Minkata test shard -- discussions and questions

[Hoikas]

I've now had two CTDs when jumping into the starry void from the bahro cave -- once from Gahreesen and just now from Eder Kemo (like Anaerin above). This second time Yeesha didn't start speaking, but I was still able to grab the pole and jump. I'll post a Hesk ticket and a compressed log.

Bugs like these are problematic, even on MOULa. Unfortunately, no developer that I know if is able to reproduce these crashes. And, of course, if we cannot reproduce them, we can't figure out what's going wrong :\

[Mac_Fife]

Hoikas raises a very important subject in that post - reproducibility. It is indeed difficult to fix a problem if you can't see why it happens and some of the kinds of bugs we're seeing mentioned that also feature in MOULa, are quite intermittent. They might be repeatable for a particular user, but impossible for a developer to replicate. You get the feeling that some bugs are actually related to video driver issues or some aspect of the user's physical setup, so it's near impossible for the developers to recreate the circumstances where the problem occurs.

So, where possible, when you find a bug it's a good policy to try to recreate it yourself (maybe with another avatar, or by resetting the age and doing it over again, if necessary) and add any additional observations to your original ticket. The more information that exists on when a bug does or doesn't occur, then the greater the chance of getting to the root of it.

Any tickets that the OpenUru.org team recognize as issues that exist in MOULa, or are clearly not issues that are specific to the shard and it's client, will be moved over into our JIRA database for developer attention. There are a couple of those in the support ticket system right now, but of course with the current holiday season we're only popping in very occasionally to keep an eye on things .

[semplerfi]

As a good practice I always defragment the game files after a crash and or a stack dump. Crashes have a tendency to fragment the game files badly which just complicates the issues.

[Christian Walther]

I’ve done a bit more investigation on my Windows XP problem.

• The PhysX_Setup.exe download loop is not related to FAT32, it also happens on an NTFS volume.
• Watching network traffic with Wireshark, it looks like every TCP conversation has something like a 20% chance of working. When it works, it works all the way through. When it doesn’t work, there are a lot of retransmissions and Dup Acks, sequence numbers never get past 1, and the two ends somehow continue talking past each other with diminishing frequency until the server finally resets the connection after about 191 seconds. At that point, the client sometimes reconnects (and gets another 20% chance of getting a step further), sometimes not (and sits there stuck indefinitely). I’m not fluent enough in TCP to diagnose what really happens, but if someone wants to have a look, I can provide the packet trace files. Maybe a simultaneous trace on the server end would be insightful too.
• In an attempt to work around the problem, I tried it on Mac OS on the same machine using the MOUL Cider wrapper. No networking problems there, however the launcher doesn’t get very far because no macExternal manifest is available on the file server.

[rarified]

Windows TCP stack has an unpleasant habit of closing TCP connections with a reset (RST) rather than the specified FIN handshake. It's been speculated that Microsoft did this to reduce resource demands when there are a lot of quick small transfers. But it still causes headaches; MOSS has code to tolerate a close-by-reset, but there may be something else going on.

I'm thinking on first thought there is a checksum problem that causes one endpoint or another to ignore traffic in this case. We've had enough XP clients now accessing the server that I don't think there is something specific to XP in general that is causing the problem.

A packet trace might be interesting if you can make it available. I should be able to trace from the server side as well if we can coordinate times.

_R

[Hoikas]

Before Windows Vista, the Windows TCP stack was pretty terrible. Before MOUL, Uru used UDP... I wish they hadn't changed it. The older network protocol was much more reliable, never blocked, was fully cross platform compatible, and integrated much more cleanly with the rest of the engine. Sadly, the servers themselves did not scale well, so Cyan decided they needed to rewrite the whole thing. Now Cyan has servers that scale well, but networking code that is finicky at best... And that's not even taking into account the benefits we got from avoiding TCP.

[manloc]

Hi,
I can't download "MinkataUruInstall.exe" :

Service inaccessible!
En raison de travaux de maintenance ou de problèmes de capacité le serveur n'est pas en mesure de répondre à votre requête pour l'instant. Veuillez réessayer plus tard.
Si vous pensez qu'il s'agit d'une erreur du serveur, veuillez contacter le gestionnaire du site.
Error 503

I'm on "Google Chrome 16.0.912.63" (Windows 7, Familial Edition x64 SP1).

[Mac_Fife]

I can confirm that the download service on the Foundry doesn't seem to be operational right now. The Minkata shard and other Foundry services seem to be OK, so I'll ask rarified to take a look as soon he's available. It's probably something that didn't get re-started following the downtime yesterday.

[manloc]

OK, thank you.

[rarified]

I was out for a bit; yes there was a missing service that provided the downloads. It should be up now.

Sorry for the oversight

_R