|
From: Doug C. <cu...@ap...> - 2008-02-29 21:10:21
|
Ning Li wrote: > A copy of a document will be stored in a stored field, right? If we don't keep complete documents in the log, then I figured that all fields would be stored, so that document retrieved from the index could be directly added to another index. > First, keeping both a doc and its inverted form in an index > means storing the doc and indexing the doc are done in the > same "transaction". A traditional document database often > store a doc first and then index it later (hopefully soon). It should be possible to configure Lucene so that adding a document never triggers any blocking merges. So folks would have to wait for the single-document inversion, but no more. Today, it won't yet be visible to search for a bit, until a commit, but perhaps in a future version of Lucene it might be possible to search unflushed segments. > Second, a traditional document database often supports > updating a doc's "metadata" such as author or date. We > don't support this or we say a document is name-value > pairs and we reconstruct from stored fields and support > such update? I assume that incrementally updateable fields must not be indexed? In any case, I hope that indexing is fast enough that updating by add+delete is okay. CouchDB seems to get away with it... > #5 with non-sync light-weight logging? Should work. Yes. Doug |