GitLab

Signed-off-by: gukq <gukaiqiang@gmail.com>

fixes #419

Also call FullClose in a goroutine; no need to block this.

(not happy with that but I'm starting to think we need to rethink stream
closing, again...)

(avoids the race condition)

This reverts commit aca83b9b, reversing
changes made to 86b8929d.

This was is not the correct fix. We already expose these addresses via the
host's AllAddrs method. The real problem is probably that we just don't ever
tell anyone about them (unless we disconnect and reconnect to our nearby DHT
nodes).

We need an address gossip protocol.

and regenerate protobuf files

related to #383

I won't call this *fixed* yet but it should help.

Currently, we leak a stream. We could use FullClose but, IMO, it's not worth it.
We might as well just reset and walk away.

This happens all the time in tests where we intentionally use fake keys for
performance.

Anyways, users probably don't want their logs spammed with errors they can't do
anything about.

(counting open connections requires copying)

Also, make the libp2p constructor fully useful. There should now be no need to
manually construct a swarm/host.

It is source of most warning noise in Warning channel and we would like
to enable warning channel some time in future by default

License: MIT
Signed-off-by: Hector Sanjuan <hector@protocol.ai>

Closes #256

This way, we actually process disconnect notifications (and reduce the lifetime
on peer addr records).

This commit prevents us from repeatedly extending the lifetimes of all observed
addresses if a peer keeps on reconnecting.

It also fixes two race conditions:

1. We may end up processing a disconnect after a re-connect and may
accidentally give the addresses associated with that peer a
RecentlyConnectedAddrTTL instead of a ConnectedAddrTTL.

2. We may end up processing a connect after the associated disconnect storing
the associated peer addresses indefinitely.

There's really no reason to *only* do this if we have relays enabled. Doing it
this way makes `go vet` happier and reduces conditional logic.

Before, on close, we:

1. Weren't completing the write.
2. Flushing the buffer without waiting the latency delay.

This fixes that by using two separate channels for close/reset and ignoring the
close channel in deliverOrWait.

* Fix the tests to work with separate reset/close methods.
* Ensure we interrupt writes on reset.
* Always delay the proper time even if we're sending short messages.
* Copy buffers as we send them. `Write` is not allowed to hang onto buffers. If
  we run into performance issues, we can always add a buffer pool.