Windl, Ulrich wrote: > Hi! > I always wondered about the scope within SyncRepl RIDs have to be unique:
The manpage says explicitly: within a consumer site. > * Per database > * Per server > * Within all servers > For a support case covering MMR replication including cn=config I got this > response: > > One cannot copy the config between machines. Syncrepl configuration includes > the RID parameter. That RID must be UNIQUE on all servers. By copying the > config, > one now has two machines with the same RID. That response is wrong. Of course a config can be copied between machines. The test suite test050 does exactly that, with replication. > > I think its OK to do so, however. Slapd handles the situation where a server > is asked to sync from itself correctly ARAIK. > > We have three servers, says s1 (sid=1), s2 (sid=2), and s3 (sid=3). And there > are three syncrepl statements making each server to syn from each of the three > servers. Now you're talking about SIDs, which are completely different from RIDs. RIDs are purely informational; their main use is to disambiguate server log messages when there's more than one consumer configured in a server. They have no effect on the replication protocol at all. > > So effectively s1 will sync from s2 and s3, s2 will sync from s1 and s3, and > s3 will sync from s1 and s2 (each server ignoring the sync from itself). > > The log message may be like this for s2: > > slapd[1811445]: conn=1001 op=2 syncprov_matchops: skipping original sid 002 > > Am I correct, or is the supports answer correct? > > Kind regards, > > Ulrich Windl -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/
