diff options
| author | David Fifield <david@bamsoftware.com> | 2020-02-21 14:27:54 -0700 |
|---|---|---|
| committer | David Fifield <david@bamsoftware.com> | 2020-02-21 20:36:07 -0700 |
| commit | a05f5efc7966bcbf71d0a9410d4656a221bcb1bb (patch) | |
| tree | 02a52051b715f3db86bb156ae448bdf5f97acb53 | |
| parent | 6b902fcad0437d2d92696a1ccc7801091d3ef667 (diff) | |
Set the smux KeepAliveTimeout (idle timeout) to 10 minutes.
Analogously to the quic IdleTimeout.
The default KeepAliveTimeout is otherwise 30 seconds.
https://github.com/xtaci/smux/blob/c6969d8a76874342611f4b544d05b1590f1a76b1/mux.go#L45
The too-short idle timeout was causing frequent hangs in the client,
with a particularly bad effect. In copyLoop, the downstream socks←webRTC
copy would terminate after 30 seconds of idleness (could be as soon as
the first 30 seconds, if you get a bad proxy right off the bat), but the
upstream webRTC←socks copy would keep running for about 2 minutes (may
be the tor SocksTimeout, I'm not sure), with no opportunity to start
working in that time because the downstream copy had already exited.
| -rw-r--r-- | server/server.go | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/server/server.go b/server/server.go index e67d6c9..98dc18e 100644 --- a/server/server.go +++ b/server/server.go @@ -311,6 +311,7 @@ func acceptStreamsKCP(conn *kcp.UDPSession) error { smuxConfig := smux.DefaultConfig() smuxConfig.Version = 2 + smuxConfig.KeepAliveTimeout = 10 * time.Minute sess, err := smux.Server(conn, smuxConfig) if err != nil { return err |