The International Simutrans Forum

Development => Bug Reports => Topic started by: Vonjo on July 10, 2011, 07:30:12 PM

Title: Re: deManglers test Simutrans-Experimental Server
Post by: Vonjo on July 10, 2011, 07:30:12 PM
(I am not sure where to write it. But I think it is still related. It is not only Experimental but also standard server.)

I often play behind an always-on NAT. Sometimes when 'transfreing game', the internet connection gets crowded and the download freezes/stalls at some point. Then I kill simutrans.

At this point, maybe the server think that I am still downloading and waits for response from me. The server pauses until I finished download, which will not happen because I killed simutrans. The server thinks it still connected to me, which in reality just the NAT server.

Then the server waits for looong time...

The server runs again after a few minutes, sometimes hours, or sometimes even the server needs to be restarted manually. Timothy once needed to restart the server because of this. Never happened with deMangler's so far as the savegame is usually still small.

Maybe we need something like timeout...
Title: Re: Causes of server crashes
Post by: jamespetts on July 10, 2011, 07:40:50 PM
I have split this from the discussion in the Experimental board, as this also affects Standard and merits consideration of a fix there.
Title: Re: deManglers test Simutrans-Experimental Server
Post by: prissi on July 10, 2011, 08:26:26 PM
Usually, when the connection drops, the server will give up. This should happen quite fast, since then the router does not respond any more. Usually this takes less than 30s on most OS.
Title: Re: Re: deManglers test Simutrans-Experimental Server
Post by: jamespetts on July 10, 2011, 09:42:38 PM
Prissi,

see the thread here (http://forum.simutrans.com/index.php?topic=7602.new#new) that I split from this discussion about a reproducible case (on both Standard and Experimental) in which that does not happen.
Title: Re: deManglers test Simutrans-Experimental Server
Post by: prissi on July 11, 2011, 09:09:14 AM
Apart from the wrong link: In the original thread with same title nothing reproducable was reported.
Title: Re: Re: deManglers test Simutrans-Experimental Server
Post by: jamespetts on July 11, 2011, 09:18:38 AM
Ahh - I thought that he had found that when there was a timeout during connexion, the server would be in a running but frozen state? I suppose that his particular network issues are hard to recreate, but could it be reproduced, I wonder, by pulling the Ethernet cable from a client on connexion?
Title: Re: Re: deManglers test Simutrans-Experimental Server
Post by: Vonjo on July 11, 2011, 10:25:22 AM
10:22 UTC i made Timothy's brit server down by unplugging the modem while transfering.
UPDATE: 10:32 still down.
UPDATE: 10:38 down.
UPDATE: 10:40 up again.
The server waited for more than 15 minutes not 30 sec.

UPDATE: 10:52 tested again with Timothy's brit server at 10:52 by unplugging the modem while transfering.
UPDATE: 11:08 still down
UPDATE: 11:15 down
UPDATE: 11:31 down
UPDATE: 12:08 UTC still down for more than an hour.

I also made 96comic down from 10:45-11:02
Title: Re: Re: deManglers test Simutrans-Experimental Server
Post by: Vonjo on July 11, 2011, 11:38:02 AM
Anyone can confirm it?
Title: Re: deManglers test Simutrans-Experimental Server
Post by: prissi on July 11, 2011, 12:39:36 PM
What OS Timothy's server runs on?
Title: Re: Re: deManglers test Simutrans-Experimental Server
Post by: Vonjo on July 11, 2011, 01:06:08 PM
Quote from: prissi on July 11, 2011, 12:39:36 PM
What OS Timothy's server runs on?
I think linux.
http://measureofchaos.blogspot.com/2011/03/simutrans-server-administration-guide.html

Note: Timothy's 128.brit server still down (paused) after more than 2 hours (since 10:52). Checked at 13:05 UTC.
Title: Re: deManglers test Simutrans-Experimental Server
Post by: prissi on July 11, 2011, 03:12:02 PM
I only tested this locally, and there is worked. Ok, maybe also simutrans has to set a timeout.
Title: Re: Re: deManglers test Simutrans-Experimental Server
Post by: jamespetts on July 14, 2011, 12:16:35 PM
Perhaps this needs to be tested further?