Does the patch that detects the corrupt indices cause proper handling of this situation? In other words, customers dont notice this anymore and indices get discarded and rebuilt?
They'll still get some "out of sync" error and get disconnected, I
think. Next login would anyway work again.But looking at the index files you sent me they weren't corrupted at
all, so the problem is that the in-memory views of the indexes get
somehow broken. Do you use mmap_disable=yes? You could try switching
that setting the other way. What about lock_method?
mmap_disable = yes lock_method = fcntl
I guess we can turn mmap_disable off, since our indexes are on local FS.
Did this problem start only with rc15? What version were you using
before?
We were using beta8 before, it was the only one that didnt have some problem we couldnt live with. We've tried some RCs, but every time we had some major problem. RC15 seems workable, but with a few issues.
I'll turn mmap_disable off. Anything else I can do?
Cor