On Tue, May 17, 2011 at 10:38, Matthew Woodward <[email protected]> wrote: > We have two CouchDB servers behind a load balancer and based on some > behavior I'm seeing I had a question about how views get generated in > that kind of setup. > > When I create a new view and hit each server in the cluster > individually, it of course takes a while for the view to be created > initially, which is expected. > > What's unexpected (at least to me) is that when I hit the cluster, > even if the views have already been created on each individual server > it seems (based on the time it takes for the view to return data) that > they have to be created again. > > Also, based on behavior I was seeing this morning, it seems that the > view has to be created for each server in the cluster even if you're > hitting the cluster URL/IP. > > To make this more concrete, assume cluster is the URL for the load > balancer, and couch1 and couch2 are the individual servers. > > 1. hit server1 -> view index is created > 2. hit server2 -> view index is created > 3. hit cluster -> time delay while view index is created even though > it already has been created on server1 and server2 > 4. hit cluster again -> if I happen to get bounced to the server I > didn't hit in 3 above, there's another delay while (presumably) the > view index is created > > Is there something to do with the host name that ties into views that > would be causing this behavior?
There is nothing tying views to hostnames. > > Happy to provide additional details if necessary. > > Thanks. > > -- If you can think of any more details that are relevant, please post them. Are you sure that your views are being re-indexed? When you experience the delay through the load balancer do you see indexing in /_active_tasks on that node? What is your load balancer? Perhaps it's buffering the reply. > Matthew Woodward > [email protected] > http://blog.mattwoodward.com > identi.ca / Twitter: @mpwoodward > > Please do not send me proprietary file formats such as Word, > PowerPoint, etc. as attachments. > http://www.gnu.org/philosophy/no-word-attachments.html >
