opencon thread massively slows down client shutdown? #2954

This is with latest 0.8.4 RC on Win7 x64, I have no idea yet what causes this, but I'm sure shutdown with previous versions was faster. This was with proxy enabled, problem seems to be non existant with proxy disabled. @sipa Any idea for such a behaviour? Any timeout blocking that thread?

Proxy enabled (opencon takes 1 1/2 minutes to interrupt):

<pre> 2013-08-30 10:25:25 addcon thread interrupt 2013-08-30 10:25:25 net thread interrupt 2013-08-30 10:25:25 dumpaddr thread stop 2013-08-30 10:25:25 msghand thread interrupt 2013-08-30 10:27:04 opencon thread interrupt 2013-08-30 10:27:04 Flush(false) 2013-08-30 10:27:04 wallet.dat refcount=0 2013-08-30 10:27:04 wallet.dat checkpoint 2013-08-30 10:27:04 wallet.dat detach 2013-08-30 10:27:04 wallet.dat closed 2013-08-30 10:27:04 DBFlush(false) ended 47ms 2013-08-30 10:27:04 StopNode() 2013-08-30 10:27:04 Flushed 14694 addresses to peers.dat 47ms 2013-08-30 10:27:04 Committing 30072 changed transactions to coin database... 2013-08-30 10:27:05 Flush(true) 2013-08-30 10:27:05 wallet.dat refcount=0 2013-08-30 10:27:05 wallet.dat checkpoint 2013-08-30 10:27:05 wallet.dat detach 2013-08-30 10:27:05 wallet.dat closed 2013-08-30 10:27:05 DBFlush(true) ended 16ms </pre>

Proxy disabled (opencon immediately interrupted): <per> 2013-08-30 10:31:26 addcon thread interrupt 2013-08-30 10:31:26 dumpaddr thread stop 2013-08-30 10:31:26 net thread interrupt 2013-08-30 10:31:26 msghand thread interrupt 2013-08-30 10:31:29 connection timeout 2013-08-30 10:31:29 opencon thread interrupt 2013-08-30 10:31:29 Flush(false) 2013-08-30 10:31:29 wallet.dat refcount=0 2013-08-30 10:31:29 wallet.dat checkpoint 2013-08-30 10:31:29 wallet.dat detach 2013-08-30 10:31:30 wallet.dat closed 2013-08-30 10:31:30 DBFlush(false) ended 42ms 2013-08-30 10:31:30 StopNode() 2013-08-30 10:31:30 Flushed 14542 addresses to peers.dat 68ms 2013-08-30 10:31:30 Committing 542 changed transactions to coin database... 2013-08-30 10:31:30 Flush(true) 2013-08-30 10:31:30 wallet.dat refcount=0 2013-08-30 10:31:30 wallet.dat checkpoint 2013-08-30 10:31:30 wallet.dat detach 2013-08-30 10:31:30 wallet.dat closed 2013-08-30 10:31:30 DBFlush(true) ended 20ms </pre>

Sometimes shutdown time is normal, sometimes it's slow as hell :-/.

I'm seeing this behaviour with a latent proxy too. Minutes for the threads to terminate.

@63 What OS are you using? Any special configuration?

@Diapolo Last build from github, OSX, proxying through Tor, not listening.

Can any of the core devs take a look please? @gmaxwell @gavinandresen @jgarzik

Sad that most core-devs seem to currently do fancy things and have no time for such bug reports...

"Shutdown is slow when I'm using a proxy" is not high on the priority list of things to fix.

Sometimes I whish, we would prefer fixing things before inventing new ones... I mean there are quite some unanswered bug reports here, which are old. I will report them in the future, but currently it's not motivating. At least IMHO all bugs should get reviewed and tagged!

Fwiw, a quick backtrace shows this waiting on the recv() in Socks5().

I've recently seen this problem as well. It appears that threads waiting for the proxy to reply can hang shutdown. Next time I encounter this issue I'll investigate further.

I also intent to debug this, when I've built a Qt debug build :).

@laanwj Any progress here? My guess is it is related to our SOCKS5 code (as @theuni said above) or perhaps even it just happens when using Tor. It seems Tor is trying at least 121 secs, before telling our client (Bitcoin Core) it can't get a connection Jun 12 15:41:27.477 [Hinweis] Tried for 121 seconds to get a connection to 83.227.173.83:8333. Giving up.... which could lead to the delay during shutdown.