At Meebo, we have to re-index when we release new versions of our couch apps. We do this by uploading the new design documents with a -stage suffix.
We grow our Lounge cluster so that it can support indexing a _design/app-stage document while serving requests for _design/app. Once app-stage has indexed completely we update app and drop the -stage. Since CouchDB checks the actual code content of the view functions renaming the design document does not cause CouchDB to re-index. We find that with multi-core systems we rarely have issues with CPU load and disk I/O is the bottleneck for both the indexing and serving. Randall On Tue, Mar 23, 2010 at 02:49, Benoit Chesneau <[email protected]> wrote: > Hi all, > > I would be curious how you manage reindexation of big datasests in > your couchdb. Say you have > 100K docs in couchdb. How do you do to > handle the cpu load ? > > Is distribution enough to dispatch the load ? Would be curious to know > how cloudant, ubuntuone or even the BBC do or what are your experience > about it. > > > - benoit >
