On 20 Feb 2009, at 17:23, Robert Newson wrote:
I expect the b-tree wastage is minimal (though not zero).
actually not, only after compaction.
Cheers
Jan
--
I've wondered what happens on filesystems that don't support sparse
files, I assume they'd just be slower and use more disk space. Given
that the holes vanish after compaction, I suspected a bad calculation
in the code (couch_db.erl, I think), but I've not found it, it seems
to do the right thing. HFS+ doesn't support holes but I'm pretty sure
NTFS does.
Btw, it's mostly around attachments. If you add lots of documents but
no attachments, ls and df are in close agreement.
B.
On Fri, Feb 20, 2009 at 4:00 PM, Jens Alfke <[email protected]>
wrote:
On Feb 20, 2009, at 6:03 AM, Pascal Borghino wrote:
I am currently compacting it... even if 'Compaction rewrites the
database
file, removing outdated document revisions and deleted
documents'... no
document should be outdate neither deleted...
In addition to the sparseness of the file, another reason for the
size
difference might be obsolete b-tree nodes. The file is append-only,
so any
time a b-tree changes, the old nodes remain in the file. If you've
done a
large number of individual insertions, that space might be
significant.
(Probably not gigabytes, though.)
[email protected] wrote:
I find the actual
consumed space is far, far less that 'ls' shows. CouchDB .couch
files
are very sparse, large gaps of unwritten data, ostensibly to keep
btree and document items separate, but these 'holes' vanish after
compaction, even if you have zero updates and deletes.
Hm. But not all filesystems support sparse files. HFS+, the Mac OS
filesystem, doesn't. (Does NTFS?) Is there an option to suppress
the gaps?
—Jens
