* Gerd Stolpmann:

>> All my crashes could be traced back to operator error and faulty
>> hardware.  Berkeley DB has a very bad reputation when it comes to data
>> integrity and durability, but in my experience, this is simply not
>> true.
>
> I think it is true.
>
> A good database must not become corrupt even if the hardware crashes.

It depends on the crash.  In my case, hardware flipped bits randomly
under high load, eventually leading to a kernel crash.  This is hard
to correct in software.  Generally, it's better to fix the broken
hardware.

I haven't lost data due to simple crashes yet (power outages,
crash-early kernel bugs, or MCEs because of thermal issues -- knock on
wood).