On Monday 19. March 2012 22:56:42 Marco Schulze wrote: > Synchronization is the reason every thread should wait. If the log is > always flushed and fred crashes, you know exactly where the last good > checkpoint was before the crash. If the log is buffered (or > asynchronous), the thread may be miles ahead from the last message > written to disk, and suddenly you have no idea where to look for the > bug. Shotgun debugging indeed.
How often does this actually happen? Considering the performance penalty it has to fix a very real problem IMHO, especially since having logging change how the code behaves in such a dramatic way will make debugging anything that depends on timing that much harder.
