karavelov@spnet.net wrote:

Hello,

I have recompiled slapd (2.4.18-release) on another machine (i386) in order to see if the bug is architecture dependent - the other servers are on amd64 architecture. It shows the same bug.

The log file could be found here: http://purgatory.spnet.net/~karavelov/d2

for the failing record, search for: filter="(&(objectClass=mailDomain)(dc=justillusion.net))"

It does not fail with "scope not ok" error, but with "bdb_search: no candidates". I have seen the same error appear on the other servers (amd64) but I had trouble to isolate the whole history of the query. On this test server I have seen the other error too ("scope not ok") so the failing mode is not architecture dependent.

This is now fixed in HEAD overlays/pcache.c. The bug occurred because the cache was seeing a child entry first, and so the dc=justillusion.net object got created as a glue entry. Later when the actual dc=justillusion.net entry was received, the modify to store its true values in the cache DB failed because it needed the manageDSAit control.