[389-users] changelog issues when re-initing a master

[Robert Viduya]

Hi, apologies if this gets posted more than once, I'm having MUA issues.

We have a 4-master, 2-hub, 6-slave set up, scattered over 2 data- 
centers.  We had been running fine with 2 masters and fedora-ds 1.0.4  
on RHEL4 (32-bit), but this past summer, we decided to upgrade to 4  
masters, RHEL5 (64-bit) and fedora-ds 1.2.0.  We did clean installs,  
not upgrades and everything went pretty smoothly.

Lately, however, we've run into a problem where we've had to  
reinitialize one of the masters due to database corruption.  The  
process I've used was, on a running master, to right-click on the  
replication agreement in the gui and select "Export".  Then I'd scp  
the resultant ldif file to the down master and import it using ldif2db.

Once the import was done, restarting the server produces the following  
in the errorlog:

	[29/Oct/2009:05:21:05 -0400] NSMMReplicationPlugin -
	replica_check_for_data_reload: Warning: data for replica
	ou=accounts,ou=gtaccounts,ou=departments,dc=gted,dc=gatech,dc=edu was
	reloaded and it no longer matches the data in the changelog (replica
	data > changelog). Recreating the changelog file. This could affect
	replication with replica's consumers in which case the consumers
	should be reinitialized.

which is perfectly fine, you expect the changelog to be invalid the  
first time the server is brought up after a reinitialize.  The problem  
is that it does this each and every time the server is restarted from  
then on.

This sounds a lot like bug 388021, but nothing I've read in there  
helps.  I've tried deleting the changelog before importing, deleting  
the changelog after importing, setting the changelog size to 1 entry,  
delete-and-recreate all agreements... nothing.  If the server has a  
changelog configured and it gets restarted, it trashes it and recreates.

This system is a production system and importing our data is a good 8  
to 12 hour process.  Management really doesn't want to give me the  
time to start from scratch, so I can't take all the machines down and  
rebuild everything.  Besides, reinitializing a single master shouldn't  
involve a complete rebuild like that.

Any help would be greatly appreciated.