Quanah Gibson-Mount quanah@stanford.edu writes:
--On Friday, October 06, 2006 9:36 PM +0200 Dieter Kluenter dieter@dkluenter.de wrote:
"Golden Butler" golden@cnt.org writes:
[...]
In general slapd writes to disk as soon as possible, depending on pending read operations, that is, a delay of a few seconds my occur. Not writing to disk for a few month seems to be quite odd.If you use a bdb or hdb database you might try to recover the database by means of db_recover. In order to force slapd to write to disk you may define a checkpoint in slapd.conf(5)
IIRC, it only flushes to disk when told to, which is either via the checkpoint settings, or a *normal* slapd shutdown. Otherwise, it does not. So his scenario makes perfect sense to me. Assuming the cache setting holds the entire DB, I believe.
AfAIK, this is only true if the flag DB_TXN_NOSYNC is set. But even than the cache is flushed at an idle state of operations.
-Dieter