On Thursday 31 May 2001 21:44, Andreas Dilger wrote:
> I noticed something interesting when running "mongo" with debugging
> on. It is adding filenames which are only sequential numbers, and the
> hash code is basically adding to only two blocks until those blocks
> are full, at which point (I
On Thursday 31 May 2001 21:44, Andreas Dilger wrote:
I noticed something interesting when running mongo with debugging
on. It is adding filenames which are only sequential numbers, and the
hash code is basically adding to only two blocks until those blocks
are full, at which point (I guess)
Daniel writes:
> On Thursday 31 May 2001 21:44, Andreas Dilger wrote:
> > I think Al's idea of doing the validation once on the initial
> > read is a good one.
>
> I'm doing that in the current patch, for leaf blocks, look at
> ext2_bread. For index blocks, ext2_bread needs help to know that a
On Thursday 31 May 2001 21:44, Andreas Dilger wrote:
> Daniel, you write:
> > - Fall back to linear search in case of corrupted index
>
> OK, I have _some_ of the code needed to do this, but in one case I'm
> not sure of what your intention was - in dx_probe() you check for
> "unused_flags & 1"
Daniel, you write:
> - Fall back to linear search in case of corrupted index
OK, I have _some_ of the code needed to do this, but in one case I'm
not sure of what your intention was - in dx_probe() you check for
"unused_flags & 1" to signal a bad/unsupported index. Why only check
the low bit
Daniel, you write:
- Fall back to linear search in case of corrupted index
OK, I have _some_ of the code needed to do this, but in one case I'm
not sure of what your intention was - in dx_probe() you check for
unused_flags 1 to signal a bad/unsupported index. Why only check
the low bit
On Thursday 31 May 2001 21:44, Andreas Dilger wrote:
Daniel, you write:
- Fall back to linear search in case of corrupted index
OK, I have _some_ of the code needed to do this, but in one case I'm
not sure of what your intention was - in dx_probe() you check for
unused_flags 1 to signal
Daniel writes:
On Thursday 31 May 2001 21:44, Andreas Dilger wrote:
I think Al's idea of doing the validation once on the initial
read is a good one.
I'm doing that in the current patch, for leaf blocks, look at
ext2_bread. For index blocks, ext2_bread needs help to know that a
block
8 matches
Mail list logo