-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Rich Megginson Sent: Tuesday, November 17, 2009 9:23 PM To: General discussion list for the 389 Directory server project. Subject: Re: [389-users] Replication and High Availalbiltiy
Bucl, Casper wrote: > -----Original Message----- > From: [email protected] > [mailto:[email protected]] On Behalf Of Rich > Megginson > Sent: Tuesday, November 17, 2009 12:35 PM > To: General discussion list for the 389 Directory server project. > Subject: Re: [389-users] Replication and High Availalbiltiy > > Bucl, Casper wrote: > >> -----Original Message----- >> From: [email protected] >> [mailto:[email protected]] On Behalf Of Rich >> Megginson >> Sent: Tuesday, November 17, 2009 8:23 AM >> To: General discussion list for the 389 Directory server project. >> Subject: Re: [389-users] Replication and High Availalbiltiy >> >> Bucl, Casper wrote: >> >> >>> Hi, >>> >>> I'm trying to create a high availability ldap for a system I have in >>> place that is currently using multimaster replication. Using a >>> shared storage system isn't an option in this case. >>> >>> To give you an idea of what our setup looks like, >>> >>> There are two nodes, that have replication set up. These are set up >>> as multimasters and have processes that write to both of them. These >>> changes replicate to the other ldap server. >>> >>> Now I need them to be in a high availability configuration. >>> >>> I have created duplicates of each node and gotten the high >>> availability portion on each of them to work correctly. >>> >>> The problem comes with fedora and replication. >>> >>> I have tried multiple ways of setting up fedora and replication and >>> they always seem to end up with changes not being replicated to the >>> other master when we have failed over to the secondary node. The two >>> most successful one's are below >>> >>> Configurations. >>> >>> Full Mesh: All links were set up as a two way replication. >>> >>> This always ends up with at least 2 nodes showing errors saying it >>> "Can't locate CSN" or "Duplicate node ID" >>> >>> Node1A ------- Node1B >>> >>> | \ / | >>> >>> | X | >>> >>> | / \ | >>> >>> Node2A ------- Node2B >>> >>> Single replication agreement between VIPs >>> >>> In this configuration, we initially copied over the slapd instance >>> directory on setup of the second HA node (Node1A to Node1B) so that >>> the settings and configurations are identical on both. Then as >>> changes were made to the ldap, we created backups using db2bak. >>> These backups are copied over to the failover box and then imported >>> on startup of fedora ds. This doesn't appear to backup the changelog >>> and ends up with an error saying "Can't locate CSN" again. >>> >>> Node1 VIP >>> >>> | >>> >>> | >>> >>> Node2 VIP >>> >>> I have tried other things as well and they were a lot less fruitful >>> than the two examples I have here. >>> >>> Has anyone set up a high availability scenario similar to this? Can >>> anyone suggest a different process or configuration that would >>> accomplish what I'm after? >>> >>> >>> >> Yes. Configurations like this have been working at high volume installations >> for several years. >> >> Let's start with - what platform are you running on your systems? What >> version of DS? What procedure did you use to set up and initialize your >> replicas? >> >> >>> Thanks, >>> >>> Casper >>> >>> -------------------------------------------------------------------- >>> - >>> - >>> -- >>> >>> -- >>> 389 users mailing list >>> [email protected] >>> https://www.redhat.com/mailman/listinfo/fedora-directory-users >>> >>> >>> >> The environment is set up using Fedora-Directory 1.0.4 >> >> > What platform? I would suggest using the latest (1.2.2 or 1.2.4 which is in > testing). We have fixed many, many bugs in replication since 1.0.4, and the > issue with the CSN you are reporting sounds like a bug that has been fixed in > 1.2.x. > >> To set up the multimaster replication I used the mmr.pl script. When >> reinitializing the consumers, I use ldapmodify and set the >> nsDS5BeginReplicaRefresh to start. >> >> Another question about the fully meshed configuration, Can there be more >> nodes? We will be wanting to add another HA node to the environment so this >> would take the total to 6 directory servers. The idea being that there is a >> central hub that is kind of the global master. Everyone replicates their >> info up to it and then it gets redistributed back out to the others. >> >> > Yes, you can have more than 4 masters. > >> Would the easier method be to copy the changelog information over to the >> standby node? >> > No. > >> Is there a method to do this? >> >> > Not really. > >> Thanks, >> Casper >> >> >> -- >> 389 users mailing list >> [email protected] >> https://www.redhat.com/mailman/listinfo/fedora-directory-users >> >> > Hi Rich, > What is the proper way to reinitialize the replication agreement in a > multimaster configuration. Whenever I try using the nsDS5BeginReplicaRefresh > method, it ends up creating a new changelog on the node being refreshed and > then it begins having replication issues. Notably the "Can't locate CSN" > error. Could this be something related to some of the bugs you were speaking > of? > Yes. See https://bugzilla.redhat.com/show_bug.cgi?id=388021 - fixed in 1.1.0 What platform are you running on? > Casper > > -- > 389 users mailing list > [email protected] > https://www.redhat.com/mailman/listinfo/fedora-directory-users > We are using a version of linux based on Redhat linux. Kernel version 2.6.26.8. Thanks for all your help. I will have to work on an upgrade plan to get our LDAP updated. Casper -- 389 users mailing list [email protected] https://www.redhat.com/mailman/listinfo/fedora-directory-users
