On Mon, 9 Jan 2012, ZHOU Eric JP wrote:
This issue only happens when power-off/power cut-off one of the
servers, and could be probably prevented by ?sending heart beat? to
verify the established connections.
[cutting out openldap-devel; this is usage]
Sure; see the associated ITS and the thread "No replication after power
The only thing I'd note is that "2 hours...is too long" doesn't seem
that strong an excuse. It's easily tunable in all modern OS I can think
of, and in some cases (Linux comes to mind) you can just setsockopt()
within the application itself so as to not alter the system-wide settings.
Actually, I'm starting to wonder if this should be an option you can pass
as a syncrepl (really libldap) configuration directive. There's something
to be said for having aggressive-ish keepalives on server-to-server
communication (with, hopefully, good bandwidth/latency/communication
costs/etc.) and keeping your site defaults for "normal" clients. OpenLDAP
already inched in this direction with the ability to set different
idletimeout, etc., for the syncrepl client. Is anybody interested in this?
Maybe I'll do it (or have a student do it) if a few people think it'd