It will take a while before we try out 4.10 again. We might just jump directly to Solr 5 (waiting for some fixes regarding faceting).
One thing I noticed today that luceneMatchVersion was still set to 4.7 and not to 4.10.4. But I would expect Solr 4.10 to handle this. > On 23 Nov 2015, at 12:21, Upayavira <u...@odoko.co.uk> wrote: > > Can you confirm that the indexes *are* in place on disk before you shut > down? Whilst it may seem like indexes are being removed, this would seem > extremely unlikely, and the explanation is likely something else. > > Try looking at your index files, post some content, commit, look again > at your index files, then shut down. What changes have you seen? Did the > commit 'take' and cause the files to be written? > > Upayavira > > On Mon, Nov 23, 2015, at 10:25 AM, Oliver Schrenk wrote: >> Yes. Hard commit using SolrJ 4.10.4 client with UpdateResponse.commit() >> which defaults to waitFlush=true and waitSearcher=true >> >> We reverted back to Solr 4.7 for the time being, which is a pity as we >> really benefitted from the concurrent commit fix. >> >>> On 17 Nov 2015, at 18:00, Erick Erickson <erickerick...@gmail.com> wrote: >>> >>> Did you commit after indexing and before shutting down? Even if you didn't, >>> I'm >>> still a bit surprised, but that's one possible explanation. >>> >>> But this is the first time I've seen this problem mentioned... >>> >>> Best, >>> Erick >>> >>> On Tue, Nov 17, 2015 at 4:08 AM, Oliver Schrenk <oschr...@elmar.nl> wrote: >>>> Hi, >>>> >>>> since we upgraded our cluster from 4.7 to 4.10.4 we are experiencing >>>> issues. When shutting down the service (with a confirmed graceful shutdown >>>> in the logs), the index is dropped, with only one lonely `segments.gen` >>>> file left for each shard and all other files being deleted. >>>> >>>> There is no message in the logs, other than graceful shutdown. Did anybody >>>> have a similar issues and has some advice? >>>> >>>> Cheers, >>>> Oliver >>
