sang jun song wrote:
The changes in the Retro-Changelog (RCL) database are not guaranteed
to be in order
even in case of a single server, because the incoming requests are
independently, and the scheduling of the worker threads – that process
– is driven by the thread library.
There is a much higher risk for out-of-order changes in the RCL
database in case
replication is deployed (for reasons other than the above mentioned
Invalid Changelog Entries
The contents of the RCL database cannot be fully trusted, because RCL
records the changes received by the directory server and not the
changes applied to
the database by the directory server. Typically, the URP code (update
protocol) may modify the received change and the RCL plugin may record
changes in the RCL database.
Is solved it?
There hasn't been any new work done in the retro changelog code for a
so I suspect the answer is no. However, perhaps you could tell us more
intended application using the rcl ? There may be a different way to
achieve what you want
BTW, there _is_ now a mechanism to prevent out-of-order processing of
updates inside a single server : it's used in replication sessions only
but if someone had a strong motivation to ensure in-order processing they
could make code changes to enable that existing mechanism.