On Mon, May 19, 2014 at 6:14 AM, Clemens Wyss DEV <clemens...@mysign.ch> wrote: > Mike, > first of all thanks for all your input, I really appreciate (as much as I > like reading your blog).
You're welcome! >> Hmm, but you swap these files over while an IndexReader is still open on the >> index? > no IndexReader is open while swapping. At least not by design. We have at > most one (current)reader per Index, which is closed just before swapping. I > will investigate on this more deeply, though, me not being impeccable ;) Hmm, double check that. The only way I can explain what you're seeing is if the reader was open on the index, then the index was blown away / replaced, then openIfChanged called ... >> Instead, use IndexWriter.deleteAll, then index all your new docs, >>then commit, then reopen the reader. > we were facing memory issues and hence decided to do "intermediate commits" > (i.e. flush the writer). And in order to still provide "atomicity" (with > respect to a full re-index) we decided to go the before mentionend approach. Can you just decrease IW's ramBufferSizeMB to relieve the memory pressure? >> Separately, it's better to use a near-real-time IndexReader so >>you don't have to commit in order to see changes. > What if I were to search right after deleteAll? That's fine, it's fully transactional: the deleteAll will not be visible until you open a new reader. If you change your mind, you can call rollback after deleteAll and the docs are back, etc. Even opening a new IW with OpenMode.CREATE is transactional: that change will not become visible to a reader until it's reopened. Mike McCandless http://blog.mikemccandless.com --------------------------------------------------------------------- To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org For additional commands, e-mail: java-user-h...@lucene.apache.org
