Dieter Maurer wrote:
Martijn Faassen writes:
For some reason, at random times, Postgresql seems to magically lose some records that were already there and should've been committed. This seems to happen within a few hours of the time the records were first created. All trace of the records disappears from the database; we can see the holes in the autoincrement primary key fields (and we never do any delete ourselves that could cause this). Several months ago, someone reported (zope-dev mailing list) a hole in the database adapters transaction handling. Jim confirmed the bug.
Hm, I'll try to look for that. Is there a fix, and if so, which Zope version? :)
When my memory is right, the hole means that database transactions may be hanging, Zope forgets to either commit or abort them.
I does not look like your problem, but maybe the two problems are related...
Well, a non-commit or abort *could* mean weird behavior like this. I've also found some ConflictErrors generated by ZODB/Connection.py that may be related. Thanks, Martijn