incubator-couchdb-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mark Deibert <mark.deib...@gmail.com>
Subject Re: Storage limitations?
Date Mon, 11 Nov 2013 14:41:12 GMT
A followup on the "1000's of images" question. I could approach this a
couple ways. Currently each image is attached to it's own Photo doc. Which
I've read is better for replication than one attachment with many
attachments. So that's fine, but will Couch have any issue managing several
thousand of these Photo docs, each with a 3MB'ish image attachment? If you
were building this Couchapp, would you...

1) Keep the photos as described above in one CouchDB
2) Move the Photo docs with attachments out into a separate CouchDB
3) Do 2, but break Photos into multiple categorized CouchDBs
4) Upload the images to the filesystem, just store the link in Couch

I want to build this Couchapp in such a way as to not make life miserable
for CouchDB :-D




On Sun, Nov 10, 2013 at 6:33 PM, Dave Cottlehuber <dch@jsonified.com> wrote:

> On 10 November 2013 23:14, Mark Deibert <mark.deibert@gmail.com> wrote:
> > I read an article somewhere that using include_docs is "hard" on memory
> or
> > disk or in some way taxes Couch and therefore you should just emit the
> doc.
> > Is this true?
>
> Like most general statements it has some truth and some lies in it :-)
>
> views and docs are stored in separate .couch btree files on disk.
>
> emit(key, doc) puts a full copy of the doc (that's already in the doc
> .couch b~tree) into the view b~tree.
>
> advantage - no need to hop over to the doc .couch file to retrieve the
> document.
> disadvantage - you now have 2 copies of the doc in separate files, wasted
> space.
>
> If you do things right, and your app fits this model, the generated
> etags from views and docs can be cached in nginx or similar, and
> repeated queries don't need to hit your couch.
>
> So yes, include_docs means extra reads, but like most of these things
> you should benchmark your situation, under a realistic load, not just
> pumping 1000 single-doc reads at it.
>
> A+
> Dave
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message