So I recently had some data loss (non solr related) on an ext4 filesystem (ubuntu jaunty alpha5). https://bugs.launchpad.net/ubuntu/+source/linux/+bug/317781
The upshot? It's not really a filesystem bug... it's slightly different semantics that are still allowed by posix, applications that aren't careful enough, applications that rewrite their files often, combined with a larger window before data hits the disk in ext4. Lucene already handles this by explicitly calling fsync on the index files before writing out and syncing the segments file that refers to all the index files. It seems like Solr should do the same when replicating files? It's especially important in the case of replicating config files... if that gets messed up from a crash-reboot, Solr wouldn't be able to come back up. -Yonik http://www.lucidimagination.com
