0.27.8/src/upgrader.js#L256. This would indicate that something is calling newStream and during protocol negotiation the socket is closed by the other side.

This may not be an issue with TCP itself but application/library code not catching errors when calling dialProtocol or connection.newStream. Looking into where this is being thrown in IPFS.

[jacobheun]

@achingbrain I added a patch in [email protected] that should help catch errors like this. Can you let me know if you continue to see this in CI after upgrading?

[dapplion]

Seeing the same issue recurrently in Lodestar CI and local tests

[achingbrain]

I haven't seen this for at least a year so I'm going to close this. @dapplion please shout if this is still causing problems.