Login Bugs and Crashes List

[John Adams]

Login finally crashed again, all that was happening was Dakadin's world trying to fetch players.

Stack

Code: Select all

20:03:31.686 D Net      Received world info request for DakadinTest
20:03:31.686 D WSList   Fetching WorldServer by name: 'DakadinTest' (21)

Program received signal SIGSEGV, Segmentation fault.
[Switcbtng to Thread 0x7ffff5c75700 (LWP 3859)]
#0  0x0000000000430e1a in std::__weak_count<(__gnu_cxx::_Lock_policy)2>::_M_get_use_count (this=0x8) at /usr/include/c++/4.8/bits/shared_ptr_base.h:672
#1  0x000000000042f3f6 in std::__weak_ptr<Client, (__gnu_cxx::_Lock_policy)2>::expired (this=0x0) at /usr/include/c++/4.8/bits/shared_ptr_base.h:1277
#2  0x000000000042e27c in std::weak_ptr<Client>::lock (this=0x0) at /usr/include/c++/4.8/bits/shared_ptr.h:497
#3  0x000000000042b010 in UDPServer::HandleWriteHelper (this=0x669d00 <net>, data=0x0, now=1408417411688) at ../common/UDPServer.cpp:497
#4  0x000000000042aefe in UDPServer::HandleWrite (this=0x669d00 <net>) at ../common/UDPServer.cpp:477
#5  0x000000000042968f in WriterThread (data=0x669d00 <net>) at ../common/UDPServer.cpp:126
#6  0x000000000042901a in ThreadRun (arg=0x685b50) at ../common/Thread.cpp:77
#7  0x00007ffff768e182 in start_thread (arg=0x7ffff5c75700) at pthread_create.c:312
#8  0x00007ffff6ea138d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:111

Looks like the same stack reported above which was never addressed.

[Dakadin]

[quote]Login finally crashed again, all that was happening was Dakadin's world trying to fetch players.[/quote]

Sorry about that. I was worried that I caused it. I was testing a few things for the waiting on data issue and I obviously did something the server didn't like.

I will see if I can figure out why it crashed so we can resolve it.

[John Adams]

Don't be sorry, not ever! Crashes are awesome, so we can make the code more stable. Hopefully someone knows what the problem is, though it seems to happen infrequently, but mostly when worlds are coming up and down rapidly (I think).

[Lokked]

[quote="John Adams"]There is one major LoginServer bug (aside from crashes) I'd like to get resolved soon, too. It seems when worlds are connected, if Login should go down (crash or a restart), the connected worlds -- never reconnect.

Such is the case for New Telon last night. Login crashed around 7:47pm MST, and came right back up. New Telon lost it's TCP connection at the same time (the hint to me that login crashed)
[quote]19:47:13.481 E TCP TCP Connection disconnected from 68.98.218.5:10102
19:47:24.011 I Net Connected to login server 68.98.218.5:10102[/quote]
But, Login itself never got the signal that the world reconnected - and thus, New Telon has appeared "Down" for the last 12 hours - when it was not.

Login crashes are very rare, so they are hard to catch without watching them closely all day. I'll try to debug it today.[/quote]

I've fixed this and committed (Rev 618):

Code: Select all

Added SendConnect() to Process(), section where tcp.reconnect() is called. This will send the proper TCP Connect packet to the Login Server, to properly re-establish connection after a Login Server crash.

It was an issue with the World Servers. Previously, when the TCP Socket was recreated on a WS after it lost connection to LS (after LS crash), the code followed a different path than when the Socket was created when first starting the WS. Was an easy one

[John Adams]

Cool, thanks. Giving it a try today.