Christoph Herrmann C.Herrmann@science-computing.de writes:
Dear all,
we want to update our openldap-2.3.39 infrastructure to openldap 2.4.19. For that we did some stress tests to an openldap-2.4.19 test infrastructure. we are running the tests in a single master environment with 5 slave servers, all with openldap-2.4.19. All machines are running ntpd and are in sync:
[...]
this is our syncrepl config:
master/provider:
overlay syncprov syncprov-checkpoint 100 10 syncprov-sessionlog 100
slave/consumer:
syncrepl rid=467 provider=ldap://ldapma01:389 type=refreshAndPersist retry=2,30,15,40,60,+ searchbase="o=example,c=de" filter="(objectClass=*)" scope=sub schemachecking=on bindmethod=simple binddn="cn=replica,ou=replica,o=example,c=de" credentials="replica" updateref ldap://ldapma01:389
During the tests we always run in to problems with syncrepl. We are doing tests in the way:
bind; add|modify|delete; unbind; wait for n ms; bind .... (10 =< n =< 100)
depending on the time between the requests, about 1% of our write requests will not be synchronized to any off the slaves. It seems, the the master send the syncreplrequests out of sync, and all slaves reject syncing with the message: CSN too old, ignoring ...
What can we do, to solve that problem?
I found that disable logging on provider and consumers solved most of my problems. in addition you should check the number of waiters and increase the number of threads if required.
-Dieter