Warning chaos!

[Mystler]

Hey, i have a quite interesting problem.
My moss_auth.log file has a size of 86MB and contains a line like "WARN: Socket 6 selected for write but could not write" VERY often.
As I could see there were the same warnings in the 1.5MB moss_backend.log file. I had some for Socket 6, Socket 15 and some more but scrolling down the files was kinda horrifying.
Additionally, the avatars that were created before those warnings appeared cannot login anymore. They are getting a network error after clicking on "Explore URU".

Do you have any idea what the heck is happening?

~Mystler

[a'moaca']

Wow! That's logged as a warning because it is not supposed to happen!

Briefly, there is a function (named "select") used to ask the OS, "Do you have any space in your buffer where I can put data to be sent over the network?" When the OS says yes, then one should be able to put data in the buffer. That error is telling us the OS said there was space to write data but then there wasn't after all. This is supposed to be an unusual circumstance.

If this happens every single time, something is messed up, and no data will make it to the client. What exactly is the network error the client gives you? It is usually not very useful, but it could be illuminating.

What you need to do is pick one of these sockets (the number identifies one particular connection, but it could be reused) and search backwards through the log until you find other events mentioning it. I would remove or set aside the current moss_auth.log and try to log in once with an affected avatar, then copy aside that log, so you have only one incident to look at. What other events were reported for that socket? If none, try setting the auth log level to DEBUG or even MSGS and repeat the experiment. I do not know what you are looking for, because this is a problem that just isn't seen very much. So let's see what you turn up.

- a'moaca'

[Mystler]

The "Socket could not write" lines are in the moss_auth.log while in the backend there are these ones:

Code: Select all

1312921303.030679  INFO backend: Backend startup
1312922301.604995  WARN backend: SQL error in VaultFetchRefs: ERROR:  stack depth limit exceeded
HINT:  Increase the configuration parameter "max_stack_depth", after ensuring the platform's stack depth limit is adequate.
CONTEXT:  PL/pgSQL function "fetchnoderefs" line 9 at FOR over SELECT rows
PL/pgSQL function "fetchnoderefs" line 13 at FOR over SELECT rows
PL/pgSQL function "fetchnoderefs" line 13 at FOR over SELECT rows
PL/pgSQL function "fetchnoderefs" line 13 at FOR over SELECT rows
...[this line repeats ca. 34000 times]

[Mystler]

Hm... everything works fine now. Seems it was a problem with a public age (had the UUID 0000000[...]) that has been created by MercAngels fan age scripts. I did it another way now and it works.

[cjkelly1]

Would it be possible for me to obtain either a copy of a fan age and the steps to reproduce the issue using Merc's tools, or perhaps a copy of the broken DB (either one is fine)? I would like to determine the root cause of this issue, and take steps to prevent it.

Thank you.