Involuntary Logouts--request for debugging suggestions

brianjbatchelor

New member
I have an odd problem that I would appreciate some suggestions as to how to get enough information to understand it. I hesitate to list it as a bug, as the problem is location specific. I can use the same DropBoxed setup at one location with no issues, but at another, I get logged out of KoL every few commands. Mafia notices this and immediately logs me back in. My CLI looks something like this:

Code:
 Password hash mismatch
Path: "/inv_use.php?pwd==3&whichitem=4502&ajax=1&_=1320353 617054"
Validating login server (www.kingdomofloathing.com)...
926 players online.
Sending login request...
Synchronizing moon data...
Requests complete.

Validating login server (www.kingdomofloathing.com)...
926 players online.
Sending login request...
Synchronizing moon data...
Requests complete.

You have the maximum number of AT buffs already.
You acquire an effect: Donho's Bubbly Ballad (duration: 60 Adventures)
Validating login server (www.kingdomofloathing.com)...
925 players online.
Sending login request...
Synchronizing moon data...
Requests complete.

Validating login server (www.kingdomofloathing.com)...
925 players online.
Sending login request...
Synchronizing moon data...
Requests complete.

I play almost exclusively in the relay browser, so this behavior manifests as a lot of odd pauses and GM scripts behaving badly as the PW hash changes out from under them. I have tried setting mafia debugging and logging to their most verbose, and I have not been able to find any instance of a log off command being given--just an endless string of logging on again and again. I'm now starting to look for things specific to one location which could cause mafia to believe that a session has been lost. We do a lot of wacky network stuff at that location--from SSL proxying to DPI and heavy packet shaping, so its easy to understand how some local eccentricity is causing this wonky behavior. However, the only changes preceding the start of my session dropping was a change our internal authentication, Novell eDirectory to MS AD.

I want to stress that I realize this is my problem and not Mafia's--I am not looking for someone to solve my problem, I am trying to get good Ideas of where to look myself. I am a programmer by trade and so while networking is not foreign to me, its not something I have an intuitive feel for. So if anyone has read this far, and has Ideas about how to investigate further, I would greatly appreciate a push in the right direction.
 

Bale

Minion
Since this is not a report of a KoLmafia bug, I'm moving this out of the bug reporting forum. 'Cause that's what a good minion does. ;)

I'm clueless as to your problem, so all I can do is wish you, "Good luck."
 

fronobulax

Developer
Staff member
Which version of mafia are you running? I seem to recall some similar behavior when KoL changed their load balancing scheme.

I have sometimes confirmed a problem is network related by running a "ping -t" aimed at the site that I am having problems with, while I try and recreate the problem. This will make bandwidth problems worse and may point at latency or lag issues that might be culprits.
 

brianjbatchelor

New member
Which version of mafia are you running? I seem to recall some similar behavior when KoL changed their load balancing scheme.

The issue started when I was using 9958, I rolled back as far as 9956 and it still occurred. I have been updating to the latest build every morning since.

On the thread of the load balancers though, what is the chance there is some legacy bit somewhere which is determining connectivity by hitting www#. rather than just www.kingdomofloathing.com? The problem persists with greasemonkey disabled, so its not a request from one of those scripts. However it has to be something that is treated differently in different network locations--is there a boundry condition where a request would not resolve in a heavily monitored environment, but would in a vanilla connection?

I have been trying to avoid it, but I am running out of Ideas beyond running a full TCPDump and trying to match timestamps for when the logoffs occur. They don't happen unless I do things in the relay browser, so there is some causal link to activity.
 

holatuwol

Developer
I don't know if the error occurs elsewhere, but the only place where I remember KoL throwing a 'password hash mismatch' is in chat? Are you in chat when these errors happen?
 

Bale

Minion
Could this actually be because of a recent KoL bug that CDMoyer explains HERE?

CDMoyer said:
As some have surmised, it's because the proxy server picks a back-end based on your IP address. (well, the class C of your IP address), and some of you unlucky folks have connections that jump between class Cs regularly.

Our long term plan is to move sessions to a central location so this won't be an issue.

Our short term plan is investigating some other modes that nginx can operate in for picking a back-end, such as cookie based sticky sessions.

Sorry about this pain, folks.
 
Last edited:
Top