I have two questions about this.
First, is the code intended to apply to masters, slaves, or both? As
the code change was written, it will apply only to masters, because
slaves always have kdb_first_time set to zero. If we want it to apply
to both, checking for kdb_last_sno != 0 would detect whether the ulog is
non-empty on the master or has ever received updates on the slave.
Second, after looking at the iprop code in detail, I'm questioning
whether the safety net is necessary or just annoying. Can I have more
detail on the reasoning behind adding it?
On the master, loading a non-iprop dump will reinitialize the ulog
header which will cause full resyncs to the slaves, which seems fine to
me. And I can certainly see wanting to restore from a backup on a
master KDC running iprop.
On a slave, loading a non-iprop dump would be weird, but perhaps kpropd
isn't working at the moment and you need to manually propagate an
update. The slave will take a full resync from the master as soon as
kpropd starts working again, since the slave's ulog header will be
reinitialized.
RT for krbdev.mit.edu
