[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: CSN too old, ignoring - and therefore not syncing

Pat Riehecky schrieb:
On Tue, 2008-12-23 at 14:33 +0100, Dieter Kluenter wrote:
Pat Riehecky <prieheck@iwu.edu> writes:

Here is the quick and dirty what I am trying to do:

ldap1 and ldap2 are supposed to be in MultiMaster.  They are time synced
to pool.ntp.org and each other (if they drift I would rather they sorta
drift together, but pool should be keeping that in check).

Right now I am just beating them up to see how 2.4.13 performs. (So far
VERY well, minus this little problem)

I have a rather small ldif (41 entries) that just wont sync (I'm
starting small). Debug gives me 

ber_scanf fmt (m}) ber:
ber_dump: buf=0xb806f120 ptr=0xb806f137 end=0xb806f175 len=62
  0000:  00 3c 72 69 64 3d 30 30  31 2c 73 69 64 3d 30
30   .<rid=001,sid=00
  0010:  32 2c 63 73 6e 3d 32 30  30 38 31 32 32 32 31 37
2,csn=2008122217
  0020:  34 37 32 31 2e 38 35 35  39 30 34 5a 23 30 30 30
4721.855904Z#000
  0030:  30 30 30 23 30 30 31 23  30 30 30 30 30 30
000#001#000000
do_syncrep2:
cookie=rid=001,sid=002,csn=20081222174721.855904Z#000000#001#000000
do_syncrep2: rid=001 CSN too old, ignoring
20081222174721.855904Z#000000#001#000000
ldap_msgfree

I am not exactly sure how it gotten to be "too old." The ldif I am
importing is not the result of a slapcat or anything that would preserve
the CSN or UUID attributes (not that syncrepl uses UUID). I am loading
one single file with ldapadd which, in my understanding, sets up the CSN
and wouldn't let me import one anyway. 

Each server has no entries until I load the one, so there shouldn't be
any weird stale CSNs causing this.  They are "sync'ed" almost instantly
after the one system is loaded - I just don't have everything.

After a sync:
ldap1 - slapcat |grep dn: |wc -l = 41
ldap2 - slapcat |grep dn: |wc -l = 18

Right now I can get them in sync with a slapcat/slapadd, but when the go
into production I wont be able to say for certain which one is
authoritative.  That is the purpose of multi-master....

OpenLDAP 2.4.13, built by me (passed all tests) on Ubuntu Linux 32 bit

Any ideas as to what I can do to stop this from happening?
I noticed same behaviour if add a url to serverID and this url is the
local host address, thus pointing to itself.

serverID 1 192.168.100.23 # this is the local address
serverID 2 192.168.100.24
serverID 3 192.168.100 25

If I omit the url it works fine.

-Dieter

Sort of an unrelated question, but why would you do this?  I will freely
admit a lack of understanding in this area....  Is there a benefit that
I have over looked to replicating to one's self or is it more a about
maintainable copy pastable config?

Pat

its recommended to use the complete serverid-string (serverid  [id + url])
for every server that is part of your mmr-set. take a look into the admin
guide:
http://www.openldap.org/doc/admin24/replication.html#N-Way%20Multi-Master%20replication