Re: [Monotone-devel] Re: mnt: fatal: std::logic_error: database.cc:1386: error

[Nathaniel J. Smith]

On Sat, Dec 02, 2006 at 10:33:35PM -0800, Zack Weinberg wrote:
> If it does not always fail the same way every time, you should check
> for hardware faults (bad RAM, especially); monotone should be wholly
> deterministic.

I wonder about bad RAM too with these errors, but we don't actually
have the data to know it's non-deterministic; the problem is that the
second pull started up where the previous one left off.  Notice it has
way fewer revisions to pull...

(Now, one might expect it to have immediately then caused the same
error, but there's some arbitrary ordering and stuff involved, maybe.)

One thing to keep in mind, though, is that even on unreliable hardware
you can use mtn with a lot of confidence -- it will notice any serious
problems that would corrupt its results, so when a mtn command
succeeds, you can trust it.  If mtn _is_ behaving
nondeterministically, though, then you might want to run memtest86
etc., just in case you use some other programs that are a little less
robust against having their memory randomly scrambled... :-).

-- Nathaniel