On Fri, Jan 10, 2014 at 8:45 PM, Paul Davis <[email protected]>wrote:
> Originally I intended for other indexers to be callback behaviors > loaded into couch_index. Since writing that code though I've come to > the conclusion that its not the best approach. If you want to collapse > it somewhere I'd collapse it back into couch_mrview and undo some of > the abstractions. > > As to collapsing it into the main couch application that makes no > sense seeing as the entire point of pulling it out was to remove the > dependency of the main couch application on indexers. > But right now all_docs is also in couch_index, which makes it difficult to completely separate from couch. So it would also require to put some code back in couch imo. Another thing that is not quite clear is how to have couch_index_server to resist to the couch application restart or crash. When it crashes (or restart in test) the notifier hangs and the couch_index_server exit: https://github.com/apache/couchdb/blob/master/src/couch_index/src/couch_index_server.erl#L127 I am not quite sure how to make it more resilient at this point. Any idea? - benoit > > On Thu, Jan 9, 2014 at 4:25 PM, Benoit Chesneau <[email protected]> > wrote: > > Hi all, > > > > Maybe Paul can answer. But I am wondering why we have couch_index in its > > own folder. For me it is quite difficult to see it separated from couchdb > > (even with the supervision). While couch_mrview fit well in its own > > repository since one can consider to replace it by another kind of > index. I > > am thinking we should put it back in the main couch app. > > > > Thoughts? > > > > - benoit >
