CrashSaver

[hemagx]

Hi,

Sometimes we get crash without know the reason and GDB log can't help so what we will do ? Q_Q

Here's my idea to help servers owners to know reason of that unknow crashes simply a little log name CrashSaver that record every thing happen in the server and every 5 min delete the log and start record again ( it's record every thing player attacked somethin,player used skill id,player used NPC,player traded, blah ) so we can check last actions happened in the server and try to know how crash happened.

some issus we will have with this High load on server , Super Duper number of SQL queries so i think we have to make it optional thing can be enabled or not only in important cases.

thanks

 

[Michi]

+1

 

[KeiKun]

i don't see gdb hard to read

>_<

 

[sketchyphoenix]

To be blunt, a stack trace or core dump should give you a general idea even if you can't figure out the exact cause. Needing something like this creates pointless overhead and really just means you should read your crash logs more carefully or work backwards with the information you do have (or use a different debugger)

 

[hemagx]

i don't see gdb hard to read

>_<

To be blunt, a stack trace or core dump should give you a general idea even if you can't figure out the exact cause. Needing something like this creates pointless overhead and really just means you should read your crash logs more carefully or work backwards with the information you do have (or use a different debugger)

Some Times even Herc Devlopers or any other Developers ( rA , brA , 3ceam , eAthena ) can't know how the crash happened and can't do tests since the CoreDump give us general things so this log can help a lot,,,

 

[KeiKun]

i don't see gdb hard to read

>_<

>To be blunt, a stack trace or core dump should give you a general idea even if you can't figure out the exact cause. Needing something like this creates pointless overhead and really just means you should read your crash logs more carefully or work backwards with the information you do have (or use a different debugger)

Some Times even Herc Devlopers or any other Developers ( rA , brA , 3ceam , eAthena ) can't know how the crash happened and can't do tests since the CoreDump give us general things so this log can help a lot,,,

Read the gdb's freaking manual/doc??

and its already helpful...

 

[sketchyphoenix]

i don't see gdb hard to read

>_<

To be blunt, a stack trace or core dump should give you a general idea even if you can't figure out the exact cause. Needing something like this creates pointless overhead and really just means you should read your crash logs more carefully or work backwards with the information you do have (or use a different debugger)

Some Times even Herc Devlopers or any other Developers ( rA , brA , 3ceam , eAthena ) can't know how the crash happened and can't do tests since the CoreDump give us general things so this log can help a lot,,,

It's not helping anything. Here's why.

What you're saying is that wasting processing time and memory (that increases with each player on the server, I might add) to clutter up the entire codebase with potentially large amounts of cache for the off-chance that 1 tiny little entry is going to have anything remotely useful to gather from it. The best part? You can still gather all that player data and not a single bit of it is useful if it turns out somebody just plain wrote buggy code. This is literally how you should not debug a program, ever.

Crashes that happen that can't be so easily figured out from crash reports are almost always buggy code. It doesn't matter what skill somebody used, what level they are, what they were doing because all that data is accurate. What's not accurate is how it's being handled. That takes time to localize the problem, studying the flow of execution, identifying the error, correcting it, then testing it.

 

[Ind]

Its hardly feasible the actions are many and logging them even if with a single player online would be catastrophic (picture 50 monsters in a map moving and doing stuff even while the player is sitting still, that alone could result in some ~100 queries a second depending on how often the monsters walk, consider that 1 cell - 1 action, so moving 7 cells - logging 7 actions), the gdb data is as best as it can get in my opinion.